feat(pipeline): upgrade built-ins to DeepSeek V4.1 Flash

docs/configuration.md

@@ -24,7 +24,7 @@ defaults:
   baseRef: main                    # optional; auto-detected when unset (origin default branch, else main/master/develop/trunk, else current branch)
   pipeline: quick                  # pipeline used when -p/--pipeline is not given
   autoAcceptJudgeModel: anthropic/claude-haiku-4-5   # model for smart auto-accept (--smart); defaults to the run's model
-  branchNameModel: openrouter/deepseek/deepseek-v4-flash-0731  # proposes worktree branch names (may look up referenced issues); you confirm the name
+  branchNameModel: openrouter/deepseek/deepseek-v4.1-flash  # proposes worktree branch names (may look up referenced issues); you confirm the name
   commitMessageModel: openai/gpt-5.6-luna     # writes the conventional commit message for automatic run compaction and close's squash-merge commit
   worktree: true                   # force a new branch + worktree for every run; false always runs in the current tree. Unset decides per branch (isolate on a trunk, run in place on a branch)
   worktreeLocation: ~/dev/worktrees/{repo}/{branch}  # where isolated worktrees are created ({repo}/{branch} placeholders, ~ = home; the branch slug is appended when {branch} is missing). A marker line in the repo's AGENTS.md/README.md outranks this; default ~/.convoy/worktrees

docs/pipelines.md

@@ -17,18 +17,18 @@ PRD → implementer → patterns → security → design → tests → measure
                                                            └ fix┘
 ```
 
-All model IDs below are public provider IDs. The former local NAN models use their OpenRouter equivalents: `openrouter/deepseek/deepseek-v4-flash-0731#high` and `openrouter/z-ai/glm-5.3-flash#high`.
+All model IDs below are public provider IDs. The former local NAN models use their OpenRouter equivalents: `openrouter/deepseek/deepseek-v4.1-flash#high` and `openrouter/z-ai/glm-5.3-flash#high`.
 
 | Step | Model | Advisor |
 |---|---|---|
-| `implementer` | DeepSeek V4 Flash 0731 · high | GPT 6 Astra · xhigh |
-| `patterns` | DeepSeek V4 Flash 0731 · high | GLM 5.3 · high |
+| `implementer` | DeepSeek V4.1 Flash · high | GPT 6 Astra · xhigh |
+| `patterns` | DeepSeek V4.1 Flash · high | GLM 5.3 · high |
 | `security` | GLM 5.3 Flash · high | GLM 5.3 · high |
 | `design` | GLM 5.3 Flash · high | GPT 6 Astra · xhigh |
-| `tests` | DeepSeek V4 Flash 0731 · high | GLM 5.3 · high |
+| `tests` | DeepSeek V4.1 Flash · high | GLM 5.3 · high |
 | Goal: independent scorers | Grok 4.6 · high + GLM 5.3 · high | — |
 | Goal: verified consensus | GPT 5.6 Sol · xhigh | — |
-| Goal: targeted fixes | DeepSeek V4 Flash 0731 · high | GPT 6 Astra · xhigh |
+| Goal: targeted fixes | DeepSeek V4.1 Flash · high | GPT 6 Astra · xhigh |
 
 The goal measures first, then fixes only reported gaps. It stops at **90/100**, after **five fix rounds**, or on the configured plateau limit (three rounds by default). A run that hits a limit can finish below the target; inspect its final score. Scorers read the original requirements and current diff without previous reports; the consensus reads only the current scorer reports and verifies their claims.
 
@@ -41,7 +41,7 @@ Select one with `-p/--pipeline`; no project config is needed. `full-cycle` is th
 | Pipeline | Changes code? | What it does |
 |---|---|---|
 | `full-cycle` | yes | Implement, audit, polish and test with the models above, then independently score and fix gaps up to five times to reach 90/100. |
-| `implement` | yes | DeepSeek V4 Flash implements with Astra xhigh advice. Patterns and tests use DeepSeek; security and design use GLM Flash. Those later steps run unadvised. Closes with a read-only DeepSeek recap at `reports/run-report.md`; no scoring loop. |
+| `implement` | yes | DeepSeek V4.1 Flash implements with Astra xhigh advice. Patterns and tests use DeepSeek; security and design use GLM Flash. Those later steps run unadvised. Closes with a read-only DeepSeek recap at `reports/run-report.md`; no scoring loop. |
 | `ship` | yes | Sync with the base; scope and review the diff across DeepSeek and GLM Flash; report, adversarially triage and fix accepted findings with Astra advice; recap; then score with Grok and GLM, reconcile on GLM and fix gaps up to five times to reach 90/100. |
 | `review` | no | The report-only review: DeepSeek scopes and reports; clean-code, security and bug audits each run on DeepSeek and GLM Flash. Grok and GLM score independently, then GLM verifies the consensus. Produces findings and a quality score. |
 | `fixer` | yes | Prove supplied findings with regression tests, apply targeted fixes, then independently rerun the checks. Terra xhigh handles all three phases; Astra xhigh advises reproduction and fixes. |

docs/worktrees.md

@@ -110,7 +110,7 @@ An isolated run gets a new branch checked out in a dedicated worktree, leaving y
 
 The branch is always agreed with you first, in a **Branch** step between Options and Review:
 
-- An `Intended Branch Name` (or `git checkout -b …`) in the prompt is used as-is — the model is not asked to reinvent it. A short prompt that is just a path to a plan file is read first, so pasting `docs/plans/foo.md` still picks up the name inside. Otherwise `defaults.branchNameModel` (DeepSeek V4 Flash 0731 via OpenRouter by default) reads the prompt and proposes a conventional name — `feat/runtime-guard-limits`, `fix/login-redirect` — always in English, even when the prompt is not, keeping the document's own words rather than paraphrasing them. Prompts that only reference an issue (`#123`, `DEV-1339`, a URL) are looked up first, so the branch is named after what the issue is about.
+- An `Intended Branch Name` (or `git checkout -b …`) in the prompt is used as-is — the model is not asked to reinvent it. A short prompt that is just a path to a plan file is read first, so pasting `docs/plans/foo.md` still picks up the name inside. Otherwise `defaults.branchNameModel` (DeepSeek V4.1 Flash via OpenRouter by default) reads the prompt and proposes a conventional name — `feat/runtime-guard-limits`, `fix/login-redirect` — always in English, even when the prompt is not, keeping the document's own words rather than paraphrasing them. Prompts that only reference an issue (`#123`, `DEV-1339`, a URL) are looked up first, so the branch is named after what the issue is about.
 - The proposed name is shown in an editable field together with the worktree path it would take. Enter accepts it and moves on to Review; nothing is created until you confirm the run there.
 - `tab` moves to the **hint** box: describe how you want it named ("name it after the budget limits") and press Enter or `ctrl+R` to re-name it. This is also what you get when the prompt is too thin to name anything, or when the naming model is unavailable — the step still opens, with a name derived from the prompt, ready to be edited.
 - Names already taken by a branch or an existing worktree are suffixed (`-2`, `-3`) instead of failing `git worktree add` after the run has been confirmed.

src/config-tui.ts

@@ -2133,7 +2133,7 @@ export function describeDefault(key: keyof ConvoyDefaults): string {
     case "autoAcceptJudgeModel":
       return "Model the smart auto-accept judge uses (falls back to the run's model)."
     case "branchNameModel":
-      return "Model that names worktree branches (default: openrouter/deepseek/deepseek-v4-flash-0731)."
+      return "Model that names worktree branches (default: openrouter/deepseek/deepseek-v4.1-flash)."
     case "commitMessageModel":
       return "Model that writes the conventional commit message for automatic run compaction and close (default: anthropic/claude-haiku-4-5)."
     case "worktree":

src/config.ts

@@ -287,7 +287,7 @@ defaults:
   # maxConcurrentAgents: 30 # optional: cap agents running at once within a parallel group
   # baseRef: main # optional: when unset, convoy auto-detects (origin default branch, else main/master/develop/trunk, else current branch)
   # pipeline: full-cycle
-  # branchNameModel: openrouter/deepseek/deepseek-v4-flash-0731 # optional: model that names worktree branches
+  # branchNameModel: openrouter/deepseek/deepseek-v4.1-flash # optional: model that names worktree branches
   # commitMessageModel: openai/gpt-5.6-luna # optional: model that writes the conventional commit message for automatic run compaction and close's squash-merge commit
   # worktree: true # optional: force a fresh branch + worktree for every run; false always runs in the current tree. Unset decides per branch: isolate on a trunk (main/master/develop/trunk or the detected base), run in place on any other branch
   # worktreeLocation: ~/dev/worktrees/{repo}/{branch} # optional: where isolated worktrees are created ({repo}/{branch} placeholders, ~ = home; the branch slug is appended when {branch} is missing). A marker in the repo's AGENTS.md/README.md outranks this; unusable locations fall back to ~/.convoy/worktrees
@@ -347,7 +347,7 @@ defaults:
 # The \`implement\` pipeline is inlined below as an editable starting point; redefining a name here overrides the built-in.
 pipelines:
   implement:
-    description: Advised implementation on DeepSeek V4 Flash consulting Astra 6, then pattern/security audits, design polish, tests, and a one-page run recap
+    description: Advised implementation on DeepSeek V4.1 Flash consulting Astra 6, then pattern/security audits, design polish, tests, and a one-page run recap
     # defaultPrompt and suggestedPrompts are optional. A defaultPrompt is used
     # when the pipeline runs without an explicit prompt — the launcher prefills
     # its field and \`convoy -p <pipeline>\` falls back to it — and the

src/pipeline.ts

@@ -15,9 +15,9 @@ const opusViaOpenRouter = "openrouter/anthropic/claude-opus-5"
 /** Grok 4.6 high: the review and ship scoring, plus one hunter audit track. */
 const grokModel = "openrouter/x-ai/grok-4.6#high"
 const kimiModel = "openrouter/moonshotai/kimi-k3"
-/** DeepSeek V4 Flash 0731 on OpenRouter: the writer for `implement`, review's report, and the goal fixers. */
-const deepseekModel = "openrouter/deepseek/deepseek-v4-flash-0731"
-/** DeepSeek V4 Flash on OpenRouter with reasoning raised: used for writing, flash audits and run recaps. */
+/** DeepSeek V4.1 Flash on OpenRouter: the writer for `implement`, review's report, and the goal fixers. */
+const deepseekModel = "openrouter/deepseek/deepseek-v4.1-flash"
+/** DeepSeek V4.1 Flash on OpenRouter with reasoning raised: used for writing, flash audits and run recaps. */
 const deepseekHighModel = `${deepseekModel}#high`
 /** GPT 5.6 Sol: the consensus reporter for full-cycle and the hunter pipelines at xhigh. */
 const solModel = "openai/gpt-5.6-sol"
@@ -462,7 +462,7 @@ export const builtInPipelines: Record<string, PipelineSpec> = {
   // Implementation follows the same model choices as full-cycle, but closes
   // with a recap instead of running the quality loop. Only the writer is advised.
   implement: {
-    description: "Advised implementation on DeepSeek V4 Flash consulting Astra 6, then pattern/security audits, design polish, tests, and a one-page run recap",
+    description: "Advised implementation on DeepSeek V4.1 Flash consulting Astra 6, then pattern/security audits, design polish, tests, and a one-page run recap",
     steps: [
       { agent: "implementer", model: defaultImplementerModel, advisor: defaultImplementAdvisorModel, reports: "none" },
       { agent: "patterns", model: defaultImplementAuditModel, advisor: false },
@@ -475,7 +475,7 @@ export const builtInPipelines: Record<string, PipelineSpec> = {
   // The default combines implementation with a terminal, independently scored goal loop.
   "full-cycle": {
     description:
-      "Implement with DeepSeek V4 Flash and GLM 5.3 Flash, advised by Astra 6 and GLM 5.3, then measure with Grok and GLM and a verified Sol consensus; fix gaps up to five times to reach 90/100.",
+      "Implement with DeepSeek V4.1 Flash and GLM 5.3 Flash, advised by Astra 6 and GLM 5.3, then measure with Grok and GLM and a verified Sol consensus; fix gaps up to five times to reach 90/100.",
     steps: [
       { agent: "implementer", model: deepseekHighModel, advisor: astraXhighModel, reports: "none" },
       { agent: "patterns", model: deepseekHighModel, advisor: glm53HighModel },
@@ -489,7 +489,7 @@ export const builtInPipelines: Record<string, PipelineSpec> = {
           improve: {
             briefStep: "fix",
             steps: [
-              // The directed fixer, on DeepSeek V4 Flash advised by Astra 6 (extra
+              // The directed fixer, on DeepSeek V4.1 Flash advised by Astra 6 (extra
               // high). It alone receives the score brief (by step name). diff:
               // true is load-bearing: as the fragment's first step it would
               // otherwise default to no diff.
@@ -520,7 +520,7 @@ export const builtInPipelines: Record<string, PipelineSpec> = {
   // Report-only audits on two flash models, followed by independent scoring.
   review: {
     description:
-      "Report-only PR review on ultra-cheap models: scope, parallel audits (DeepSeek V4 Flash + GLM 5.3 Flash) and report on flash models; the scoring uses GLM 5.3 high + Grok 4.6 high and the score consensus stays on GLM 5.3 high. Makes no changes.",
+      "Report-only PR review on ultra-cheap models: scope, parallel audits (DeepSeek V4.1 Flash + GLM 5.3 Flash) and report on flash models; the scoring uses GLM 5.3 high + Grok 4.6 high and the score consensus stays on GLM 5.3 high. Makes no changes.",
     defaultPrompt: "Review the current branch against its base and report prioritized findings with a verified quality score.",
     suggestedPrompts: ["Review the open PR for this branch", "Review only the last commit's diff"],
     steps: [

src/worktree.ts

@@ -49,7 +49,7 @@ export type BranchNameProposal = {
 }
 
 /** Cheap, fast model used to synthesize a branch name from the prompt. */
-export const defaultBranchNameModel = "openrouter/deepseek/deepseek-v4-flash-0731"
+export const defaultBranchNameModel = "openrouter/deepseek/deepseek-v4.1-flash"
 
 /** Registered so the namer replaces opencode's default coding agent instead of merely appending to it. */
 const namerAgentName = "convoy-branch-namer"

test/agents.test.ts

@@ -277,12 +277,12 @@ describe("throughput routing in the opencode config", () => {
     ], undefined, {
       throughputModels: [
         { providerID: "openrouter", modelID: "z-ai/glm-5.3" },
-        { providerID: "openrouter", modelID: "deepseek/deepseek-v4-flash-0731" },
+        { providerID: "openrouter", modelID: "deepseek/deepseek-v4.1-flash" },
       ],
     })
 
     expect(config.provider?.openrouter?.models?.["z-ai/glm-5.3"]).toEqual({ options: { provider: { sort: "throughput" } } })
-    expect(config.provider?.openrouter?.models?.["deepseek/deepseek-v4-flash-0731"]).toEqual({ options: { provider: { sort: "throughput" } } })
+    expect(config.provider?.openrouter?.models?.["deepseek/deepseek-v4.1-flash"]).toEqual({ options: { provider: { sort: "throughput" } } })
     // Options-only entries: no name or limit is invented over the catalog's real model.
     expect(config.provider?.openrouter?.models?.["z-ai/glm-5.3"]?.name).toBeUndefined()
     expect(config.provider?.openrouter?.models?.["z-ai/glm-5.3"]?.limit).toBeUndefined()

test/attach.test.ts

@@ -167,11 +167,11 @@ describe("reconstructedPhases", () => {
       "pre-hook-1",
       "sync",
       "scope",
-      "clean-code__openrouter-deepseek-deepseek-v4-flash-0731-high",
+      "clean-code__openrouter-deepseek-deepseek-v4-1-flash-high",
       "clean-code__openrouter-z-ai-glm-5-3-flash-high",
-      "security__openrouter-deepseek-deepseek-v4-flash-0731-high",
+      "security__openrouter-deepseek-deepseek-v4-1-flash-high",
       "security__openrouter-z-ai-glm-5-3-flash-high",
-      "bugs__openrouter-deepseek-deepseek-v4-flash-0731-high",
+      "bugs__openrouter-deepseek-deepseek-v4-1-flash-high",
       "bugs__openrouter-z-ai-glm-5-3-flash-high",
       "report",
       "triage",

test/config.test.ts

@@ -1156,7 +1156,7 @@ describe("default config init", () => {
     expect(body).toContain("# maxConcurrentAgents: 30")
     expect(body).toContain("# baseRef: main")
     expect(body).toContain("# pipeline: full-cycle")
-    expect(body).toContain("# branchNameModel: openrouter/deepseek/deepseek-v4-flash-0731")
+    expect(body).toContain("# branchNameModel: openrouter/deepseek/deepseek-v4.1-flash")
     expect(body).toContain("# hooks:")
     expect(body).toContain("#           command: gh pr create --fill")
     expect(body).toContain("# agents:")

test/coordinate.test.ts

@@ -294,15 +294,15 @@ describe("runCoordinateBoot", () => {
     const unresolved = advisedImplementerStep()
     const resolved = {
       ...unresolved,
-      model: "openrouter/deepseek/deepseek-v4-flash-0731",
+      model: "openrouter/deepseek/deepseek-v4.1-flash",
       resolvedModel: {
         configured: "nan/deepseek-v4-flash#high",
         logical: "nan/deepseek-v4-flash#high",
         gateway: "nitro" as const,
         providerID: "openrouter",
-        modelID: "deepseek/deepseek-v4-flash-0731",
+        modelID: "deepseek/deepseek-v4.1-flash",
         variant: "high",
-        target: "openrouter/deepseek/deepseek-v4-flash-0731#high",
+        target: "openrouter/deepseek/deepseek-v4.1-flash#high",
       },
       resolvedAdvisor: {
         configured: "openrouter/x-ai/grok-4.6#high",

test/launch-tui.test.ts

@@ -409,7 +409,7 @@ describe("launch TUI narrow-width row budgets", () => {
         { stepName: "score-report", groupId: "g2", kind: "agent" as const, modelLabel: "x-ai-grok-5", advisorLabel: "" },
       ],
       improve: [
-        { stepName: "fix", groupId: "g3", kind: "agent" as const, modelLabel: "deepseek-v4-flash", advisorLabel: "x-ai-grok-5 advisor ×3" },
+        { stepName: "fix", groupId: "g3", kind: "agent" as const, modelLabel: "deepseek-v4.1-flash", advisorLabel: "x-ai-grok-5 advisor ×3" },
       ],
     } satisfies GoalPreview
     for (const width of widths) {
@@ -713,7 +713,7 @@ describe("launch TUI pipeline preview", () => {
       { stepName: "score-report", groupId: "g2", kind: "agent", modelLabel: "x-ai-grok-5", advisorLabel: "" },
     ],
     improve: [
-      { stepName: "fix", groupId: "g3", kind: "agent", modelLabel: "deepseek-v4-flash", advisorLabel: "x-ai-grok-5 advisor ×3" },
+      { stepName: "fix", groupId: "g3", kind: "agent", modelLabel: "deepseek-v4.1-flash", advisorLabel: "x-ai-grok-5 advisor ×3" },
     ],
   } satisfies GoalPreview
 
@@ -736,7 +736,7 @@ describe("launch TUI pipeline preview", () => {
     // relationship, with their resolved models.
     expect(lines).toContain("    ○ score  · 2 models")
     expect(lines).toContain("    ○ score-report  · x-ai-grok-5")
-    expect(lines).toContain("    ○ fix  · deepseek-v4-flash → x-ai-grok-5 advisor ×3")
+    expect(lines).toContain("    ○ fix  · deepseek-v4.1-flash → x-ai-grok-5 advisor ×3")
   })
 
   test("goalLines collapses policy chips and fragment roles when the panel is narrow", () => {
@@ -1569,7 +1569,7 @@ describe("launch TUI goal classification", () => {
             { stepName: "score", groupId: "g1", kind: "agent", modelLabel: "glm-5.3-high", advisorLabel: "" },
             { stepName: "score-report", groupId: "g2", kind: "agent", modelLabel: "x-ai-grok-5", advisorLabel: "" },
           ],
-          improve: [{ stepName: "fix", groupId: "g3", kind: "agent", modelLabel: "deepseek-v4-flash", advisorLabel: "" }],
+          improve: [{ stepName: "fix", groupId: "g3", kind: "agent", modelLabel: "deepseek-v4.1-flash", advisorLabel: "" }],
         },
       }],
       "configured",
@@ -1587,7 +1587,7 @@ describe("launch TUI goal classification", () => {
       expect(detail).not.toContain("4 measurements")
       expect(detail).toContain("measure  · score ← score-report")
       expect(detail).toContain("improve  · brief → fix  · then re-measure")
-      expect(detail).toContain("○ fix  · deepseek-v4-flash")
+      expect(detail).toContain("○ fix  · deepseek-v4.1-flash")
     } finally {
       await closeLauncher(launcher)
     }