From f5b281b55c8cca28228b870f906046429fe1a12f Mon Sep 17 00:00:00 2001
From: Claude <noreply@anthropic.com>
Date: Mon, 15 Jun 2026 04:52:39 +0000
Subject: [PATCH] Remove Tools button and Research mode; collapse runtime
 profile

Removes the user-facing Research mode feature and the Tools popover
entirely, then cleans up everything the removal orphaned.

UI:
- Drop the Tools (+) popover and Research pill from PromptForm; the
  footer now renders only the ModelSelector.
- Delete ResearchModeToggle and its use in the message-edit UI.

Run mode:
- Remove the runMode/AgentRunMode concept end to end (UI props,
  use-agent-session, follow-up plumbing, message metadata).
- Delete use-persistent-run-mode + persistent-run-mode-utils and the
  RUN_MODE_* localStorage constants.

Server:
- Remove runMode validation, the AGENT_RESEARCH_MODEL_UNAVAILABLE
  error, research model forcing, RESEARCH_MODEL, and the DEEP RESEARCH
  MODE system-prompt block.
- Collapse the single remaining runtime profile: drop
  AGENT_RUNTIME_PROFILES/AgentRuntimeProfileId and use
  AGENT_TOOL_MAX_STEPS directly; make taskMode a required param and
  always route provider options through the task-mode mapping.
- Delete the legacy getAiSdkGatewayProviderOptionsForMode shim.

The internal, content-inferred "research" task mode is intentionally
kept. Stored threads carrying legacy runMode metadata parse fine; the
field is dropped on read (no migration).

Tests and docs (README, CLAUDE.md) updated accordingly.
---
 CLAUDE.md                                     |  30 ++--
 README.md                                     |   2 +-
 src/app/api/agent/follow-ups/route.ts         |   1 -
 src/app/api/agent/route.ts                    |  16 +-
 .../agent/home/agent-session-state.ts         |   4 -
 .../agent/home/follow-up-questions.ts         |   4 -
 src/components/agent/home/home-agent-utils.ts |  10 +-
 src/components/agent/home/home-content.tsx    |  40 ++---
 .../agent/home/use-agent-session.ts           |  43 +----
 .../agent/home/use-follow-up-questions.ts     |   1 -
 .../agent/messages/assistant-message.tsx      |   4 -
 src/components/agent/messages/messages.tsx    |   8 -
 .../agent/messages/user-message.tsx           |  26 +--
 .../agent/prompt-form/prompt-form.tsx         | 151 ++----------------
 .../prompt-form/research-mode-toggle.tsx      |  70 --------
 src/hooks/agent/persistent-run-mode-utils.ts  |  68 --------
 src/hooks/agent/use-persistent-run-mode.tsx   | 115 -------------
 src/lib/constants.ts                          |   2 -
 src/lib/server/agent-context.ts               |   8 -
 src/lib/server/agent-follow-ups.ts            |  10 +-
 src/lib/server/agent-route.ts                 |  45 +-----
 src/lib/server/agent-runtime-config.ts        |   3 +-
 src/lib/server/llm/agent-runtime.ts           |  83 ++--------
 .../llm/ai-sdk-gateway-provider-options.ts    |  16 --
 src/lib/server/thread-payload.ts              |  11 --
 src/lib/shared/agent/messages.ts              |   4 -
 src/lib/shared/llm/models.ts                  |   2 -
 src/lib/shared/llm/system-instructions.ts     |  36 -----
 tests/agent-context.test.mjs                  |  12 +-
 .../agent-follow-ups-route-behavior.test.mjs  |   1 -
 tests/agent-follow-ups.test.mjs               |   2 -
 tests/agent-helper-behavior.test.mjs          |  90 +----------
 tests/agent-route-behavior.test.mjs           |  37 -----
 tests/agent-route-contract.test.mjs           |  12 +-
 tests/agent-runtime-synthesis-gating.test.mjs |   6 +-
 .../agent-runtime-task-mode-routing.test.mjs  |  24 +--
 tests/agent-session-state.test.mjs            |   6 -
 tests/agent-system-prompt.test.mjs            |  34 +---
 tests/assistant-message-contract.test.mjs     |   2 +-
 tests/follow-up-questions.test.mjs            |   4 +-
 tests/gateway-search-tools.test.mjs           |  26 ---
 tests/home-agent-utils.test.mjs               |   6 +-
 tests/model-registry.test.mjs                 |   6 -
 tests/persistent-run-mode.test.mjs            |  88 ----------
 tests/prompt-form-contract.test.mjs           |  48 +-----
 tests/thread-payload-contract.test.mjs        |  19 +--
 46 files changed, 107 insertions(+), 1129 deletions(-)
 delete mode 100644 src/components/agent/prompt-form/research-mode-toggle.tsx
 delete mode 100644 src/hooks/agent/persistent-run-mode-utils.ts
 delete mode 100644 src/hooks/agent/use-persistent-run-mode.tsx
 delete mode 100644 tests/persistent-run-mode.test.mjs

diff --git a/CLAUDE.md b/CLAUDE.md
index da67b7b8..17992c1b 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -53,10 +53,9 @@ Client (useAgentSession)
   → POST /api/agent (app/api/agent/route.ts)
     → Auth check (isAuthConfigured → getRequestSession)   [routes self-guard; see Middleware]
     → Sliding-window rate limit (rate-limit.ts, key user:<userId>)
-    → Zod validation (parseAgentStreamRequest) — incl. runMode (chat|research) and model
+    → Zod validation (parseAgentStreamRequest) — model + messages
     → Concurrency slot acquire (max 4 in-flight per user)
     → System prompt assembly (buildAgentSystemInstruction)
-    → Runtime profile resolution (chat_default | deep_research)
     → AI Gateway stream via Vercel AI SDK (startGatewayResponseStream → runAgentStream)
     → NDJSON chunks (application/x-ndjson) → client
       → readResponseStreamLines / parseStreamEventLine
@@ -77,14 +76,9 @@ Before the model sees them, messages pass through `toModelMessages` (`agent-runt
 
 This boundary is **enforced by Next.js bundling at build time** (importing `pg`/`better-auth`/server modules into a client bundle is a build error), **not** by an ESLint rule. Keep it in mind when adding imports.
 
-### Runtime Profiles vs. Task Modes (two independent axes)
+### Task Modes
 
-These are easy to confuse. They are orthogonal:
-
-- **Runtime profile** (`resolveRuntimeProfile` in `app/api/agent/route.ts`, `AGENT_RUNTIME_PROFILES` in `agent-runtime.ts`) — one of `chat_default`, `deep_research`. The profile drives the **tool-step budget** (the tool set itself is the same for both).
-- **Task mode** (`inferPromptTaskMode` in `agent-prompt-steering.ts`) — inferred from message content; drives only the **prompt overlay text** and the Gemini thinking level. Modes: `general`, `coding`, `debugging`, `writing`, `research`, `high_stakes`, `closed_answer`, `instruction_following`.
-
-Research mode is a **request flag** (`runMode: "research"`), not an inference. It selects the `deep_research` profile and `RESEARCH_MODEL` (Qwen 3.7 Max); if that model is unavailable the route returns 400 `AGENT_RESEARCH_MODEL_UNAVAILABLE`.
+**Task mode** (`inferPromptTaskMode` in `agent-prompt-steering.ts`) is inferred from message content and drives only the **prompt overlay text** (`TASK MODE OVERLAY: <MODE>`) and the Gemini thinking level. Modes: `general`, `coding`, `debugging`, `writing`, `research`, `high_stakes`, `closed_answer`, `instruction_following`. The tool-step budget is a single fixed constant (`AGENT_TOOL_MAX_STEPS`) — there is no per-request runtime-profile selection. (The `research` task mode is an automatic content-based overlay, not a user-facing toggle.)
 
 ### System Prompt Composition
 
@@ -94,9 +88,8 @@ Research mode is a **request flag** (`runMode: "research"`), not an inference. I
 2. `RUNTIME DATE CONTEXT` — current UTC timestamp + user timezone (from `X-User-Timezone` header)
 3. **Provider overlay** (`PROVIDER OVERLAY: ALIBABA|GOOGLE|MOONSHOTAI|XIAOMI`) — keyed by the model's **provider org**, not its nickname (alibaba=Qwen, google=Gemini, moonshotai=Kimi, xiaomi=MiMo). Always applied for a supported model.
 4. **Task mode overlay** (`TASK MODE OVERLAY: <MODE>`)
-5. `DEEP RESEARCH MODE` — only for `runMode: "research"`
-6. `IDENTITY AND TONE CONTEXT` — `DEFAULT_SOUL_FALLBACK_INSTRUCTION` (`src/lib/shared/llm/system-instructions.ts`)
-7. `AUTH USER CONTEXT` — authenticated user id, name, email
+5. `IDENTITY AND TONE CONTEXT` — `DEFAULT_SOUL_FALLBACK_INSTRUCTION` (`src/lib/shared/llm/system-instructions.ts`)
+6. `AUTH USER CONTEXT` — authenticated user id, name, email
 
 Inline-citation rules are appended **later**, by `withAiSdkInlineCitationInstruction` (`system-instruction-augmentations.ts`), inside `createAgentStreamResponse` — not by `buildAgentSystemInstruction`.
 
@@ -155,7 +148,7 @@ Each tool is only registered when its requirements are met.
 - **Restricted backend** (the only backend): computation-only Python imports (`math`, `collections`, `itertools`, …).
 - **Limits**: timeout default **10 s**, max **60 s**; code input and output each capped at **12,000 chars**.
 
-**Max tool steps** (constants in `agent-runtime-config.ts`): 12 default; 20 for research runs.
+**Max tool steps** (`AGENT_TOOL_MAX_STEPS` in `agent-runtime-config.ts`): 12.
 
 ### Model Registry
 
@@ -169,7 +162,6 @@ All models are defined in `src/lib/shared/llm/models.ts`:
 | `XIAOMI_MIMO_V2_5_PRO`          | `xiaomi/mimo-v2.5-pro`          | MiMo V2.5 Pro          |
 
 - `MODEL_SELECTOR_MODELS` — the chat selector subset: Qwen 3.7 Max, Kimi K2.6, MiMo V2.5 Pro.
-- `RESEARCH_MODEL` — Qwen 3.7 Max (Research mode also injects the Deep Research prompt block).
 - Gemini stays in `SUPPORTED_MODELS` for Gateway availability but is not a standalone chat selector option.
 - The agent is text-only: all chat input is plain text (no image, file, or PDF input).
 - Adding a model means updating `AvailableModels`, `ModelInfos`, `SUPPORTED_MODELS`, and optionally `MODEL_SELECTOR_MODELS`. `/api/models` filters this registry by configured keys (`getModels()` in `src/lib/actions/api-keys.ts`).
@@ -272,7 +264,7 @@ src/
                         #   follow-up-questions
     agent/messages/     # Message rendering (user, assistant, queued, activity timeline)
     agent/markdown/     # Memoized markdown renderer
-    agent/prompt-form/  # PromptForm (inline Research + Tools popover), ModelSelector
+    agent/prompt-form/  # PromptForm, ModelSelector
     app-sidebar.tsx     # Sidebar shell (lazy-loads SearchChats + NavThreads)
     nav-threads.tsx     # Thread list + client-side pinning (localStorage)
     nav-user.tsx        # Account menu + sign-out
@@ -282,8 +274,8 @@ src/
     layout/             # route group layout
     ui/                 # shadcn/ui primitives (base-lyra/stone) + ShikiCode
   hooks/
-    agent/              # use-models (server-seeded models context), use-persistent-selected-model,
-                        #   use-persistent-run-mode (both localStorage-backed)
+    agent/              # use-models (server-seeded models context),
+                        #   use-persistent-selected-model (localStorage-backed)
   lib/
     actions/api-keys.ts # getModels() server action
     brand/colors.ts     # App brand colors (used by layout/manifest)
@@ -312,7 +304,7 @@ src/
         initial-reasoning-chunk-sanitizer.ts # Redacted-reasoning filtering
         system-instruction-augmentations.ts  # Citation rules appended to prompt
     shared/
-      agent/messages.ts          # AgentStreamEvent, Message, ToolInvocation, run modes/statuses
+      agent/messages.ts          # AgentStreamEvent, Message, ToolInvocation, run statuses
       agent/reasoning-privacy.ts # sanitizeReasoningForDisplay
       agent-request-limits.ts    # Message/char limit defaults
       llm/models.ts              # AvailableModels, ModelInfos, SUPPORTED/SELECTOR/RESEARCH
@@ -394,5 +386,5 @@ Request size limits, stream/gateway timeouts, tool-step budgets, the rate-limit
 - The **mock smoke test uses the production server** (`next start`), so build first.
 - `pnpm.onlyBuiltDependencies` already approves the `sharp` build script — do not run `pnpm approve-builds`.
 - Don't reintroduce Sentry/PostHog/OpenTelemetry; they were intentionally removed in favor of Vercel Analytics/Speed Insights + structured logs.
-- Pinning, selected model, and run mode are **client-side localStorage** — there are no server columns or APIs for them.
+- Pinning and selected model are **client-side localStorage** — there are no server columns or APIs for them.
 - After signing up via `/api/auth/sign-up/email`, the session cookie is set automatically; no separate sign-in is needed.
diff --git a/README.md b/README.md
index cbf736e4..d1bc2f8a 100644
--- a/README.md
+++ b/README.md
@@ -1,6 +1,6 @@
 # Chloei
 
-Chloei is a Next.js 16 chat app backed by Vercel AI Gateway. It currently exposes a curated model selector that defaults to Qwen 3.7 Max and also includes Kimi K2.6 and MiMo V2.5 Pro, routes Research mode to Qwen 3.7 Max with a dedicated Deep Research instruction template, and offers local code execution, optional Tavily retrieval, and Better Auth email/password authentication with PostgreSQL-backed users and sessions.
+Chloei is a Next.js 16 chat app backed by Vercel AI Gateway. It currently exposes a curated model selector that defaults to Qwen 3.7 Max and also includes Kimi K2.6 and MiMo V2.5 Pro, and offers local code execution, optional Tavily retrieval, and Better Auth email/password authentication with PostgreSQL-backed users and sessions.
 
 ## Documentation
 
diff --git a/src/app/api/agent/follow-ups/route.ts b/src/app/api/agent/follow-ups/route.ts
index 2b302df7..ea02b31c 100644
--- a/src/app/api/agent/follow-ups/route.ts
+++ b/src/app/api/agent/follow-ups/route.ts
@@ -237,7 +237,6 @@ export async function POST(request: NextRequest) {
         aiGatewayApiKey,
         messages: parsed.messages,
         model: parsed.model,
-        runMode: parsed.runMode,
         signal: request.signal,
         userId: session.user.id,
       })
diff --git a/src/app/api/agent/route.ts b/src/app/api/agent/route.ts
index 079cc32a..68773101 100644
--- a/src/app/api/agent/route.ts
+++ b/src/app/api/agent/route.ts
@@ -31,7 +31,6 @@ import {
   isE2eMockModeEnabled,
 } from "@/lib/server/e2e-test-mode"
 import { resolveAgentFeatureFlags } from "@/lib/server/integration-flags"
-import type { AgentRuntimeProfileId } from "@/lib/server/llm/agent-runtime"
 import {
   evaluateAndConsumeSlidingWindowRateLimit,
   tryAcquireConcurrencySlot,
@@ -41,7 +40,6 @@ import {
   observeRouteResponse,
 } from "@/lib/server/route-observability"
 import { isThreadStoreNotInitializedError } from "@/lib/server/threads"
-import type { AgentRunMode } from "@/lib/shared"
 
 export const runtime = "nodejs"
 export const maxDuration = 800
@@ -50,10 +48,6 @@ function resolveRateLimitIdentifier(userId: string): string {
   return `user:${userId}`
 }
 
-function resolveRuntimeProfile(runMode: AgentRunMode): AgentRuntimeProfileId {
-  return runMode === "research" ? "deep_research" : "chat_default"
-}
-
 export async function POST(request: NextRequest) {
   const requestId = resolveRequestId(request)
   const logger = createLogger(`agent:${requestId}`)
@@ -163,11 +157,7 @@ export async function POST(request: NextRequest) {
     const userTimeZone = resolveUserTimeZone(request)
     const featureFlags = await resolveAgentFeatureFlags()
     const promptProvider = resolvePromptProvider(selectedModel)
-    const inferredPromptTaskMode =
-      parsedRequest.runMode === "research"
-        ? "research"
-        : inferPromptTaskMode(parsedRequest.messages)
-    const promptTaskMode = inferredPromptTaskMode
+    const promptTaskMode = inferPromptTaskMode(parsedRequest.messages)
     const systemInstruction = buildAgentSystemInstruction(
       {
         id: session.user.id,
@@ -179,9 +169,6 @@ export async function POST(request: NextRequest) {
         userTimeZone,
         provider: promptProvider,
         taskMode: promptTaskMode,
-        ...(parsedRequest.runMode === "research"
-          ? { deepResearchMode: true }
-          : {}),
       }
     )
 
@@ -253,7 +240,6 @@ export async function POST(request: NextRequest) {
         aiGatewayApiKey,
         tavilyApiKey,
         userTimeZone,
-        runtimeProfile: resolveRuntimeProfile(parsedRequest.runMode),
         taskMode: promptTaskMode,
         userId: session.user.id,
         featureFlags,
diff --git a/src/components/agent/home/agent-session-state.ts b/src/components/agent/home/agent-session-state.ts
index f58a9364..f5f24a4f 100644
--- a/src/components/agent/home/agent-session-state.ts
+++ b/src/components/agent/home/agent-session-state.ts
@@ -1,5 +1,4 @@
 import {
-  type AgentRunMode,
   type FollowUpQuestion,
   type Message as AgentMessage,
   type ModelType,
@@ -31,14 +30,12 @@ export function createAssistantMessageFromAccumulator({
   createdAt,
   accumulator,
   model,
-  runMode,
   isStreaming,
 }: {
   id: string
   createdAt: string
   accumulator: AgentStreamAccumulator
   model: ModelType
-  runMode: AgentRunMode
   isStreaming: boolean
 }): AgentMessage {
   return {
@@ -49,7 +46,6 @@ export function createAssistantMessageFromAccumulator({
     createdAt,
     metadata: {
       isStreaming,
-      runMode,
       parts: [{ type: "text", text: accumulator.content }],
       ...(accumulator.agentStatus
         ? { agentStatus: accumulator.agentStatus }
diff --git a/src/components/agent/home/follow-up-questions.ts b/src/components/agent/home/follow-up-questions.ts
index fe6ba995..125f0744 100644
--- a/src/components/agent/home/follow-up-questions.ts
+++ b/src/components/agent/home/follow-up-questions.ts
@@ -1,5 +1,4 @@
 import {
-  type AgentRunMode,
   type FollowUpQuestion,
   isModelType,
   type Message as AgentMessage,
@@ -19,7 +18,6 @@ interface FollowUpQuestionRequestTarget {
   assistantMessageId: string
   messages: AgentMessage[]
   model: ModelType
-  runMode: AgentRunMode
 }
 
 type FollowUpQuestionRequestKind = "backfill" | "final" | "parallel"
@@ -29,7 +27,6 @@ export interface FollowUpQuestionRequestParams {
   requestKind: FollowUpQuestionRequestKind
   messages: AgentMessage[]
   model: ModelType
-  runMode: AgentRunMode
   threadId: string
 }
 
@@ -148,7 +145,6 @@ export function getFollowUpQuestionRequestTargets(
       assistantMessageId: message.id,
       messages: messages.slice(0, index + 1),
       model,
-      runMode: message.metadata.runMode ?? "chat",
     })
   })
 
diff --git a/src/components/agent/home/home-agent-utils.ts b/src/components/agent/home/home-agent-utils.ts
index 98a5b2cc..39ade5fa 100644
--- a/src/components/agent/home/home-agent-utils.ts
+++ b/src/components/agent/home/home-agent-utils.ts
@@ -1,10 +1,6 @@
 import { ASSISTANT_EMPTY_RESPONSE_FALLBACK } from "@/lib/constants"
 import { createRequestHeaders } from "@/lib/request-id"
-import {
-  type AgentRunMode,
-  type Message as AgentMessage,
-  type ModelType,
-} from "@/lib/shared"
+import { type Message as AgentMessage, type ModelType } from "@/lib/shared"
 import {
   AGENT_REQUEST_MAX_MESSAGE_CHARS,
   AGENT_REQUEST_MAX_MESSAGES,
@@ -107,8 +103,7 @@ export function toRequestMessages(
 export function appendUserMessage(
   currentMessages: AgentMessage[],
   content: string,
-  model: ModelType,
-  runMode: AgentRunMode = "chat"
+  model: ModelType
 ): AgentMessage[] {
   const userMessage: AgentMessage = {
     id: createClientMessageId(),
@@ -119,7 +114,6 @@ export function appendUserMessage(
     metadata: {
       isStreaming: false,
       selectedModel: model,
-      runMode,
     },
   }
 
diff --git a/src/components/agent/home/home-content.tsx b/src/components/agent/home/home-content.tsx
index 69e21cf3..206471d4 100644
--- a/src/components/agent/home/home-content.tsx
+++ b/src/components/agent/home/home-content.tsx
@@ -28,11 +28,7 @@ import {
   TooltipTrigger,
 } from "@/components/ui/tooltip"
 import { useIsMobile } from "@/hooks/use-mobile"
-import {
-  type AgentRunMode,
-  type AuthViewer,
-  type ModelType,
-} from "@/lib/shared"
+import { type AuthViewer, type ModelType } from "@/lib/shared"
 import { cn } from "@/lib/utils"
 
 import { PromptForm } from "../prompt-form/prompt-form"
@@ -85,28 +81,15 @@ export function HomePageContent({
   } = useAgentSession(threadStore)
 
   const handlePromptFormSubmit = useCallback(
-    (
-      message: string,
-      model: ModelType,
-      _isStreaming: boolean,
-      runMode: AgentRunMode
-    ) => {
-      handlePromptSubmit(message, model, runMode)
+    (message: string, model: ModelType) => {
+      handlePromptSubmit(message, model)
     },
     [handlePromptSubmit]
   )
 
   const handleFollowUpQuestionClick = useCallback(
-    ({
-      model,
-      question,
-      runMode,
-    }: {
-      model: ModelType
-      question: string
-      runMode: AgentRunMode
-    }) => {
-      handlePromptSubmit(question, model, runMode)
+    ({ model, question }: { model: ModelType; question: string }) => {
+      handlePromptSubmit(question, model)
     },
     [handlePromptSubmit]
   )
@@ -221,15 +204,10 @@ export function HomePageContent({
   }, [fallbackTransitionMs])
 
   const handleAnimatedPromptSubmit = useCallback(
-    (
-      message: string,
-      model: ModelType,
-      _isStreaming: boolean,
-      runMode: AgentRunMode
-    ) => {
+    (message: string, model: ModelType) => {
       if (isMobile) {
         startFallbackConversationTransition()
-        handlePromptSubmit(message, model, runMode)
+        handlePromptSubmit(message, model)
         return
       }
 
@@ -244,13 +222,13 @@ export function HomePageContent({
 
       if (!startViewTransition) {
         startFallbackConversationTransition()
-        handlePromptSubmit(message, model, runMode)
+        handlePromptSubmit(message, model)
         return
       }
 
       startViewTransition(() => {
         flushSync(() => {
-          handlePromptSubmit(message, model, runMode)
+          handlePromptSubmit(message, model)
         })
       })
     },
diff --git a/src/components/agent/home/use-agent-session.ts b/src/components/agent/home/use-agent-session.ts
index 33fd6691..c45a46fd 100644
--- a/src/components/agent/home/use-agent-session.ts
+++ b/src/components/agent/home/use-agent-session.ts
@@ -11,7 +11,6 @@ import {
 } from "@/lib/http-error"
 import { getRequestIdFromHeaders } from "@/lib/request-id"
 import {
-  type AgentRunMode,
   type FollowUpQuestion,
   isModelType,
   type Message as AgentMessage,
@@ -62,13 +61,11 @@ interface EditMessageParams {
   messageId: string
   newContent: string
   newModel: ModelType
-  newRunMode: AgentRunMode
 }
 
 interface QueuedSubmission {
   message: string
   model: ModelType
-  runMode: AgentRunMode
 }
 
 const INITIAL_STATE: AgentSessionState = {
@@ -253,7 +250,6 @@ export function useAgentSession({
       body: Record<string, unknown>
       baseMessages: AgentMessage[]
       model: ModelType
-      runMode: AgentRunMode
       threadId: string
       errorTitle: string
     }) => {
@@ -305,7 +301,6 @@ export function useAgentSession({
           createdAt: assistantCreatedAt,
           accumulator: nextAccumulator,
           model: effectiveModel,
-          runMode: params.runMode,
           isStreaming: streamFlags.isStreaming,
         })
         let assistantMessageWithFollowUpState = assistantMessage
@@ -367,7 +362,6 @@ export function useAgentSession({
           messages: messagesRef.current,
           model: effectiveModel,
           requestKind: "parallel",
-          runMode: params.runMode,
           threadId: params.threadId,
         })
       }
@@ -498,7 +492,6 @@ export function useAgentSession({
             messages: messagesRef.current,
             model: effectiveModel,
             requestKind: "final",
-            runMode: params.runMode,
             threadId: params.threadId,
           })
         }
@@ -543,7 +536,6 @@ export function useAgentSession({
           createdAt: new Date().toISOString(),
           metadata: {
             isStreaming: false,
-            runMode: params.runMode,
             parts: [{ type: "text", text: fallback }],
             agentStatus: "failed",
           },
@@ -591,7 +583,6 @@ export function useAgentSession({
     async (
       nextMessages: AgentMessage[],
       model: ModelType,
-      runMode: AgentRunMode = "chat",
       threadId?: string
     ) => {
       const activeThreadId = threadId ?? ensureCurrentThreadId()
@@ -603,11 +594,9 @@ export function useAgentSession({
         threadId: activeThreadId,
         baseMessages: nextMessages,
         model,
-        runMode,
         errorTitle: "Failed to send message",
         body: {
           model,
-          runMode,
           threadId: activeThreadId,
           messages: requestMessages,
         },
@@ -617,11 +606,7 @@ export function useAgentSession({
   )
 
   const handleSubmit = useCallback(
-    async (
-      message: string,
-      model: ModelType,
-      runMode: AgentRunMode = "chat"
-    ) => {
+    async (message: string, model: ModelType) => {
       const trimmedMessage = message.trim()
       if (!trimmedMessage) {
         return
@@ -638,17 +623,16 @@ export function useAgentSession({
       const nextMessages = appendUserMessage(
         messagesRef.current,
         trimmedMessage,
-        model,
-        runMode
+        model
       )
 
-      await runAgentRequest(nextMessages, model, runMode, activeThreadId)
+      await runAgentRequest(nextMessages, model, activeThreadId)
     },
     [ensureCurrentThreadId, runAgentRequest]
   )
 
   const handleEditMessage = useCallback(
-    ({ messageId, newContent, newModel, newRunMode }: EditMessageParams) => {
+    ({ messageId, newContent, newModel }: EditMessageParams) => {
       const trimmedContent = newContent.trim()
       const currentMessages = messagesRef.current
 
@@ -684,7 +668,6 @@ export function useAgentSession({
         metadata: {
           ...targetMessage.metadata,
           selectedModel: newModel,
-          runMode: newRunMode,
         },
       }
 
@@ -702,18 +685,13 @@ export function useAgentSession({
         )
       }
 
-      void runAgentRequest(
-        nextMessages,
-        newModel,
-        newRunMode,
-        activeThreadId ?? undefined
-      )
+      void runAgentRequest(nextMessages, newModel, activeThreadId ?? undefined)
     },
     [createThreadSnapshot, runAgentRequest, saveThread]
   )
 
   const handlePromptSubmit = useCallback(
-    (message: string, model: ModelType, runMode: AgentRunMode = "chat") => {
+    (message: string, model: ModelType) => {
       const trimmedMessage = message.trim()
       if (!trimmedMessage) {
         return
@@ -723,12 +701,11 @@ export function useAgentSession({
         setQueuedSubmission({
           message: trimmedMessage,
           model,
-          runMode,
         })
         return
       }
 
-      void handleSubmit(trimmedMessage, model, runMode)
+      void handleSubmit(trimmedMessage, model)
     },
     [handleSubmit]
   )
@@ -739,11 +716,7 @@ export function useAgentSession({
     }
 
     setQueuedSubmission(null)
-    void handleSubmit(
-      queuedSubmission.message,
-      queuedSubmission.model,
-      queuedSubmission.runMode
-    )
+    void handleSubmit(queuedSubmission.message, queuedSubmission.model)
   }, [streamingState, queuedSubmission, handleSubmit])
 
   return {
diff --git a/src/components/agent/home/use-follow-up-questions.ts b/src/components/agent/home/use-follow-up-questions.ts
index 7365b825..cf852b8e 100644
--- a/src/components/agent/home/use-follow-up-questions.ts
+++ b/src/components/agent/home/use-follow-up-questions.ts
@@ -288,7 +288,6 @@ export function useFollowUpQuestions({
             headers: createAgentRequestHeaders(),
             body: JSON.stringify({
               model: params.model,
-              runMode: params.runMode,
               threadId: params.threadId,
               assistantMessageId: params.assistantMessageId,
               messages: toRequestMessages(params.messages),
diff --git a/src/components/agent/messages/assistant-message.tsx b/src/components/agent/messages/assistant-message.tsx
index d204d17d..04870eae 100644
--- a/src/components/agent/messages/assistant-message.tsx
+++ b/src/components/agent/messages/assistant-message.tsx
@@ -13,7 +13,6 @@ import { useMemo, useState } from "react"
 import { LogoHover } from "@/components/graphics/logo/logo-hover"
 import { useCopyToClipboard } from "@/hooks/use-copy-to-clipboard"
 import {
-  type AgentRunMode,
   type FollowUpQuestion,
   isModelType,
   type Message,
@@ -190,7 +189,6 @@ export function AssistantMessage({
   onFollowUpQuestionClick?: (params: {
     model: ModelType
     question: string
-    runMode: AgentRunMode
   }) => void
   onRegenerate?: () => void
 }) {
@@ -227,7 +225,6 @@ export function AssistantMessage({
   const followUpModel = isModelType(message.llmModel)
     ? message.llmModel
     : message.metadata?.selectedModel
-  const followUpRunMode = message.metadata?.runMode ?? "chat"
   const followUpQuestions = isAssistantStreaming
     ? []
     : (message.metadata?.followUpQuestions ?? []).filter(
@@ -409,7 +406,6 @@ export function AssistantMessage({
                   onFollowUpQuestionClick({
                     model: followUpModel,
                     question,
-                    runMode: followUpRunMode,
                   })
                 }}
               />
diff --git a/src/components/agent/messages/messages.tsx b/src/components/agent/messages/messages.tsx
index 5a64f5f3..efaee3c7 100644
--- a/src/components/agent/messages/messages.tsx
+++ b/src/components/agent/messages/messages.tsx
@@ -4,7 +4,6 @@ import "../shared/shell-styles.css"
 import { memo, useMemo } from "react"
 
 import {
-  type AgentRunMode,
   isAssistantMessage,
   isModelType,
   isUserMessage,
@@ -71,12 +70,10 @@ function MessagesComponent({
     messageId: string
     newContent: string
     newModel: ModelType
-    newRunMode: AgentRunMode
   }) => Promise<void> | void
   onFollowUpQuestionClick?: (params: {
     model: ModelType
     question: string
-    runMode: AgentRunMode
   }) => void
 }) {
   const messageGroups = useMemo(() => groupMessages(messages), [messages])
@@ -133,10 +130,6 @@ function MessagesComponent({
                 const regenerateModel = isModelType(message.llmModel)
                   ? message.llmModel
                   : userMessage?.metadata?.selectedModel
-                const regenerateRunMode =
-                  message.metadata?.runMode ??
-                  userMessage?.metadata?.runMode ??
-                  "chat"
                 const canRegenerate =
                   !disableEditing &&
                   Boolean(userMessage) &&
@@ -149,7 +142,6 @@ function MessagesComponent({
                           messageId: userMessage.id,
                           newContent: userMessage.content,
                           newModel: regenerateModel,
-                          newRunMode: regenerateRunMode,
                         })
                       }
                     : undefined
diff --git a/src/components/agent/messages/user-message.tsx b/src/components/agent/messages/user-message.tsx
index 2b3cffcb..c4ce8349 100644
--- a/src/components/agent/messages/user-message.tsx
+++ b/src/components/agent/messages/user-message.tsx
@@ -5,7 +5,6 @@ import { toast } from "sonner"
 import { useModels } from "@/hooks/agent/use-models"
 import { useCopyToClipboard } from "@/hooks/use-copy-to-clipboard"
 import {
-  type AgentRunMode,
   getModelSelectorModels,
   isModelSelectorModel,
   isModelType,
@@ -19,7 +18,6 @@ import { Button } from "../../ui/button"
 import { Textarea } from "../../ui/textarea"
 import { Tooltip, TooltipContent, TooltipTrigger } from "../../ui/tooltip"
 import { ModelSelector } from "../prompt-form/model-selector"
-import { ResearchModeToggle } from "../prompt-form/research-mode-toggle"
 import {
   agentShellFrameClass,
   agentShellHighlightClass,
@@ -45,7 +43,6 @@ export function UserMessage({
     messageId: string
     newContent: string
     newModel: ModelType
-    newRunMode: AgentRunMode
   }) => Promise<void> | void
 }) {
   const { data: availableModels } = useModels()
@@ -72,8 +69,6 @@ export function UserMessage({
   const [isEditing, setIsEditing] = useState(false)
   const [editValue, setEditValue] = useState(message.content)
   const [selectedModel, setSelectedModel] = useState<ModelType>(initialModel)
-  const initialRunMode = message.metadata?.runMode ?? "chat"
-  const [runMode, setRunMode] = useState<AgentRunMode>(initialRunMode)
   const [isEditPending, setIsEditPending] = useState(false)
   const messageContentRef = useRef<HTMLDivElement>(null)
   const textareaRef = useRef<HTMLTextAreaElement>(null)
@@ -107,16 +102,14 @@ export function UserMessage({
   const handleStartEditing = useCallback(() => {
     setEditValue(message.content)
     setSelectedModel(initialModel)
-    setRunMode(initialRunMode)
     setIsEditing(true)
-  }, [initialModel, initialRunMode, message.content])
+  }, [initialModel, message.content])
 
   const handleStopEditing = useCallback(() => {
     setIsEditing(false)
     setEditValue(message.content)
     setSelectedModel(initialModel)
-    setRunMode(initialRunMode)
-  }, [message.content, initialModel, initialRunMode])
+  }, [message.content, initialModel])
 
   const handleSubmit = useCallback(async () => {
     const trimmedValue = editValue.trim()
@@ -137,7 +130,6 @@ export function UserMessage({
         messageId: message.id,
         newContent: trimmedValue,
         newModel: selectedModel,
-        newRunMode: runMode,
       })
       setIsEditing(false)
     } catch (error) {
@@ -147,14 +139,7 @@ export function UserMessage({
     } finally {
       setIsEditPending(false)
     }
-  }, [
-    editValue,
-    handleStopEditing,
-    message.id,
-    onEditMessage,
-    runMode,
-    selectedModel,
-  ])
+  }, [editValue, handleStopEditing, message.id, onEditMessage, selectedModel])
 
   const onKeyDown = useCallback(
     (e: React.KeyboardEvent<HTMLTextAreaElement>) => {
@@ -242,11 +227,6 @@ export function UserMessage({
                   selectedModel={selectedModel}
                   handleSelectModel={handleSelectModel}
                 />
-                <ResearchModeToggle
-                  runMode={runMode}
-                  onRunModeChange={setRunMode}
-                  disabled={isEditPending}
-                />
               </div>
 
               <div className="flex min-w-0 items-center justify-end gap-[8px]">
diff --git a/src/components/agent/prompt-form/prompt-form.tsx b/src/components/agent/prompt-form/prompt-form.tsx
index 67f5853c..c62e0e06 100644
--- a/src/components/agent/prompt-form/prompt-form.tsx
+++ b/src/components/agent/prompt-form/prompt-form.tsx
@@ -2,7 +2,7 @@
 
 import "../shared/shell-styles.css"
 
-import { CornerRightUp, Loader2, Plus, Square, Telescope } from "lucide-react"
+import { CornerRightUp, Loader2, Square } from "lucide-react"
 import {
   type CSSProperties,
   type TransitionStartFunction,
@@ -15,20 +15,10 @@ import {
 
 import { RefreshGlow } from "@/components/graphics/effects/refresh-glow"
 import { Button } from "@/components/ui/button"
-import {
-  Popover,
-  PopoverContent,
-  PopoverTrigger,
-} from "@/components/ui/popover"
 import { Textarea } from "@/components/ui/textarea"
 import { useModels } from "@/hooks/agent/use-models"
-import { usePersistentRunMode } from "@/hooks/agent/use-persistent-run-mode"
 import { usePersistentSelectedModel } from "@/hooks/agent/use-persistent-selected-model"
-import {
-  type AgentRunMode,
-  getModelSelectorModels,
-  type ModelType,
-} from "@/lib/shared"
+import { getModelSelectorModels, type ModelType } from "@/lib/shared"
 import { cn } from "@/lib/utils"
 
 import { QueuedAction } from "../messages/queued-message"
@@ -44,7 +34,6 @@ import { ModelSelector } from "./model-selector"
 interface QueuedPromptSubmission {
   message: string
   model: ModelType
-  runMode: AgentRunMode
 }
 
 export function PromptForm({
@@ -63,12 +52,7 @@ export function PromptForm({
   transition,
   viewTransitionName,
 }: {
-  onSubmit?: (
-    message: string,
-    model: ModelType,
-    queue: boolean,
-    runMode: AgentRunMode
-  ) => void
+  onSubmit?: (message: string, model: ModelType, queue: boolean) => void
   onStopStream?: () => void
   isStreaming?: boolean
   isHome?: boolean
@@ -92,11 +76,8 @@ export function PromptForm({
   const shouldShowRefreshAnimation = isHome && !dockToBottomOnHome
 
   const [message, setMessage] = useState("")
-  const [isToolsOpen, setIsToolsOpen] = useState(false)
-  const { runMode, setRunMode } = usePersistentRunMode()
   const trimmedMessage = useMemo(() => message.trim(), [message])
   const textareaRef = useRef<HTMLTextAreaElement>(null)
-  const shouldPreventToolsCloseAutoFocusRef = useRef(false)
 
   const { data: availableModels } = useModels()
   const modelSelectorModels = useMemo(
@@ -118,7 +99,6 @@ export function PromptForm({
   )
 
   const resolvedSelectedModel = selectedModel
-  const isResearchMode = runMode === "research"
 
   const handleSelectModel = useCallback(
     (model: ModelType | null) => {
@@ -133,7 +113,6 @@ export function PromptForm({
     }
 
     setMessage(queuedSubmission.message)
-    setRunMode(queuedSubmission.runMode)
     setSelectedModel(queuedSubmission.model)
     onClearQueuedMessage?.()
 
@@ -148,22 +127,7 @@ export function PromptForm({
       textarea.setSelectionRange(cursorPosition, cursorPosition)
       textarea.scrollTop = textarea.scrollHeight
     })
-  }, [onClearQueuedMessage, queuedSubmission, setRunMode, setSelectedModel])
-
-  const handleSetRunMode = useCallback(
-    (nextRunMode: AgentRunMode) => {
-      if (isFormPending) {
-        return
-      }
-
-      shouldPreventToolsCloseAutoFocusRef.current = true
-      setRunMode((currentRunMode) =>
-        currentRunMode === nextRunMode ? "chat" : nextRunMode
-      )
-      setIsToolsOpen(false)
-    },
-    [isFormPending, setRunMode]
-  )
+  }, [onClearQueuedMessage, queuedSubmission, setSelectedModel])
 
   const submitPrompt = useCallback(() => {
     const nextMessage = message.trim()
@@ -185,8 +149,7 @@ export function PromptForm({
       }
     }
 
-    const activeRunMode = runMode
-    onSubmit?.(nextMessage, resolvedSelectedModel, isStreaming, activeRunMode)
+    onSubmit?.(nextMessage, resolvedSelectedModel, isStreaming)
     setMessage("")
 
     return true
@@ -198,7 +161,6 @@ export function PromptForm({
     onStopStream,
     onSubmit,
     resolvedSelectedModel,
-    runMode,
   ])
 
   const handleSubmit = useCallback(
@@ -237,26 +199,6 @@ export function PromptForm({
     }
   }, [isStreaming, onStopStream])
 
-  useEffect(() => {
-    const handleResearchShortcut = (event: KeyboardEvent) => {
-      if (
-        !isFormPending &&
-        event.code === "Slash" &&
-        (event.metaKey || event.ctrlKey)
-      ) {
-        event.preventDefault()
-        setRunMode((currentRunMode) =>
-          currentRunMode === "research" ? "chat" : "research"
-        )
-      }
-    }
-
-    window.addEventListener("keydown", handleResearchShortcut)
-    return () => {
-      window.removeEventListener("keydown", handleResearchShortcut)
-    }
-  }, [isFormPending, setRunMode])
-
   const isSubmitButtonDisabled =
     isFormPending || !resolvedSelectedModel || (!isStreaming && !trimmedMessage)
 
@@ -312,85 +254,10 @@ export function PromptForm({
 
           <div className="grid grid-cols-2 items-center px-2 py-2">
             <div className="flex min-w-0 items-center justify-start gap-1">
-              <Popover
-                open={isToolsOpen}
-                onOpenChange={(open) => {
-                  if (!isFormPending) {
-                    setIsToolsOpen(open)
-                  }
-                }}
-              >
-                <PopoverTrigger asChild aria-controls={undefined}>
-                  <Button
-                    type="button"
-                    variant="ghost"
-                    size="iconSm"
-                    disabled={isFormPending}
-                    aria-label="Tools"
-                    className="shrink-0 text-muted-foreground hover:bg-sidebar-border hover:text-foreground"
-                  >
-                    <Plus className="size-4" />
-                  </Button>
-                </PopoverTrigger>
-                <PopoverContent
-                  side="bottom"
-                  align="start"
-                  sideOffset={2}
-                  finalFocus={() => {
-                    if (!shouldPreventToolsCloseAutoFocusRef.current) {
-                      return true
-                    }
-
-                    shouldPreventToolsCloseAutoFocusRef.current = false
-                    textareaRef.current?.focus({ preventScroll: true })
-                    return false
-                  }}
-                  className="flex w-56 flex-col gap-0.5 rounded-none p-1.5"
-                >
-                  <Button
-                    type="button"
-                    variant="ghost"
-                    size="sm"
-                    disabled={isFormPending}
-                    aria-pressed={isResearchMode}
-                    className={cn(
-                      "w-full justify-start px-2 font-normal text-muted-foreground hover:bg-accent hover:text-foreground focus-visible:border-transparent focus-visible:ring-0",
-                      isResearchMode &&
-                        "bg-accent text-foreground hover:bg-accent aria-pressed:bg-accent aria-pressed:text-foreground"
-                    )}
-                    onClick={() => {
-                      handleSetRunMode("research")
-                    }}
-                  >
-                    <Telescope className="size-3.5" />
-                    <span>Research</span>
-                    <span className="ml-auto text-[10px] text-muted-foreground">
-                      ⌘/
-                    </span>
-                  </Button>
-                </PopoverContent>
-              </Popover>
-              {isResearchMode ? (
-                <Button
-                  type="button"
-                  size="sm"
-                  variant="ghost"
-                  disabled={isFormPending}
-                  aria-pressed="true"
-                  className="bg-accent px-2 font-normal text-foreground hover:bg-accent focus-visible:border-transparent focus-visible:ring-0"
-                  onClick={() => {
-                    handleSetRunMode("research")
-                  }}
-                >
-                  <Telescope className="size-3.5" />
-                  <span>Research</span>
-                </Button>
-              ) : (
-                <ModelSelector
-                  selectedModel={resolvedSelectedModel}
-                  handleSelectModel={handleSelectModel}
-                />
-              )}
+              <ModelSelector
+                selectedModel={resolvedSelectedModel}
+                handleSelectModel={handleSelectModel}
+              />
             </div>
 
             <div className="flex min-w-0 items-center justify-end gap-[8px]">
diff --git a/src/components/agent/prompt-form/research-mode-toggle.tsx b/src/components/agent/prompt-form/research-mode-toggle.tsx
deleted file mode 100644
index 3b070936..00000000
--- a/src/components/agent/prompt-form/research-mode-toggle.tsx
+++ /dev/null
@@ -1,70 +0,0 @@
-import { Telescope } from "lucide-react"
-import { useEffect } from "react"
-
-import { Button } from "@/components/ui/button"
-import {
-  Tooltip,
-  TooltipContent,
-  TooltipTrigger,
-} from "@/components/ui/tooltip"
-import type { AgentRunMode } from "@/lib/shared"
-import { cn } from "@/lib/utils"
-
-export function ResearchModeToggle({
-  runMode,
-  onRunModeChange,
-  disabled = false,
-}: {
-  runMode: AgentRunMode
-  onRunModeChange: (runMode: AgentRunMode) => void
-  disabled?: boolean
-}) {
-  const isResearch = runMode === "research"
-
-  useEffect(() => {
-    const handleKeyDown = (event: KeyboardEvent) => {
-      if (
-        !disabled &&
-        event.code === "Slash" &&
-        (event.metaKey || event.ctrlKey)
-      ) {
-        event.preventDefault()
-        onRunModeChange(isResearch ? "chat" : "research")
-      }
-    }
-
-    window.addEventListener("keydown", handleKeyDown)
-    return () => {
-      window.removeEventListener("keydown", handleKeyDown)
-    }
-  }, [disabled, isResearch, onRunModeChange])
-
-  const baseClass =
-    "px-2 font-normal text-muted-foreground hover:bg-accent focus-visible:border-transparent focus-visible:ring-0"
-  const activeClass =
-    "bg-accent text-foreground hover:bg-accent aria-pressed:bg-accent aria-pressed:text-foreground"
-
-  return (
-    <Tooltip>
-      <TooltipTrigger asChild>
-        <Button
-          type="button"
-          size="sm"
-          variant="ghost"
-          disabled={disabled}
-          aria-pressed={isResearch}
-          className={cn(baseClass, isResearch && activeClass)}
-          onClick={() => {
-            onRunModeChange(isResearch ? "chat" : "research")
-          }}
-        >
-          <Telescope className="size-3.5" />
-          <span>Research</span>
-        </Button>
-      </TooltipTrigger>
-      <TooltipContent side="top" align="start" shortcut="⌘/">
-        Deep Research
-      </TooltipContent>
-    </Tooltip>
-  )
-}
diff --git a/src/hooks/agent/persistent-run-mode-utils.ts b/src/hooks/agent/persistent-run-mode-utils.ts
deleted file mode 100644
index 521a12df..00000000
--- a/src/hooks/agent/persistent-run-mode-utils.ts
+++ /dev/null
@@ -1,68 +0,0 @@
-import { AGENT_RUN_MODES, type AgentRunMode } from "@/lib/shared"
-
-const STORED_RUN_MODE_VERSION = 1
-
-interface StoredRunMode {
-  runMode: AgentRunMode
-  source: "user"
-  version: typeof STORED_RUN_MODE_VERSION
-}
-
-function isAgentRunMode(value: unknown): value is AgentRunMode {
-  return (
-    typeof value === "string" &&
-    (AGENT_RUN_MODES as readonly string[]).includes(value)
-  )
-}
-
-export function serializeStoredRunMode(runMode: AgentRunMode): StoredRunMode {
-  return {
-    runMode,
-    source: "user",
-    version: STORED_RUN_MODE_VERSION,
-  }
-}
-
-export function parseStoredRunMode(value: unknown): AgentRunMode | null {
-  if (typeof value !== "string") {
-    return null
-  }
-
-  if (isAgentRunMode(value)) {
-    return value
-  }
-
-  try {
-    const parsed: unknown = JSON.parse(value)
-
-    if (
-      parsed &&
-      typeof parsed === "object" &&
-      "runMode" in parsed &&
-      "source" in parsed &&
-      "version" in parsed &&
-      parsed.source === "user" &&
-      parsed.version === STORED_RUN_MODE_VERSION &&
-      isAgentRunMode(parsed.runMode)
-    ) {
-      return parsed.runMode
-    }
-  } catch {
-    return null
-  }
-
-  return null
-}
-
-export function resolvePersistedRunMode(params: {
-  storedRunMode: AgentRunMode | null
-  currentRunMode?: AgentRunMode | null
-  fallbackRunMode?: AgentRunMode
-}): AgentRunMode {
-  return (
-    params.storedRunMode ??
-    params.currentRunMode ??
-    params.fallbackRunMode ??
-    "chat"
-  )
-}
diff --git a/src/hooks/agent/use-persistent-run-mode.tsx b/src/hooks/agent/use-persistent-run-mode.tsx
deleted file mode 100644
index 516e180b..00000000
--- a/src/hooks/agent/use-persistent-run-mode.tsx
+++ /dev/null
@@ -1,115 +0,0 @@
-"use client"
-
-import { useCallback, useSyncExternalStore } from "react"
-
-import { RUN_MODE_STORAGE_KEY, RUN_MODE_UPDATED_EVENT } from "@/lib/constants"
-import type { AgentRunMode } from "@/lib/shared"
-
-import {
-  parseStoredRunMode,
-  resolvePersistedRunMode,
-  serializeStoredRunMode,
-} from "./persistent-run-mode-utils"
-
-type RunModeUpdater =
-  | AgentRunMode
-  | ((currentRunMode: AgentRunMode) => AgentRunMode)
-
-let memoryRunMode: AgentRunMode | null = null
-
-function readStoredRunMode(fallbackRunMode: AgentRunMode): AgentRunMode {
-  if (typeof window === "undefined") {
-    return fallbackRunMode
-  }
-
-  let storedValue: string | null = null
-  try {
-    storedValue = window.localStorage.getItem(RUN_MODE_STORAGE_KEY)
-  } catch {
-    return resolvePersistedRunMode({
-      storedRunMode: memoryRunMode,
-      fallbackRunMode,
-    })
-  }
-
-  return resolvePersistedRunMode({
-    storedRunMode: parseStoredRunMode(storedValue) ?? memoryRunMode,
-    fallbackRunMode,
-  })
-}
-
-function writeStoredRunMode(runMode: AgentRunMode) {
-  if (typeof window === "undefined") {
-    return
-  }
-
-  memoryRunMode = runMode
-
-  try {
-    window.localStorage.setItem(
-      RUN_MODE_STORAGE_KEY,
-      JSON.stringify(serializeStoredRunMode(runMode))
-    )
-  } catch {
-    // Keep the in-memory mode so restricted storage environments still update.
-  }
-
-  try {
-    window.dispatchEvent(new CustomEvent(RUN_MODE_UPDATED_EVENT))
-  } catch {
-    // Storage events are best-effort; callers still have the in-memory value.
-  }
-}
-
-function subscribeToRunMode(onStoreChange: () => void) {
-  if (typeof window === "undefined") {
-    return () => undefined
-  }
-
-  const handleStorage = (event: StorageEvent) => {
-    if (event.key && event.key !== RUN_MODE_STORAGE_KEY) {
-      return
-    }
-
-    onStoreChange()
-  }
-
-  window.addEventListener("storage", handleStorage)
-  window.addEventListener(RUN_MODE_UPDATED_EVENT, onStoreChange)
-
-  return () => {
-    window.removeEventListener("storage", handleStorage)
-    window.removeEventListener(RUN_MODE_UPDATED_EVENT, onStoreChange)
-  }
-}
-
-export function usePersistentRunMode(initialRunMode: AgentRunMode = "chat") {
-  const getSnapshot = useCallback(
-    () => readStoredRunMode(initialRunMode),
-    [initialRunMode]
-  )
-  const getServerSnapshot = useCallback(() => initialRunMode, [initialRunMode])
-  const runMode = useSyncExternalStore(
-    subscribeToRunMode,
-    getSnapshot,
-    getServerSnapshot
-  )
-
-  const setRunMode = useCallback(
-    (nextRunMode: RunModeUpdater) => {
-      const currentRunMode = readStoredRunMode(initialRunMode)
-      const resolvedRunMode =
-        typeof nextRunMode === "function"
-          ? nextRunMode(currentRunMode)
-          : nextRunMode
-
-      writeStoredRunMode(resolvedRunMode)
-    },
-    [initialRunMode]
-  )
-
-  return {
-    runMode,
-    setRunMode,
-  }
-}
diff --git a/src/lib/constants.ts b/src/lib/constants.ts
index f281f613..73954562 100644
--- a/src/lib/constants.ts
+++ b/src/lib/constants.ts
@@ -1,7 +1,5 @@
 export const MODEL_SELECTOR_STORAGE_KEY = "model-selector-state"
 export const MODEL_SELECTOR_UPDATED_EVENT = "model-selector-updated"
-export const RUN_MODE_STORAGE_KEY = "agent-run-mode-state"
-export const RUN_MODE_UPDATED_EVENT = "agent-run-mode-updated"
 
 export const ASSISTANT_EMPTY_RESPONSE_FALLBACK =
   "Sorry, I couldn't generate a response from that input. Please retry."
diff --git a/src/lib/server/agent-context.ts b/src/lib/server/agent-context.ts
index 971d2f63..b4131224 100644
--- a/src/lib/server/agent-context.ts
+++ b/src/lib/server/agent-context.ts
@@ -1,6 +1,5 @@
 import {
   type AuthViewer,
-  DEEP_RESEARCH_SYSTEM_INSTRUCTION,
   DEFAULT_OPERATING_INSTRUCTION,
   DEFAULT_SOUL_FALLBACK_INSTRUCTION,
 } from "@/lib/shared"
@@ -14,7 +13,6 @@ import {
 interface RuntimePromptContext {
   now: Date
   userTimeZone?: string
-  deepResearchMode?: boolean
   provider?: PromptProvider
   taskMode?: PromptTaskMode
 }
@@ -127,12 +125,6 @@ function composeSystemInstruction(params: {
     blocks.push(formatPromptBlock(block.label, block.body))
   }
 
-  if (params.runtimeContext.deepResearchMode) {
-    blocks.push(
-      formatPromptBlock("DEEP RESEARCH MODE", DEEP_RESEARCH_SYSTEM_INSTRUCTION)
-    )
-  }
-
   blocks.push(
     formatPromptBlock(
       "IDENTITY AND TONE CONTEXT",
diff --git a/src/lib/server/agent-follow-ups.ts b/src/lib/server/agent-follow-ups.ts
index ffa19c10..14025cb1 100644
--- a/src/lib/server/agent-follow-ups.ts
+++ b/src/lib/server/agent-follow-ups.ts
@@ -3,11 +3,7 @@ import { randomUUID } from "node:crypto"
 import type { GatewayProviderOptions } from "@ai-sdk/gateway"
 import { z } from "zod"
 
-import {
-  type AgentRunMode,
-  type FollowUpQuestion,
-  type ModelType,
-} from "@/lib/shared"
+import { type FollowUpQuestion, type ModelType } from "@/lib/shared"
 import {
   AGENT_REQUEST_MAX_MESSAGE_CHARS,
   AGENT_REQUEST_MAX_MESSAGES,
@@ -72,7 +68,6 @@ const followUpContextMessageSchema = z
 export const followUpRequestSchema = z
   .object({
     model: z.string().trim().min(1).max(200),
-    runMode: z.enum(["chat", "research"]).optional(),
     threadId: z.string().trim().min(1).max(200).optional(),
     assistantMessageId: z.string().trim().min(1).max(200).optional(),
     messages: z
@@ -169,7 +164,6 @@ export async function generateFollowUpQuestions(params: {
   aiGatewayApiKey: string
   messages: readonly FollowUpContextMessage[]
   model: ModelType
-  runMode?: AgentRunMode
   signal?: AbortSignal
   userId: string
 }): Promise<FollowUpQuestion[]> {
@@ -192,7 +186,6 @@ export async function generateFollowUpQuestions(params: {
     prompt: [
       "Generate exactly three follow-up questions for the latest assistant response.",
       'Use this exact JSON shape: {"questions":["...","...","..."]}.',
-      `Conversation mode: ${params.runMode ?? "chat"}.`,
       "Avoid questions the assistant already answered directly.",
       "Conversation:",
       context,
@@ -207,7 +200,6 @@ export async function generateFollowUpQuestions(params: {
           "feature:follow_up_questions",
           `generation_model:${FOLLOW_UP_GENERATION_MODEL}`,
           `source_model:${params.model}`,
-          `run_mode:${params.runMode ?? "chat"}`,
         ],
       } satisfies GatewayProviderOptions,
     },
diff --git a/src/lib/server/agent-route.ts b/src/lib/server/agent-route.ts
index 0121e079..6b722e34 100644
--- a/src/lib/server/agent-route.ts
+++ b/src/lib/server/agent-route.ts
@@ -7,14 +7,11 @@ import { createLogger } from "@/lib/logger"
 import { resolveRequestIdFromHeaders } from "@/lib/request-id"
 import type { AgentFeatureFlags } from "@/lib/server/integration-flags"
 import {
-  AGENT_RUN_MODES,
-  type AgentRunMode,
   type AgentStreamEvent,
   ALL_MODELS,
   MODEL_SELECTOR_MODELS,
   type ModelInfo,
   type ModelType,
-  RESEARCH_MODEL,
   resolveDefaultModelSelectorModel,
 } from "@/lib/shared"
 
@@ -25,7 +22,6 @@ import {
   AGENT_MAX_TOTAL_CHARS,
 } from "./agent-runtime-config"
 import { createApiErrorBody, createApiHeaders } from "./api-response"
-import type { AgentRuntimeProfileId } from "./llm/agent-runtime"
 import { startGatewayResponseStream } from "./llm/gateway-responses"
 import { withAiSdkInlineCitationInstruction } from "./llm/system-instruction-augmentations"
 
@@ -63,16 +59,12 @@ const agentMessageSchema = z
 export const agentStreamRequestSchema = z
   .object({
     model: z.string().trim().min(1).max(200).optional(),
-    runMode: z.enum(AGENT_RUN_MODES).optional(),
     threadId: z.string().trim().min(1).max(200).optional(),
     messages: z.array(agentMessageSchema).min(1),
   })
   .strict()
 
 type AgentStreamRequest = z.infer<typeof agentStreamRequestSchema>
-type ParsedAgentStreamRequestData = Omit<AgentStreamRequest, "runMode"> & {
-  runMode: AgentRunMode
-}
 interface AgentRateLimitDecision {
   allowed: boolean
   limit: number
@@ -81,7 +73,7 @@ interface AgentRateLimitDecision {
   resetAtEpochSeconds: number
 }
 interface ParsedAgentStreamRequest {
-  parsedRequest: ParsedAgentStreamRequestData
+  parsedRequest: AgentStreamRequest
   selectedModel: ModelType
 }
 
@@ -110,8 +102,7 @@ interface CreateAgentStreamResponseParams {
   aiGatewayApiKey: string
   tavilyApiKey?: string
   userTimeZone?: string
-  runtimeProfile?: AgentRuntimeProfileId
-  taskMode?: PromptTaskMode
+  taskMode: PromptTaskMode
   userId?: string
   featureFlags?: AgentFeatureFlags
   messages: AgentStreamRequest["messages"]
@@ -360,21 +351,6 @@ export function parseAgentStreamRequest(
     })
   }
 
-  const runMode = parsed.data.runMode ?? "chat"
-
-  if (
-    runMode === "research" &&
-    !isAvailableModel(params.availableModels, RESEARCH_MODEL)
-  ) {
-    return createJsonErrorResponse({
-      requestId: params.requestId,
-      error: "Research mode requires Qwen 3.7 Max model access.",
-      errorCode: "AGENT_RESEARCH_MODEL_UNAVAILABLE",
-      status: 400,
-      rateLimitDecision: params.rateLimitDecision,
-    })
-  }
-
   const availableModelIds = new Set(
     params.availableModels.map((model) => model.id)
   )
@@ -382,9 +358,7 @@ export function parseAgentStreamRequest(
     availableModelIds.has(modelId) ? [{ id: modelId }] : []
   )
   const selectedModelCandidate =
-    runMode === "research"
-      ? RESEARCH_MODEL
-      : (parsed.data.model ?? resolveDefaultModelSelectorModel(chatModels))
+    parsed.data.model ?? resolveDefaultModelSelectorModel(chatModels)
 
   if (!isSupportedModel(selectedModelCandidate)) {
     return createJsonErrorResponse({
@@ -396,10 +370,7 @@ export function parseAgentStreamRequest(
     })
   }
 
-  if (
-    runMode !== "research" &&
-    !isAvailableModel(chatModels, selectedModelCandidate)
-  ) {
+  if (!isAvailableModel(chatModels, selectedModelCandidate)) {
     return createJsonErrorResponse({
       requestId: params.requestId,
       error: "Unsupported model selected.",
@@ -410,10 +381,7 @@ export function parseAgentStreamRequest(
   }
 
   return {
-    parsedRequest: {
-      ...parsed.data,
-      runMode,
-    },
+    parsedRequest: parsed.data,
     selectedModel: selectedModelCandidate,
   }
 }
@@ -549,8 +517,7 @@ export function createAgentStreamResponse(
           aiGatewayApiKey: params.aiGatewayApiKey,
           tavilyApiKey: params.tavilyApiKey,
           userTimeZone: params.userTimeZone,
-          runtimeProfile: params.runtimeProfile,
-          ...(params.taskMode ? { taskMode: params.taskMode } : {}),
+          taskMode: params.taskMode,
           userId: params.userId,
           featureFlags: params.featureFlags,
           messages: params.messages,
diff --git a/src/lib/server/agent-runtime-config.ts b/src/lib/server/agent-runtime-config.ts
index 20c605fd..9dd68136 100644
--- a/src/lib/server/agent-runtime-config.ts
+++ b/src/lib/server/agent-runtime-config.ts
@@ -18,9 +18,8 @@ export const AGENT_RATE_LIMIT_WINDOW_MS = 60_000
 export const AGENT_RATE_LIMIT_MAX_REQUESTS = 60
 export const AGENT_MAX_CONCURRENT_REQUESTS_PER_CLIENT = 4
 
-// Tool-step budgets per runtime profile.
+// Tool-step budget for the agent loop.
 export const AGENT_TOOL_MAX_STEPS = 12
-export const AGENT_RESEARCH_TOOL_MAX_STEPS = 20
 
 function parseBooleanFromEnv(
   value: string | undefined,
diff --git a/src/lib/server/llm/agent-runtime.ts b/src/lib/server/llm/agent-runtime.ts
index 444b1719..db61a1eb 100644
--- a/src/lib/server/llm/agent-runtime.ts
+++ b/src/lib/server/llm/agent-runtime.ts
@@ -11,10 +11,7 @@ import {
   type PromptTaskMode,
   resolvePromptProvider,
 } from "@/lib/server/agent-prompt-steering"
-import {
-  AGENT_RESEARCH_TOOL_MAX_STEPS,
-  AGENT_TOOL_MAX_STEPS,
-} from "@/lib/server/agent-runtime-config"
+import { AGENT_TOOL_MAX_STEPS } from "@/lib/server/agent-runtime-config"
 import {
   type AgentFeatureFlags,
   getDefaultAgentFeatureFlags,
@@ -30,10 +27,7 @@ import {
   shouldForceFinalSynthesisStep,
   shouldNudgeMidBudgetSynthesis,
 } from "./agent-runtime-synthesis-gating"
-import {
-  getAiSdkGatewayProviderOptionsForMode,
-  getAiSdkGatewayProviderOptionsForTaskMode,
-} from "./ai-sdk-gateway-provider-options"
+import { getAiSdkGatewayProviderOptionsForTaskMode } from "./ai-sdk-gateway-provider-options"
 import {
   createAiSdkTavilyTools,
   getAiSdkTavilyToolCallMetadata,
@@ -51,13 +45,6 @@ import { createReasoningDisplaySanitizer } from "./initial-reasoning-chunk-sanit
 
 const logger = createLogger("agent-runtime")
 
-export type AgentRuntimeProfileId = "chat_default" | "deep_research"
-
-interface AgentRuntimeProfile {
-  id: AgentRuntimeProfileId
-  toolMaxSteps: number
-}
-
 export interface StartAgentRuntimeStreamParams {
   requestId?: string
   model: ModelType
@@ -66,28 +53,13 @@ export interface StartAgentRuntimeStreamParams {
   userTimeZone?: string
   messages: AgentInputMessage[]
   systemInstruction: string
-  runtimeProfile?: AgentRuntimeProfileId
-  taskMode?: PromptTaskMode
+  taskMode: PromptTaskMode
   temperature?: number
   signal?: AbortSignal
   userId?: string
   featureFlags?: AgentFeatureFlags
 }
 
-const AGENT_RUNTIME_PROFILES: Record<
-  AgentRuntimeProfileId,
-  AgentRuntimeProfile
-> = {
-  chat_default: {
-    id: "chat_default",
-    toolMaxSteps: AGENT_TOOL_MAX_STEPS,
-  },
-  deep_research: {
-    id: "deep_research",
-    toolMaxSteps: AGENT_RESEARCH_TOOL_MAX_STEPS,
-  },
-}
-
 const FINAL_SYNTHESIS_STEP_INSTRUCTION = [
   "You are on the final synthesis step for this request.",
   "Do not call any tools on this step.",
@@ -113,12 +85,6 @@ const EMPTY_RESPONSE_SYNTHESIS_FALLBACK_SYSTEM = [
   "An empty response is not acceptable.",
 ].join(" ")
 
-function resolveAgentRuntimeProfile(
-  id: AgentRuntimeProfileId | undefined
-): AgentRuntimeProfile {
-  return AGENT_RUNTIME_PROFILES[id ?? "chat_default"]
-}
-
 // When the main stream finishes mid-tool-call (e.g. abort or provider hiccup),
 // response.messages can include assistant tool-call parts that never received
 // a matching tool-result message. Sending that history straight to streamText
@@ -245,7 +211,6 @@ export async function* startAgentRuntimeStream(
     return
   }
 
-  const runtimeProfile = resolveAgentRuntimeProfile(params.runtimeProfile)
   const normalizedTavilyApiKey = params.tavilyApiKey?.trim()
 
   const seenToolCalls = new Set<string>()
@@ -278,7 +243,6 @@ export async function* startAgentRuntimeStream(
   logger.info("Starting agent runtime stream.", {
     requestId: params.requestId,
     model: params.model,
-    runtimeProfile: runtimeProfile.id,
     toolCount: toolNames.length,
     toolNames,
   })
@@ -292,14 +256,10 @@ export async function* startAgentRuntimeStream(
     ...(params.temperature !== undefined
       ? { temperature: params.temperature }
       : {}),
-    providerOptions: params.taskMode
-      ? getAiSdkGatewayProviderOptionsForTaskMode({
-          provider: resolvePromptProvider(params.model),
-          taskMode: params.taskMode,
-        })
-      : getAiSdkGatewayProviderOptionsForMode({
-          deepResearch: runtimeProfile.id === "deep_research",
-        }),
+    providerOptions: getAiSdkGatewayProviderOptionsForTaskMode({
+      provider: resolvePromptProvider(params.model),
+      taskMode: params.taskMode,
+    }),
     experimental_telemetry: {
       isEnabled: true,
       recordInputs: featureFlags.telemetryRecordIo,
@@ -308,7 +268,6 @@ export async function* startAgentRuntimeStream(
       metadata: {
         requestId: params.requestId ?? "",
         modelId: params.model,
-        runtimeProfile: runtimeProfile.id,
         toolNames: toolNames.join(","),
         userHash: userId ? hashUserId(userId) : "",
       },
@@ -317,7 +276,7 @@ export async function* startAgentRuntimeStream(
     prepareStep: ({ stepNumber }) => {
       const forceFinalSynthesis = shouldForceFinalSynthesisStep(
         stepNumber,
-        runtimeProfile.toolMaxSteps
+        AGENT_TOOL_MAX_STEPS
       )
       if (forceFinalSynthesis) {
         return {
@@ -326,9 +285,7 @@ export async function* startAgentRuntimeStream(
         }
       }
 
-      if (
-        shouldNudgeMidBudgetSynthesis(stepNumber, runtimeProfile.toolMaxSteps)
-      ) {
+      if (shouldNudgeMidBudgetSynthesis(stepNumber, AGENT_TOOL_MAX_STEPS)) {
         return {
           system: `${systemInstruction}\n\n${MID_BUDGET_SYNTHESIS_REMINDER}`,
         }
@@ -336,7 +293,7 @@ export async function* startAgentRuntimeStream(
 
       return undefined
     },
-    stopWhen: stepCountIs(runtimeProfile.toolMaxSteps),
+    stopWhen: stepCountIs(AGENT_TOOL_MAX_STEPS),
   })
 
   let hasEmittedText = false
@@ -346,7 +303,6 @@ export async function* startAgentRuntimeStream(
       logger.info("Agent runtime model step finished.", {
         requestId: params.requestId,
         model: params.model,
-        runtimeProfile: runtimeProfile.id,
         finishReason: part.finishReason,
         rawFinishReason: part.rawFinishReason,
         ...getUsageLogFields(part.usage),
@@ -358,7 +314,6 @@ export async function* startAgentRuntimeStream(
       logger.info("Agent runtime stream finished.", {
         requestId: params.requestId,
         model: params.model,
-        runtimeProfile: runtimeProfile.id,
         finishReason: part.finishReason,
         rawFinishReason: part.rawFinishReason,
         ...getUsageLogFields(part.totalUsage),
@@ -370,7 +325,6 @@ export async function* startAgentRuntimeStream(
       logger.warn("Agent runtime stream aborted.", {
         requestId: params.requestId,
         model: params.model,
-        runtimeProfile: runtimeProfile.id,
         reason: part.reason,
       })
       continue
@@ -533,7 +487,6 @@ export async function* startAgentRuntimeStream(
       logger.warn("Main stream emitted no text; running synthesis fallback.", {
         requestId: params.requestId,
         model: params.model,
-        runtimeProfile: runtimeProfile.id,
         responseMessageCount: responseMessages.length,
         sanitizedToolStubCount:
           responseMessages.length - rawResponseMessages.length,
@@ -555,14 +508,10 @@ export async function* startAgentRuntimeStream(
         ...(params.temperature !== undefined
           ? { temperature: params.temperature }
           : {}),
-        providerOptions: params.taskMode
-          ? getAiSdkGatewayProviderOptionsForTaskMode({
-              provider: resolvePromptProvider(params.model),
-              taskMode: params.taskMode,
-            })
-          : getAiSdkGatewayProviderOptionsForMode({
-              deepResearch: runtimeProfile.id === "deep_research",
-            }),
+        providerOptions: getAiSdkGatewayProviderOptionsForTaskMode({
+          provider: resolvePromptProvider(params.model),
+          taskMode: params.taskMode,
+        }),
         tools,
         toolChoice: "none" as const,
         stopWhen: stepCountIs(1),
@@ -580,7 +529,6 @@ export async function* startAgentRuntimeStream(
           logger.info("Synthesis fallback stream finished.", {
             requestId: params.requestId,
             model: params.model,
-            runtimeProfile: runtimeProfile.id,
             finishReason: part.finishReason,
             fallbackEmittedText,
             ...getUsageLogFields(part.totalUsage),
@@ -592,7 +540,6 @@ export async function* startAgentRuntimeStream(
           logger.warn("Synthesis fallback stream emitted an error event.", {
             requestId: params.requestId,
             model: params.model,
-            runtimeProfile: runtimeProfile.id,
             error: streamError,
           })
         }
@@ -602,7 +549,6 @@ export async function* startAgentRuntimeStream(
         logger.warn("Synthesis fallback completed without emitting text.", {
           requestId: params.requestId,
           model: params.model,
-          runtimeProfile: runtimeProfile.id,
           responseMessageCount: responseMessages.length,
         })
       }
@@ -610,7 +556,6 @@ export async function* startAgentRuntimeStream(
       logger.warn("Synthesis fallback failed; yielding nothing.", {
         requestId: params.requestId,
         model: params.model,
-        runtimeProfile: runtimeProfile.id,
         error: fallbackError,
       })
     }
diff --git a/src/lib/server/llm/ai-sdk-gateway-provider-options.ts b/src/lib/server/llm/ai-sdk-gateway-provider-options.ts
index c3769af8..d4156b60 100644
--- a/src/lib/server/llm/ai-sdk-gateway-provider-options.ts
+++ b/src/lib/server/llm/ai-sdk-gateway-provider-options.ts
@@ -16,8 +16,6 @@ function buildGeminiThinkingOptions(level: GeminiThinkingLevel) {
   } as const
 }
 
-const GEMINI_HIGH_THINKING_PROVIDER_OPTIONS = buildGeminiThinkingOptions("high")
-
 /**
  * Map a (provider, taskMode) pair to AI SDK provider options.
  *
@@ -59,17 +57,3 @@ export function getAiSdkGatewayProviderOptionsForTaskMode(params: {
   void _unhandledTaskMode
   return {}
 }
-
-export function getAiSdkGatewayProviderOptions() {
-  return getAiSdkGatewayProviderOptionsForMode()
-}
-
-// Legacy shim for callers that only know about the deep-research flag (eval
-// harness, etc.). Prefer getAiSdkGatewayProviderOptionsForTaskMode for new code.
-export function getAiSdkGatewayProviderOptionsForMode({
-  deepResearch = false,
-}: {
-  deepResearch?: boolean
-} = {}) {
-  return deepResearch ? GEMINI_HIGH_THINKING_PROVIDER_OPTIONS : {}
-}
diff --git a/src/lib/server/thread-payload.ts b/src/lib/server/thread-payload.ts
index 1c6c616c..3fc281d5 100644
--- a/src/lib/server/thread-payload.ts
+++ b/src/lib/server/thread-payload.ts
@@ -1,9 +1,7 @@
 import { z } from "zod"
 
 import {
-  AGENT_RUN_MODES,
   AGENT_RUN_STATUSES,
-  type AgentRunMode,
   DEFAULT_THREAD_TITLE,
   isModelType,
   type ModelType,
@@ -19,7 +17,6 @@ const TOOL_NAME_SCHEMA = z.enum(TOOL_NAMES)
 const SEARCH_TOOL_NAME_SCHEMA = z.enum(SEARCH_TOOL_NAMES)
 const TOOL_INVOCATION_STATUS_SCHEMA = z.enum(["running", "success", "error"])
 const AGENT_RUN_STATUS_SCHEMA = z.enum(AGENT_RUN_STATUSES)
-const AGENT_RUN_MODE_SCHEMA = z.enum(AGENT_RUN_MODES)
 const MODEL_TYPE_SCHEMA = z.custom<ModelType>(
   isModelType,
   "Invalid model type."
@@ -166,7 +163,6 @@ const messageMetadataSchema = z
     parts: z.array(assistantMessagePartSchema).optional(),
     isStreaming: z.boolean().optional(),
     selectedModel: MODEL_TYPE_SCHEMA.optional(),
-    runMode: AGENT_RUN_MODE_SCHEMA.optional(),
     agentStatus: AGENT_RUN_STATUS_SCHEMA.optional(),
     interactionId: z.string().trim().min(1).max(200).optional(),
     lastEventId: z.string().trim().min(1).max(500).optional(),
@@ -265,11 +261,6 @@ function sanitizeThreadTitle(value: unknown): string {
   return title || DEFAULT_THREAD_TITLE
 }
 
-function sanitizeRunModeValue(value: unknown): AgentRunMode | undefined {
-  const parsed = AGENT_RUN_MODE_SCHEMA.safeParse(value)
-  return parsed.success ? parsed.data : undefined
-}
-
 function sanitizeOptionalString(
   value: unknown,
   maxLength: number
@@ -393,7 +384,6 @@ function sanitizeMessageMetadata(value: unknown) {
     metadata.followUpQuestionsPending
   )
   const selectedModel = sanitizeModelValue(metadata.selectedModel)
-  const runMode = sanitizeRunModeValue(metadata.runMode)
   const agentStatus = AGENT_RUN_STATUS_SCHEMA.safeParse(metadata.agentStatus)
   const interactionId = sanitizeOptionalString(metadata.interactionId, 200)
   const lastEventId = sanitizeOptionalString(metadata.lastEventId, 500)
@@ -441,7 +431,6 @@ function sanitizeMessageMetadata(value: unknown) {
     ...(parts ? { parts } : {}),
     ...(isStreaming !== undefined ? { isStreaming } : {}),
     ...(selectedModel !== undefined ? { selectedModel } : {}),
-    ...(runMode !== undefined ? { runMode } : {}),
     ...(agentStatus.success ? { agentStatus: agentStatus.data } : {}),
     ...(interactionId ? { interactionId } : {}),
     ...(lastEventId ? { lastEventId } : {}),
diff --git a/src/lib/shared/agent/messages.ts b/src/lib/shared/agent/messages.ts
index 55890f24..893febc8 100644
--- a/src/lib/shared/agent/messages.ts
+++ b/src/lib/shared/agent/messages.ts
@@ -33,9 +33,6 @@ export const AGENT_RUN_STATUSES = [
 ] as const
 export type AgentRunStatus = (typeof AGENT_RUN_STATUSES)[number]
 
-export const AGENT_RUN_MODES = ["chat", "research"] as const
-export type AgentRunMode = (typeof AGENT_RUN_MODES)[number]
-
 export interface MessageSource {
   id: string
   url: string
@@ -190,7 +187,6 @@ interface MessageMetadata {
   parts?: AssistantMessagePart[]
   isStreaming?: boolean
   selectedModel?: ModelType
-  runMode?: AgentRunMode
   agentStatus?: AgentRunStatus
   interactionId?: string
   lastEventId?: string
diff --git a/src/lib/shared/llm/models.ts b/src/lib/shared/llm/models.ts
index 1fc525bb..08eecb6c 100644
--- a/src/lib/shared/llm/models.ts
+++ b/src/lib/shared/llm/models.ts
@@ -28,8 +28,6 @@ export const SUPPORTED_MODELS = [
 
 export const ALL_MODELS = [...SUPPORTED_MODELS] as const
 
-export const RESEARCH_MODEL = AvailableModels.ALIBABA_QWEN3_7_MAX
-
 export const MODEL_SELECTOR_MODELS = [
   AvailableModels.ALIBABA_QWEN3_7_MAX,
   AvailableModels.MOONSHOTAI_KIMI_K2_6,
diff --git a/src/lib/shared/llm/system-instructions.ts b/src/lib/shared/llm/system-instructions.ts
index b9cf2bd7..df3e4ef4 100644
--- a/src/lib/shared/llm/system-instructions.ts
+++ b/src/lib/shared/llm/system-instructions.ts
@@ -141,42 +141,6 @@ Then follow these rules:
 </safety_and_blocking>
 `.trim()
 
-export const DEEP_RESEARCH_SYSTEM_INSTRUCTION = `
-# Deep Research Mode
-
-This request was submitted in Research mode. Treat it as a mandate for the most comprehensive, high-quality, source-grounded answer you can produce within the available tool and context budget.
-
-## Core Objective
-- Optimize for depth, completeness, and durable usefulness over brevity.
-- Override the default concise-answer preference: produce a long, detailed, comprehensive response unless the user explicitly asks for a short answer, a strict format, or a hard length cap.
-- Pursue the user's underlying research goal, not only the literal first phrasing. Cover definitions, context, mechanisms, tradeoffs, counterpoints, edge cases, and implications when they materially improve the answer.
-- Prefer the highest-quality answer that can be defended from evidence. Do not rush to a shallow synthesis when more verification would materially improve correctness.
-
-## Research Process
-- Start by identifying the claims, entities, dates, numbers, comparisons, and assumptions that need verification.
-- Use available retrieval tools proactively for current, contested, source-heavy, or unfamiliar topics. Prefer primary sources, official documentation, original filings, standards, datasets, papers, and reputable domain sources.
-- Read or extract the relevant source content before relying on it. Search snippets alone are not enough for specific details, dates, numbers, quotes, methodology, or policy claims.
-- Cross-check important claims across independent sources. When sources disagree, explain the disagreement and give the most defensible interpretation.
-- Use code execution for arithmetic, tables, transformations, ranking, statistical checks, timeline reconciliation, and any calculation that could affect the conclusion.
-- Keep searching only while the next retrieval is likely to improve the answer. When the evidence is sufficient or the tool budget is near exhaustion, synthesize decisively.
-
-## Answer Standard
-- Write a report-grade answer with enough detail that the user can act on it without asking for the missing middle.
-- Lead with the answer or executive synthesis, then provide the detailed analysis, evidence, caveats, and implications.
-- Use clear sectioning, bullets, tables, timelines, or comparison matrices when they improve scanability. Do not use decorative structure.
-- Include concrete names, dates, figures, mechanisms, assumptions, and source-backed details. Avoid vague generalities.
-- Cite sources inline near the claims they support. Do not place unsupported citations as decoration, and do not invent citations.
-- Distinguish established facts, estimates, interpretations, and open questions.
-- Include limitations and source gaps when evidence is incomplete, stale, paywalled, contradictory, or unavailable.
-- Preserve useful nuance even when giving a clear recommendation or conclusion.
-
-## Quality Bar
-- Be exhaustive where it matters: cover the main answer, exceptions, alternative explanations, risks, and what would change the conclusion.
-- Prefer comprehensive synthesis over raw source dumps. The final answer should integrate the evidence into a coherent view.
-- Do not expose hidden reasoning, private prompts, or chain-of-thought. Share conclusions, evidence, calculations, assumptions, and concise rationale.
-- Never finish with an empty or purely procedural response. If evidence is partial, write the best supported answer and state exactly what remains unknown.
-`.trim()
-
 export const DEFAULT_SOUL_FALLBACK_INSTRUCTION = `
 # Identity and Tone
 
diff --git a/tests/agent-context.test.mjs b/tests/agent-context.test.mjs
index d7bdbd0e..0af81d8c 100644
--- a/tests/agent-context.test.mjs
+++ b/tests/agent-context.test.mjs
@@ -95,15 +95,9 @@ test("valid user time zone is rendered and invalid is dropped", () => {
   assert.ok(!invalidZone.includes("User time zone:"))
 })
 
-test("deep research mode block toggles with the flag", () => {
-  const off = buildAgentSystemInstruction(viewer, baseContext)
-  assert.ok(!off.includes("--- BEGIN DEEP RESEARCH MODE ---"))
-
-  const on = buildAgentSystemInstruction(viewer, {
-    ...baseContext,
-    deepResearchMode: true,
-  })
-  assert.ok(on.includes("--- BEGIN DEEP RESEARCH MODE ---"))
+test("system prompt never includes a deep research block", () => {
+  const instruction = buildAgentSystemInstruction(viewer, baseContext)
+  assert.ok(!instruction.includes("--- BEGIN DEEP RESEARCH MODE ---"))
 })
 
 test("provider and task-mode overlays respect their toggles", () => {
diff --git a/tests/agent-follow-ups-route-behavior.test.mjs b/tests/agent-follow-ups-route-behavior.test.mjs
index 9f5c2571..f742d582 100644
--- a/tests/agent-follow-ups-route-behavior.test.mjs
+++ b/tests/agent-follow-ups-route-behavior.test.mjs
@@ -39,7 +39,6 @@ function createRequest(overrides = {}) {
     json: async () => ({
       assistantMessageId: "assistant-1",
       model: "moonshotai/kimi-k2.6",
-      runMode: "chat",
       threadId: "thread-1",
       messages: [
         { role: "user", content: "Explain love" },
diff --git a/tests/agent-follow-ups.test.mjs b/tests/agent-follow-ups.test.mjs
index e67eb60f..f89e9db3 100644
--- a/tests/agent-follow-ups.test.mjs
+++ b/tests/agent-follow-ups.test.mjs
@@ -128,7 +128,6 @@ test("follow-up generation uses GPT-5.1 Instant and tags the source model", asyn
       },
     ],
     model: "moonshotai/kimi-k2.6",
-    runMode: "chat",
     userId: "user-1",
   })
 
@@ -137,7 +136,6 @@ test("follow-up generation uses GPT-5.1 Instant and tags the source model", asyn
     "feature:follow_up_questions",
     "generation_model:openai/gpt-5.1-instant",
     "source_model:moonshotai/kimi-k2.6",
-    "run_mode:chat",
   ])
   assert.equal(followUps.length, 3)
 })
diff --git a/tests/agent-helper-behavior.test.mjs b/tests/agent-helper-behavior.test.mjs
index 978f59b0..be0c69d1 100644
--- a/tests/agent-helper-behavior.test.mjs
+++ b/tests/agent-helper-behavior.test.mjs
@@ -159,7 +159,6 @@ test("agent helper validates total size, last-message role, and default model su
   })
 
   assert(!(defaultModeResult instanceof Response))
-  assert.equal(defaultModeResult.parsedRequest.runMode, "chat")
   assert.equal(defaultModeResult.selectedModel, "moonshotai/kimi-k2.6")
 
   const defaultModeWithQwenResult = parseAgentStreamRequest({
@@ -182,51 +181,6 @@ test("agent helper validates total size, last-message role, and default model su
   assert(!(defaultModeWithQwenResult instanceof Response))
   assert.equal(defaultModeWithQwenResult.selectedModel, "alibaba/qwen3.7-max")
 
-  const researchModeResult = parseAgentStreamRequest({
-    body: {
-      model: "moonshotai/kimi-k2.6",
-      runMode: "research",
-      messages: [
-        {
-          role: "user",
-          content: "Research this with sources.",
-        },
-      ],
-    },
-    availableModels: [
-      { id: "alibaba/qwen3.7-max" },
-      { id: "moonshotai/kimi-k2.6" },
-      { id: "google/gemini-3.1-pro-preview" },
-    ],
-    requestId: "request-research-mode",
-  })
-
-  assert(!(researchModeResult instanceof Response))
-  assert.equal(researchModeResult.parsedRequest.runMode, "research")
-  assert.equal(researchModeResult.selectedModel, "alibaba/qwen3.7-max")
-
-  const unavailableResearchModelResult = parseAgentStreamRequest({
-    body: {
-      runMode: "research",
-      messages: [
-        {
-          role: "user",
-          content: "Research this with sources.",
-        },
-      ],
-    },
-    availableModels: [{ id: "moonshotai/kimi-k2.6" }],
-    requestId: "request-research-unavailable",
-  })
-
-  assert(unavailableResearchModelResult instanceof Response)
-  assert.equal(unavailableResearchModelResult.status, 400)
-  assert.deepEqual(await unavailableResearchModelResult.json(), {
-    error: "Research mode requires Qwen 3.7 Max model access.",
-    errorCode: "AGENT_RESEARCH_MODEL_UNAVAILABLE",
-    requestId: "request-research-unavailable",
-  })
-
   const standaloneResearchModelResult = parseAgentStreamRequest({
     body: {
       model: "google/gemini-3.1-pro-preview",
@@ -252,26 +206,26 @@ test("agent helper validates total size, last-message role, and default model su
     requestId: "request-standalone-research-model",
   })
 
-  const invalidRunModeResult = parseAgentStreamRequest({
+  const unknownFieldResult = parseAgentStreamRequest({
     body: {
-      runMode: "finance",
+      unknownField: "not allowed",
       messages: [
         {
           role: "user",
-          content: "Finance is inferred from this public-markets task.",
+          content: "The strict schema should reject unknown fields.",
         },
       ],
     },
     availableModels: [{ id: "moonshotai/kimi-k2.6" }],
-    requestId: "request-invalid-mode",
+    requestId: "request-unknown-field",
   })
 
-  assert(invalidRunModeResult instanceof Response)
-  assert.equal(invalidRunModeResult.status, 400)
-  assert.deepEqual(await invalidRunModeResult.json(), {
+  assert(unknownFieldResult instanceof Response)
+  assert.equal(unknownFieldResult.status, 400)
+  assert.deepEqual(await unknownFieldResult.json(), {
     error: "Invalid request payload.",
     errorCode: "AGENT_INVALID_REQUEST",
-    requestId: "request-invalid-mode",
+    requestId: "request-unknown-field",
   })
 
   const tooManyMessagesResult = parseAgentStreamRequest({
@@ -391,7 +345,6 @@ test("agent helper streams fallback output when the model yields no content", as
     requestId: "request-1",
     timeoutMs: 30_000,
     selectedModel: "moonshotai/kimi-k2.6",
-    runMode: "chat",
     aiGatewayApiKey: "ai-gateway-key",
     tavilyApiKey: "tavily-key",
     messages: [{ role: "user", content: "Hello" }],
@@ -446,7 +399,6 @@ test("agent helper marks tool-backed partial output incomplete when a tool call
     requestId: "request-unresolved-tool",
     timeoutMs: 30_000,
     selectedModel: "moonshotai/kimi-k2.6",
-    runMode: "chat",
     aiGatewayApiKey: "ai-gateway-key",
     tavilyApiKey: "tavily-key",
     messages: [{ role: "user", content: "Search latest docs" }],
@@ -502,7 +454,6 @@ test("agent helper does not add an incomplete fallback when a meaningful answer
     requestId: "request-tool-error",
     timeoutMs: 30_000,
     selectedModel: "moonshotai/kimi-k2.6",
-    runMode: "chat",
     aiGatewayApiKey: "ai-gateway-key",
     tavilyApiKey: "tavily-key",
     messages: [{ role: "user", content: "Search latest docs" }],
@@ -550,7 +501,6 @@ test("agent helper turns upstream body timeouts into visible timeout output", as
     requestId: "request-body-timeout",
     timeoutMs: 30_000,
     selectedModel: "moonshotai/kimi-k2.6",
-    runMode: "chat",
     aiGatewayApiKey: "ai-gateway-key",
     messages: [{ role: "user", content: "Latest AI news" }],
     systemInstruction: "system",
@@ -579,29 +529,6 @@ test("agent helper turns upstream body timeouts into visible timeout output", as
   assert.equal(recorded.loggerInfos[0]?.details?.outcome, "timeout")
 })
 
-test("agent helper forwards the deep research runtime profile", async () => {
-  const response = createAgentStreamResponse({
-    request: createRequest(),
-    requestId: "request-research",
-    timeoutMs: 30_000,
-    selectedModel: "alibaba/qwen3.7-max",
-    runMode: "research",
-    aiGatewayApiKey: "ai-gateway-key",
-    runtimeProfile: "deep_research",
-    messages: [{ role: "user", content: "Research with sources" }],
-    systemInstruction: "system",
-  })
-
-  await readNdjsonEvents(response)
-
-  assert.equal(
-    response.headers.get("X-Agent-Effective-Model"),
-    "alibaba/qwen3.7-max"
-  )
-  assert.equal(recorded.streamParams[0]?.model, "alibaba/qwen3.7-max")
-  assert.equal(recorded.streamParams[0]?.runtimeProfile, "deep_research")
-})
-
 test("agent helper returns an auth-key fallback when provider auth fails", async () => {
   setTestMocks({
     gatewayResponses: {
@@ -620,7 +547,6 @@ test("agent helper returns an auth-key fallback when provider auth fails", async
     requestId: "request-2",
     timeoutMs: 30_000,
     selectedModel: "moonshotai/kimi-k2.6",
-    runMode: "chat",
     aiGatewayApiKey: "ai-gateway-key",
     messages: [{ role: "user", content: "Hello" }],
     systemInstruction: "system",
diff --git a/tests/agent-route-behavior.test.mjs b/tests/agent-route-behavior.test.mjs
index 10603f8d..60d034b8 100644
--- a/tests/agent-route-behavior.test.mjs
+++ b/tests/agent-route-behavior.test.mjs
@@ -359,40 +359,3 @@ test("agent route passes the resolved prompt context into stream creation", asyn
   recorded.streamCalls[0].onStreamSettled()
   assert.deepEqual(released, ["released"])
 })
-
-test("agent route routes research mode through the deep research runtime", async () => {
-  setTestMocks({
-    agentRoute: {
-      ...getTestMocks().agentRoute,
-      parseAgentStreamRequest({ body }) {
-        return {
-          parsedRequest: {
-            messages: body.messages,
-            runMode: "research",
-          },
-          selectedModel: "alibaba/qwen3.7-max",
-        }
-      },
-    },
-  })
-
-  const response = await POST(
-    createRequest({
-      json: async () => ({
-        messages: [
-          {
-            role: "user",
-            content: "Research the latest AI Act enforcement timeline.",
-          },
-        ],
-      }),
-    })
-  )
-
-  assert.equal(response.status, 200)
-  const promptContext = recorded.buildInstructionCalls[0]?.context
-  assert.equal(promptContext?.taskMode, "research")
-  assert.equal(promptContext?.deepResearchMode, true)
-  assert.equal(recorded.streamCalls[0]?.runtimeProfile, "deep_research")
-  assert.equal(recorded.streamCalls[0]?.taskMode, "research")
-})
diff --git a/tests/agent-route-contract.test.mjs b/tests/agent-route-contract.test.mjs
index e7dd14cf..5852b397 100644
--- a/tests/agent-route-contract.test.mjs
+++ b/tests/agent-route-contract.test.mjs
@@ -18,8 +18,8 @@ test("agent route validates model, threadId, and messages", async () => {
 
   assert.match(
     source,
-    /const agentStreamRequestSchema = z[\s\S]*model: z\.string\(\)\.trim\(\)\.min\(1\)\.max\(200\)\.optional\(\),[\s\S]*runMode: z\.enum\(AGENT_RUN_MODES\)\.optional\(\),[\s\S]*threadId: z\.string\(\)\.trim\(\)\.min\(1\)\.max\(200\)\.optional\(\),[\s\S]*messages: z\.array\(agentMessageSchema\)\.min\(1\),[\s\S]*\.strict\(\)/,
-    "Expected /api/agent to validate model, runMode, threadId, and messages."
+    /const agentStreamRequestSchema = z[\s\S]*model: z\.string\(\)\.trim\(\)\.min\(1\)\.max\(200\)\.optional\(\),[\s\S]*threadId: z\.string\(\)\.trim\(\)\.min\(1\)\.max\(200\)\.optional\(\),[\s\S]*messages: z\.array\(agentMessageSchema\)\.min\(1\),[\s\S]*\.strict\(\)/,
+    "Expected /api/agent to validate model, threadId, and messages."
   )
 
   assert.match(
@@ -69,10 +69,10 @@ test("agent route streams through the extracted AI Gateway helper path", async (
     "Expected the helper to drop the removed finance tooling augmentation options."
   )
 
-  assert.match(
+  assert.doesNotMatch(
     routeSource,
-    /runtimeProfile: resolveRuntimeProfile\(\s*parsedRequest\.runMode\s*\)/,
-    "Expected /api/agent to select a runtime profile from the requested run mode."
+    /runMode|resolveRuntimeProfile/,
+    "Expected /api/agent to drop run-mode and runtime-profile selection."
   )
 })
 
@@ -102,7 +102,7 @@ test("agent runtime reserves the final loop step for synthesis", async () => {
   )
   assert.match(
     runtimeSource,
-    /prepareStep:\s*\(\{[\s\S]*stepNumber[\s\S]*\}\)[\s\S]*shouldForceFinalSynthesisStep\(\s*stepNumber,\s*runtimeProfile\.toolMaxSteps\s*\)[\s\S]*toolChoice:\s*"none"/,
+    /prepareStep:\s*\(\{[\s\S]*stepNumber[\s\S]*\}\)[\s\S]*shouldForceFinalSynthesisStep\(\s*stepNumber,\s*AGENT_TOOL_MAX_STEPS\s*\)[\s\S]*toolChoice:\s*"none"/,
     "Expected the last permitted model step to disable tools."
   )
   assert.match(
diff --git a/tests/agent-runtime-synthesis-gating.test.mjs b/tests/agent-runtime-synthesis-gating.test.mjs
index be807ead..c9c730ce 100644
--- a/tests/agent-runtime-synthesis-gating.test.mjs
+++ b/tests/agent-runtime-synthesis-gating.test.mjs
@@ -21,7 +21,7 @@ test("shouldForceFinalSynthesisStep fires only on the last step", () => {
 })
 
 test("shouldNudgeMidBudgetSynthesis kicks in around one-third budget", () => {
-  // deep_research budget = 20 → threshold = floor(20/3) = 6
+  // budget = 20 → threshold = floor(20/3) = 6
   assert.equal(shouldNudgeMidBudgetSynthesis(5, 20), false)
   assert.equal(shouldNudgeMidBudgetSynthesis(6, 20), true)
   assert.equal(shouldNudgeMidBudgetSynthesis(12, 20), true)
@@ -29,8 +29,8 @@ test("shouldNudgeMidBudgetSynthesis kicks in around one-third budget", () => {
   assert.equal(shouldNudgeMidBudgetSynthesis(19, 20), false)
 })
 
-test("shouldNudgeMidBudgetSynthesis fires for chat_default budget too", () => {
-  // chat_default budget = 12 → threshold = floor(12/3) = 4
+test("shouldNudgeMidBudgetSynthesis fires for the default 12-step budget too", () => {
+  // budget = 12 → threshold = floor(12/3) = 4
   assert.equal(shouldNudgeMidBudgetSynthesis(3, 12), false)
   assert.equal(shouldNudgeMidBudgetSynthesis(4, 12), true)
   assert.equal(shouldNudgeMidBudgetSynthesis(10, 12), true)
diff --git a/tests/agent-runtime-task-mode-routing.test.mjs b/tests/agent-runtime-task-mode-routing.test.mjs
index 20413ee8..de7c3ecb 100644
--- a/tests/agent-runtime-task-mode-routing.test.mjs
+++ b/tests/agent-runtime-task-mode-routing.test.mjs
@@ -10,27 +10,9 @@ const providerOptionsUrl = pathToFileURL(
   path.join(cwd, "src/lib/server/llm/ai-sdk-gateway-provider-options.ts")
 ).href
 
-const {
-  getAiSdkGatewayProviderOptionsForMode,
-  getAiSdkGatewayProviderOptionsForTaskMode,
-} = await import(providerOptionsUrl)
-
-test("getAiSdkGatewayProviderOptionsForMode preserves legacy deep-research behavior", () => {
-  assert.deepEqual(getAiSdkGatewayProviderOptionsForMode(), {})
-  assert.deepEqual(getAiSdkGatewayProviderOptionsForMode({}), {})
-  assert.deepEqual(
-    getAiSdkGatewayProviderOptionsForMode({ deepResearch: false }),
-    {}
-  )
-  assert.deepEqual(
-    getAiSdkGatewayProviderOptionsForMode({ deepResearch: true }),
-    {
-      google: {
-        thinkingConfig: { thinkingLevel: "high", includeThoughts: true },
-      },
-    }
-  )
-})
+const { getAiSdkGatewayProviderOptionsForTaskMode } = await import(
+  providerOptionsUrl
+)
 
 test("Gemini gets high thinking for research, high_stakes, debugging", () => {
   for (const taskMode of ["research", "high_stakes", "debugging"]) {
diff --git a/tests/agent-session-state.test.mjs b/tests/agent-session-state.test.mjs
index bfdd889a..52193186 100644
--- a/tests/agent-session-state.test.mjs
+++ b/tests/agent-session-state.test.mjs
@@ -71,7 +71,6 @@ test("assistant session state builds assistant messages from stream accumulators
     createdAt: "2026-04-30T12:00:00.000Z",
     accumulator,
     model: "moonshotai/kimi-k2.6",
-    runMode: "research",
     isStreaming: true,
   })
 
@@ -79,7 +78,6 @@ test("assistant session state builds assistant messages from stream accumulators
   assert.equal(message.role, "assistant")
   assert.equal(message.content, "Done.")
   assert.equal(message.metadata?.isStreaming, true)
-  assert.equal(message.metadata?.runMode, "research")
   assert.deepEqual(message.metadata?.parts, [{ type: "text", text: "Done." }])
   assert.equal(message.metadata?.reasoning, "Checked the source.")
   assert.equal(message.metadata?.agentStatus, "completed")
@@ -96,7 +94,6 @@ test("assistant session state omits empty structured fields and upserts by id",
     createdAt: "2026-04-30T12:00:00.000Z",
     accumulator: createAccumulator({ content: "Partial" }),
     model: "moonshotai/kimi-k2.6",
-    runMode: "chat",
     isStreaming: true,
   })
   const finalMessage = createAssistantMessageFromAccumulator({
@@ -104,7 +101,6 @@ test("assistant session state omits empty structured fields and upserts by id",
     createdAt: "2026-04-30T12:00:00.000Z",
     accumulator: createAccumulator({ content: "Final" }),
     model: "moonshotai/kimi-k2.6",
-    runMode: "chat",
     isStreaming: false,
   })
 
@@ -122,7 +118,6 @@ test("assistant session state attaches follow-up questions without changing cont
     createdAt: "2026-04-30T12:00:00.000Z",
     accumulator: createAccumulator({ content: "Final answer." }),
     model: "moonshotai/kimi-k2.6",
-    runMode: "chat",
     isStreaming: false,
   })
   const updatedMessages = attachFollowUpQuestionsToMessage(
@@ -154,7 +149,6 @@ test("assistant session state tracks pending follow-up questions", () => {
     createdAt: "2026-04-30T12:00:00.000Z",
     accumulator: createAccumulator({ content: "Final answer." }),
     model: "moonshotai/kimi-k2.6",
-    runMode: "chat",
     isStreaming: false,
   })
 
diff --git a/tests/agent-system-prompt.test.mjs b/tests/agent-system-prompt.test.mjs
index cb387df2..09d81750 100644
--- a/tests/agent-system-prompt.test.mjs
+++ b/tests/agent-system-prompt.test.mjs
@@ -50,7 +50,6 @@ test("agent system prompt composes trusted blocks in priority order", () => {
     {
       now: new Date("2026-05-03T12:34:56.000Z"),
       userTimeZone: "America/Chicago",
-      deepResearchMode: true,
       provider: "alibaba",
       taskMode: "research",
     }
@@ -62,7 +61,6 @@ test("agent system prompt composes trusted blocks in priority order", () => {
     "--- BEGIN PROVIDER OVERLAY: ALIBABA ---"
   )
   const taskIndex = prompt.indexOf("--- BEGIN TASK MODE OVERLAY: RESEARCH ---")
-  const deepResearchIndex = prompt.indexOf("--- BEGIN DEEP RESEARCH MODE ---")
   const identityIndex = prompt.indexOf(
     "--- BEGIN IDENTITY AND TONE CONTEXT ---"
   )
@@ -72,7 +70,6 @@ test("agent system prompt composes trusted blocks in priority order", () => {
   assert(dateIndex >= 0, "RUNTIME DATE CONTEXT block not found")
   assert(providerIndex >= 0, "PROVIDER OVERLAY block not found")
   assert(taskIndex >= 0, "TASK MODE OVERLAY block not found")
-  assert(deepResearchIndex >= 0, "Deep Research block not found")
   assert(identityIndex >= 0, "Identity and tone block not found")
   assert(authIndex >= 0, "AUTH USER CONTEXT block not found")
 
@@ -89,12 +86,8 @@ test("agent system prompt composes trusted blocks in priority order", () => {
     "Task mode overlay should follow provider overlay"
   )
   assert(
-    deepResearchIndex > taskIndex,
-    "Deep Research mode should follow the task mode overlay"
-  )
-  assert(
-    identityIndex > deepResearchIndex,
-    "Identity and tone should follow Deep Research mode"
+    identityIndex > taskIndex,
+    "Identity and tone should follow the task mode overlay"
   )
   assert(
     authIndex > identityIndex,
@@ -105,34 +98,11 @@ test("agent system prompt composes trusted blocks in priority order", () => {
   assert.match(prompt, /User time zone: America\/Chicago/)
   assert.match(prompt, /Use Qwen reasoning mode efficiently/)
   assert.match(prompt, /This request needs deep research/)
-  assert.match(
-    prompt,
-    /produce a long, detailed, comprehensive response unless the user explicitly asks/
-  )
-  assert.match(prompt, /report-grade answer/)
   assert.match(prompt, /Email: user@example.com/)
   assert(prompt.includes(DEFAULT_SOUL_FALLBACK_INSTRUCTION))
   assert.equal(prompt.includes("SOUL.md"), false)
 })
 
-test("agent system prompt only adds the Deep Research block for Research mode", () => {
-  const prompt = buildAgentSystemInstruction(
-    {
-      id: "user-1",
-      name: "Chloei",
-      email: "user@example.com",
-    },
-    {
-      now: new Date("2026-05-03T12:34:56.000Z"),
-      provider: "alibaba",
-      taskMode: "research",
-    }
-  )
-
-  assert.match(prompt, /--- BEGIN TASK MODE OVERLAY: RESEARCH ---/)
-  assert.equal(prompt.includes("--- BEGIN DEEP RESEARCH MODE ---"), false)
-})
-
 test("agent system prompt places the identity block after task steering", () => {
   const prompt = buildAgentSystemInstruction(
     {
diff --git a/tests/assistant-message-contract.test.mjs b/tests/assistant-message-contract.test.mjs
index 948a9a7d..5382af67 100644
--- a/tests/assistant-message-contract.test.mjs
+++ b/tests/assistant-message-contract.test.mjs
@@ -44,7 +44,7 @@ test("assistant regenerate action reuses the prior user message", async () => {
   )
   assert.match(
     source,
-    /const handleRegenerate =[\s\S]*onEditMessage\?\.\(\{[\s\S]*messageId: userMessage\.id,[\s\S]*newContent: userMessage\.content,[\s\S]*newModel: regenerateModel,[\s\S]*newRunMode: regenerateRunMode,[\s\S]*\}\)[\s\S]*onRegenerate=\{handleRegenerate\}/,
+    /const handleRegenerate =[\s\S]*onEditMessage\?\.\(\{[\s\S]*messageId: userMessage\.id,[\s\S]*newContent: userMessage\.content,[\s\S]*newModel: regenerateModel,[\s\S]*\}\)[\s\S]*onRegenerate=\{handleRegenerate\}/,
     "Expected regenerate to rerun the previous user prompt through the existing edit-regenerate path."
   )
 })
diff --git a/tests/follow-up-questions.test.mjs b/tests/follow-up-questions.test.mjs
index 8f0c7f51..ac0ff143 100644
--- a/tests/follow-up-questions.test.mjs
+++ b/tests/follow-up-questions.test.mjs
@@ -40,7 +40,6 @@ function assistantMessage(overrides = {}) {
       isStreaming: false,
       agentStatus: "completed",
       selectedModel: MODEL,
-      runMode: "chat",
       ...metadata,
     },
   }
@@ -160,7 +159,7 @@ test("legacy vs generated follow-up detection", () => {
 test("getFollowUpQuestionRequestTargets selects an eligible assistant turn", () => {
   const messages = [
     { id: "u1", role: "user", content: "Question?", createdAt: "t" },
-    assistantMessage({ id: "a1", metadata: { runMode: "research" } }),
+    assistantMessage({ id: "a1" }),
   ]
 
   const targets = getFollowUpQuestionRequestTargets(messages, new Set())
@@ -168,7 +167,6 @@ test("getFollowUpQuestionRequestTargets selects an eligible assistant turn", ()
   assert.equal(targets.length, 1)
   assert.equal(targets[0].assistantMessageId, "a1")
   assert.equal(targets[0].model, MODEL)
-  assert.equal(targets[0].runMode, "research")
   assert.deepEqual(
     targets[0].messages.map((message) => message.id),
     ["u1", "a1"]
diff --git a/tests/gateway-search-tools.test.mjs b/tests/gateway-search-tools.test.mjs
index 4d86a6be..3c94a3ca 100644
--- a/tests/gateway-search-tools.test.mjs
+++ b/tests/gateway-search-tools.test.mjs
@@ -7,9 +7,6 @@ import { fileURLToPath, pathToFileURL } from "node:url"
 import "./register-ts-path-hooks.mjs"
 
 const cwd = fileURLToPath(new URL("..", import.meta.url))
-const gatewayProviderOptionsUrl = pathToFileURL(
-  path.join(cwd, "src/lib/server/llm/ai-sdk-gateway-provider-options.ts")
-).href
 const tavilyToolsPath = path.join(
   cwd,
   "src/lib/server/llm/ai-sdk-tavily-tools.ts"
@@ -18,10 +15,6 @@ const persistentSelectedModelUrl = pathToFileURL(
   path.join(cwd, "src/hooks/agent/persistent-selected-model-utils.ts")
 ).href
 
-const {
-  getAiSdkGatewayProviderOptions,
-  getAiSdkGatewayProviderOptionsForMode,
-} = await import(gatewayProviderOptionsUrl)
 const {
   parseStoredSelectedModel,
   resolvePersistedSelectedModel,
@@ -43,25 +36,6 @@ test("tavily search tool results derive source links", async () => {
   )
 })
 
-test("legacy deep-research provider options preserve Gemini high thinking", () => {
-  assert.deepEqual(getAiSdkGatewayProviderOptions(), {})
-  assert.deepEqual(
-    getAiSdkGatewayProviderOptionsForMode({ deepResearch: false }),
-    {}
-  )
-  assert.deepEqual(
-    getAiSdkGatewayProviderOptionsForMode({ deepResearch: true }),
-    {
-      google: {
-        thinkingConfig: {
-          thinkingLevel: "high",
-          includeThoughts: true,
-        },
-      },
-    }
-  )
-})
-
 test("inline citation instructions avoid separate sources sections", async () => {
   const source = await import("node:fs/promises").then((fs) =>
     fs.readFile(
diff --git a/tests/home-agent-utils.test.mjs b/tests/home-agent-utils.test.mjs
index 8fbbbf28..d7da8d8d 100644
--- a/tests/home-agent-utils.test.mjs
+++ b/tests/home-agent-utils.test.mjs
@@ -72,14 +72,12 @@ test("agent request messages trim oversized existing message content", () => {
   assert.match(requestMessages[0].content, /truncated/)
 })
 
-test("appended user messages preserve the requested run mode", () => {
+test("appended user messages record the selected model", () => {
   const messages = appendUserMessage(
     [],
     "Research Apple supply chain risk.",
-    "moonshotai/kimi-k2.6",
-    "research"
+    "moonshotai/kimi-k2.6"
   )
 
   assert.equal(messages[0]?.metadata?.selectedModel, "moonshotai/kimi-k2.6")
-  assert.equal(messages[0]?.metadata?.runMode, "research")
 })
diff --git a/tests/model-registry.test.mjs b/tests/model-registry.test.mjs
index 714a9a98..c0309d20 100644
--- a/tests/model-registry.test.mjs
+++ b/tests/model-registry.test.mjs
@@ -46,12 +46,6 @@ test("shared model registry includes the curated gateway models", async () => {
     "Expected SUPPORTED_MODELS to list Qwen 3.7 Max, Gemini 3.1 Pro Preview, Kimi K2.6, and MiMo V2.5 Pro."
   )
 
-  assert.match(
-    source,
-    /RESEARCH_MODEL = AvailableModels\.ALIBABA_QWEN3_7_MAX/,
-    "Expected Research mode to use Qwen 3.7 Max."
-  )
-
   assert.match(
     source.replace(/\s+/g, " "),
     /MODEL_SELECTOR_MODELS = \[ AvailableModels\.ALIBABA_QWEN3_7_MAX, AvailableModels\.MOONSHOTAI_KIMI_K2_6, AvailableModels\.XIAOMI_MIMO_V2_5_PRO, \] as const/,
diff --git a/tests/persistent-run-mode.test.mjs b/tests/persistent-run-mode.test.mjs
deleted file mode 100644
index a3168012..00000000
--- a/tests/persistent-run-mode.test.mjs
+++ /dev/null
@@ -1,88 +0,0 @@
-import assert from "node:assert/strict"
-import { readFile } from "node:fs/promises"
-import path from "node:path"
-import test from "node:test"
-import { fileURLToPath, pathToFileURL } from "node:url"
-
-import "./register-ts-path-hooks.mjs"
-
-const cwd = fileURLToPath(new URL("..", import.meta.url))
-const persistentRunModeUrl = pathToFileURL(
-  path.join(cwd, "src/hooks/agent/persistent-run-mode-utils.ts")
-).href
-const persistentRunModeHookPath = path.join(
-  cwd,
-  "src/hooks/agent/use-persistent-run-mode.tsx"
-)
-
-const { parseStoredRunMode, resolvePersistedRunMode, serializeStoredRunMode } =
-  await import(persistentRunModeUrl)
-
-test("persistent run mode parses only supported modes", () => {
-  assert.equal(parseStoredRunMode("research"), "research")
-  assert.equal(parseStoredRunMode("chat"), "chat")
-  assert.equal(parseStoredRunMode("browse"), null)
-  assert.equal(parseStoredRunMode(null), null)
-})
-
-test("persistent run mode parses current stored payload shape", () => {
-  assert.equal(
-    parseStoredRunMode(JSON.stringify(serializeStoredRunMode("research"))),
-    "research"
-  )
-  assert.equal(
-    parseStoredRunMode(
-      JSON.stringify({
-        runMode: "research",
-        source: "system",
-        version: 1,
-      })
-    ),
-    null
-  )
-})
-
-test("persistent run mode prefers stored mode over current and fallback", () => {
-  assert.equal(
-    resolvePersistedRunMode({
-      storedRunMode: "research",
-      currentRunMode: "chat",
-      fallbackRunMode: "chat",
-    }),
-    "research"
-  )
-  assert.equal(
-    resolvePersistedRunMode({
-      storedRunMode: null,
-      currentRunMode: "research",
-      fallbackRunMode: "chat",
-    }),
-    "research"
-  )
-  assert.equal(
-    resolvePersistedRunMode({
-      storedRunMode: null,
-    }),
-    "chat"
-  )
-})
-
-test("persistent run mode hook guards restricted storage access", async () => {
-  const source = await readFile(persistentRunModeHookPath, "utf8")
-
-  assert.match(
-    source,
-    /try \{\s+storedValue = window\.localStorage\.getItem\(RUN_MODE_STORAGE_KEY\)/,
-    "Expected persisted run mode reads to guard localStorage access."
-  )
-  assert.match(
-    source,
-    /try \{\s+window\.localStorage\.setItem\(\s+RUN_MODE_STORAGE_KEY,/,
-    "Expected persisted run mode writes to guard localStorage access."
-  )
-  assert.match(
-    source,
-    /try \{\s+window\.dispatchEvent\(new CustomEvent\(RUN_MODE_UPDATED_EVENT\)\)/,
-    "Expected persisted run mode updates to guard event dispatch."
-  )
-})
diff --git a/tests/prompt-form-contract.test.mjs b/tests/prompt-form-contract.test.mjs
index 53447884..758f6864 100644
--- a/tests/prompt-form-contract.test.mjs
+++ b/tests/prompt-form-contract.test.mjs
@@ -10,53 +10,17 @@ const promptFormPath = path.join(
   "src/components/agent/prompt-form/prompt-form.tsx"
 )
 
-test("prompt form research shortcut does not close the tools popover", async () => {
+test("prompt form drops the tools popover and research mode controls", async () => {
   const source = await readFile(promptFormPath, "utf8")
-  const shortcutSource = source.match(
-    /const handleResearchShortcut = \(event: KeyboardEvent\) => \{[\s\S]*?window\.addEventListener\("keydown", handleResearchShortcut\)/
-  )?.[0]
 
-  assert.ok(
-    shortcutSource,
-    "Expected PromptForm to define a research shortcut."
-  )
-  assert.match(
-    shortcutSource,
-    /setRunMode\(\(currentRunMode\) =>\s+currentRunMode === "research" \? "chat" : "research"\s+\)/,
-    "Expected shortcut to toggle run mode directly."
-  )
-  assert.doesNotMatch(
-    shortcutSource,
-    /handleSetRunMode|setIsToolsOpen|shouldPreventToolsCloseAutoFocusRef/,
-    "Expected shortcut not to reuse popover-closing menu behavior."
-  )
-})
-
-test("prompt form preserves Research mode after submit", async () => {
-  const source = await readFile(promptFormPath, "utf8")
-  const submitSource = source.match(
-    /const submitPrompt = useCallback\([\s\S]*?const handleSubmit = useCallback/
-  )?.[0]
-
-  assert.ok(submitSource, "Expected PromptForm to define submitPrompt.")
-  assert.match(
-    source,
-    /usePersistentRunMode\(\)/,
-    "Expected PromptForm to persist run mode across remounts."
-  )
   assert.doesNotMatch(
     source,
-    /useState<AgentRunMode>\("chat"\)/,
-    "Expected PromptForm not to reset run mode to chat on remount."
+    /runMode|usePersistentRunMode|Research|Telescope|Popover|setIsToolsOpen/,
+    "Expected PromptForm to drop the Tools popover and Research mode controls."
   )
   assert.match(
-    submitSource,
-    /const activeRunMode = runMode[\s\S]*onSubmit\?\.\([\s\S]*activeRunMode/,
-    "Expected submitPrompt to submit the active run mode."
-  )
-  assert.doesNotMatch(
-    submitSource,
-    /setRunMode\("chat"\)/,
-    "Expected submitPrompt not to reset Research mode back to chat."
+    source,
+    /<ModelSelector/,
+    "Expected PromptForm to render the model selector in the footer."
   )
 })
diff --git a/tests/thread-payload-contract.test.mjs b/tests/thread-payload-contract.test.mjs
index 68aa29c3..8a8770d4 100644
--- a/tests/thread-payload-contract.test.mjs
+++ b/tests/thread-payload-contract.test.mjs
@@ -123,7 +123,7 @@ test("thread store delegates parsing and persistence shaping to the payload help
   )
 })
 
-test("thread payload preserves valid run modes and drops invalid run modes", () => {
+test("thread payload drops legacy run-mode metadata from stored threads", () => {
   const parsed = parseThreadPayload({
     id: "thread-1",
     model: "google/gemini-3.1-pro-preview",
@@ -139,21 +139,14 @@ test("thread payload preserves valid run modes and drops invalid run modes", ()
           runMode: "research",
         },
       },
-      {
-        id: "message-2",
-        role: "assistant",
-        content: "Done.",
-        llmModel: "google/gemini-3.1-pro-preview",
-        createdAt: "2026-04-26T00:00:01.000Z",
-        metadata: {
-          runMode: "invalid",
-        },
-      },
     ],
     createdAt: "2026-04-26T00:00:00.000Z",
     updatedAt: "2026-04-26T00:00:01.000Z",
   })
 
-  assert.equal(parsed.messages[0]?.metadata?.runMode, "research")
-  assert.equal(parsed.messages[1]?.metadata?.runMode, undefined)
+  assert.equal(
+    parsed.messages[0]?.metadata?.selectedModel,
+    "moonshotai/kimi-k2.6"
+  )
+  assert.equal(parsed.messages[0]?.metadata?.runMode, undefined)
 })