diff --git a/components/LLMProviderPanel.jsx b/components/LLMProviderPanel.jsx
index 2d80c01..8b54fc3 100644
--- a/components/LLMProviderPanel.jsx
+++ b/components/LLMProviderPanel.jsx
@@ -101,7 +101,10 @@ export default function LLMProviderPanel({ onSave, onClose }) {
   };
 
   const isCustom = config.provider === "custom";
-  const showUrlField = isCustom || config.provider === "ollama";
+  const showUrlField =
+    isCustom
+    || config.provider === "ollama"
+    || config.provider === "ollama_local";
 
   return (
     <Panel>
diff --git a/engine/adapters/ollamaLocal.js b/engine/adapters/ollamaLocal.js
new file mode 100644
index 0000000..1e60a77
--- /dev/null
+++ b/engine/adapters/ollamaLocal.js
@@ -0,0 +1,81 @@
+/**
+ * Ollama (local) streaming adapter — OpenAI-compat path.
+ *
+ * Sends chat to Ollama's /v1/chat/completions through the Django proxy.
+ * The OpenAI-compat endpoint has a stable OpenAI-spec wire contract:
+ * tool_calls[].function.arguments is always a JSON string. The native
+ * /api/chat endpoint flipped this field's expected shape between Ollama
+ * versions (object on <=0.16.2, string on >0.16.2 for inbound history),
+ * which broke multi-round tool-call sessions for everyone running a
+ * newer local install. See chatbox-core PR #22 → reverted by #23, whose
+ * follow-up note named this adapter as the right architectural fix.
+ *
+ * Routes through /apps/tethysdash/ollama-proxy/v1/chat/completions/ so
+ * the browser doesn't hit CORS against a non-publicly-exposed Ollama
+ * daemon and so CSRF/auth wrap the call uniformly with other LLM
+ * providers.
+ *
+ * Model discovery + capability gating still go through the Ollama
+ * /api/tags + /api/show paths — see helpers/index.js listModels for
+ * the ollama_local branch.
+ */
+import OpenAI from "openai";
+import { mergeToolCalls } from "../../helpers/index.js";
+
+const PROXY_BASE = "/apps/tethysdash/ollama-proxy/v1";
+
+export async function streamChat({
+  baseUrl, apiKey, model,
+  messages, tools, csrfToken, signal,
+  onThinkingChunk, onContentChunk,
+}) {
+  const absoluteProxyBase = `${globalThis.location.origin}${PROXY_BASE}`;
+
+  const client = new OpenAI({
+    baseURL: absoluteProxyBase,
+    apiKey: apiKey || "ollama",
+    dangerouslyAllowBrowser: true,
+    defaultHeaders: {
+      ...(csrfToken ? { "x-csrftoken": csrfToken } : {}),
+      ...(baseUrl ? { "x-ollama-host": baseUrl } : {}),
+      ...(apiKey ? { "x-ollama-key": apiKey } : {}),
+    },
+  });
+
+  const mergedMessage = { role: "assistant", content: "", thinking: "", tool_calls: null };
+
+  const stream = await client.chat.completions.create(
+    {
+      model,
+      messages,
+      tools: tools?.length ? tools : undefined,
+      stream: true,
+      max_completion_tokens: 16384,
+    },
+    { signal },
+  );
+
+  for await (const chunk of stream) {
+    if (signal?.aborted) break;
+
+    const delta = chunk.choices?.[0]?.delta;
+    if (!delta) continue;
+
+    if (typeof delta.content === "string" && delta.content) {
+      mergedMessage.content += delta.content;
+      onContentChunk?.(delta.content);
+    }
+
+    if (typeof delta.reasoning === "string" && delta.reasoning) {
+      mergedMessage.thinking += delta.reasoning;
+      onThinkingChunk?.(delta.reasoning);
+    }
+
+    if (Array.isArray(delta.tool_calls)) {
+      mergedMessage.tool_calls = mergeToolCalls(mergedMessage.tool_calls ?? [], delta.tool_calls);
+    }
+  }
+
+  if (mergedMessage.tool_calls === null) delete mergedMessage.tool_calls;
+  return { message: mergedMessage };
+}
diff --git a/engine/adapters/ollamaLocal.test.js b/engine/adapters/ollamaLocal.test.js
new file mode 100644
index 0000000..6012210
--- /dev/null
+++ b/engine/adapters/ollamaLocal.test.js
@@ -0,0 +1,217 @@
+/**
+ * engine/adapters/ollamaLocal.test.js — coverage for the OpenAI-compat
+ * adapter that targets Ollama's /v1/chat/completions via the Django proxy.
+ *
+ * Pinned behaviors (in order of how they catch regressions):
+ *   - Routes through /apps/tethysdash/ollama-proxy/v1 (NOT direct browser
+ *     call to localhost:11434). Catches a regression where the adapter
+ *     starts hitting the daemon directly and runs into CORS.
+ *   - Forwards Django proxy headers (x-csrftoken, x-ollama-host,
+ *     x-ollama-key) via SDK defaultHeaders so the proxy can route to the
+ *     user-configured Ollama host.
+ *   - Multi-round tool-call wire contract: when the engine echoes a prior
+ *     assistant tool_call back as part of `messages`, the request shape
+ *     must follow OpenAI spec (arguments stringified). This is the
+ *     regression-catcher PR #23 named in its follow-up — "the test that
+ *     would have caught this regression".
+ *   - Return shape matches `{ message: { content, tool_calls } }` — the
+ *     contract engine's getMessage() reads from.
+ */
+
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+
+// Module-level capture so the mock factory can record what the adapter
+// passed in without exporting hooks from the SUT.
+let capturedConfig = null;
+let capturedCreateParams = null;
+let mockStream = null;
+
+vi.mock("openai", () => {
+  return {
+    default: vi.fn(function (config) {
+      capturedConfig = config;
+      this.chat = {
+        completions: {
+          create: vi.fn(async (params) => {
+            capturedCreateParams = params;
+            return mockStream;
+          }),
+        },
+      };
+    }),
+  };
+});
+
+async function* makeStream(chunks) {
+  for (const chunk of chunks) yield chunk;
+}
+
+beforeEach(() => {
+  capturedConfig = null;
+  capturedCreateParams = null;
+  mockStream = makeStream([]);
+  // The adapter reads globalThis.location.origin to build the proxy URL.
+  // jsdom sets it to "http://localhost:3000" by default; pin explicitly so
+  // assertions don't drift with vitest config changes.
+  if (!globalThis.location) {
+    globalThis.location = { origin: "http://localhost:3000" };
+  }
+});
+
+afterEach(() => {
+  vi.restoreAllMocks();
+});
+
+async function call(opts = {}) {
+  const { streamChat } = await import("./ollamaLocal.js");
+  return streamChat({
+    baseUrl: "http://localhost:11434",
+    apiKey: "ollama-test-key",
+    model: "qwen3:30b-a3b-instruct-2507-q4_K_M",
+    messages: [{ role: "user", content: "hi" }],
+    tools: [],
+    csrfToken: "csrf-test",
+    ...opts,
+  });
+}
+
+describe("ollamaLocal adapter — Django proxy routing", () => {
+  it("uses /apps/tethysdash/ollama-proxy/v1 as the SDK baseURL (absolute)", async () => {
+    await call();
+    expect(capturedConfig.baseURL).toBe(
+      "http://localhost:3000/apps/tethysdash/ollama-proxy/v1",
+    );
+  });
+
+  it("does NOT call the Ollama daemon directly (no localhost:11434 in baseURL)", async () => {
+    await call({ baseUrl: "http://localhost:11434" });
+    expect(capturedConfig.baseURL).not.toContain("11434");
+  });
+
+  it("forwards x-csrftoken, x-ollama-host, x-ollama-key via defaultHeaders", async () => {
+    await call({
+      baseUrl: "http://localhost:11434",
+      apiKey: "secret-key",
+      csrfToken: "csrf-abc",
+    });
+    expect(capturedConfig.defaultHeaders).toEqual({
+      "x-csrftoken": "csrf-abc",
+      "x-ollama-host": "http://localhost:11434",
+      "x-ollama-key": "secret-key",
+    });
+  });
+
+  it("omits header entries when their source values are empty", async () => {
+    await call({ baseUrl: "", apiKey: "", csrfToken: "" });
+    expect(capturedConfig.defaultHeaders).toEqual({});
+  });
+});
+
+describe("ollamaLocal adapter — multi-round wire contract", () => {
+  it(
+    "second-round request includes prior assistant tool_calls with arguments as STRING " +
+      "(the regression PR #23 named — Ollama /v1 follows OpenAI spec)",
+    async () => {
+      // Construct the kind of `messages` array the engine builds AFTER a
+      // first-round tool call: user → assistant(tool_calls) → tool result.
+      // Arguments come from mergeToolCalls; for /v1/chat/completions the
+      // wire shape must be a JSON string.
+      const messages = [
+        { role: "user", content: "list dates" },
+        {
+          role: "assistant",
+          content: "",
+          tool_calls: [
+            {
+              id: "call_1",
+              type: "function",
+              function: {
+                name: "list_available_dates",
+                arguments: '{"model":"cfe_nom"}',
+              },
+            },
+          ],
+        },
+        {
+          role: "tool",
+          tool_call_id: "call_1",
+          content: '{"dates": []}',
+        },
+      ];
+
+      await call({ messages });
+
+      const echoed = capturedCreateParams.messages.find((m) => m.role === "assistant");
+      expect(echoed).toBeDefined();
+      expect(echoed.tool_calls).toBeDefined();
+      expect(typeof echoed.tool_calls[0].function.arguments).toBe("string");
+    },
+  );
+});
+
+describe("ollamaLocal adapter — streaming + return shape", () => {
+  it("accumulates content deltas and returns { message: { content, ... } }", async () => {
+    mockStream = makeStream([
+      { choices: [{ delta: { content: "Hello" } }] },
+      { choices: [{ delta: { content: " world" } }] },
+    ]);
+    const chunks = [];
+    const result = await call({ onContentChunk: (c) => chunks.push(c) });
+    expect(chunks).toEqual(["Hello", " world"]);
+    expect(result).toEqual({
+      message: expect.objectContaining({
+        role: "assistant",
+        content: "Hello world",
+      }),
+    });
+  });
+
+  it("accumulates tool_calls and drops the field when nothing streamed", async () => {
+    mockStream = makeStream([
+      { choices: [{ delta: { content: "ok" } }] },
+    ]);
+    const result = await call();
+    expect("tool_calls" in result.message).toBe(false);
+  });
+
+  it("preserves tool_calls when the stream emits them", async () => {
+    mockStream = makeStream([
+      {
+        choices: [
+          {
+            delta: {
+              tool_calls: [
+                {
+                  index: 0,
+                  id: "call_1",
+                  type: "function",
+                  function: { name: "list_available_dates", arguments: '{"model":"cfe_nom"}' },
+                },
+              ],
+            },
+          },
+        ],
+      },
+    ]);
+    const result = await call();
+    expect(Array.isArray(result.message.tool_calls)).toBe(true);
+    expect(result.message.tool_calls[0].function.name).toBe(
+      "list_available_dates",
+    );
+  });
+
+  it("passes model, messages, tools, and max_completion_tokens to chat.completions.create", async () => {
+    await call({
+      model: "qwen3:30b-a3b-instruct-2507-q4_K_M",
+      messages: [{ role: "user", content: "hi" }],
+      tools: [{ type: "function", function: { name: "f", parameters: {} } }],
+    });
+    expect(capturedCreateParams.model).toBe(
+      "qwen3:30b-a3b-instruct-2507-q4_K_M",
+    );
+    expect(capturedCreateParams.messages).toEqual([{ role: "user", content: "hi" }]);
+    expect(capturedCreateParams.tools).toBeDefined();
+    expect(capturedCreateParams.stream).toBe(true);
+    expect(capturedCreateParams.max_completion_tokens).toBe(16384);
+  });
+});
diff --git a/engine/index.js b/engine/index.js
index 24bdf3c..c0bc504 100644
--- a/engine/index.js
+++ b/engine/index.js
@@ -43,12 +43,14 @@ import { streamChat as openaiStreamChat } from "./adapters/openai.js";
 import { streamChat as anthropicStreamChat } from "./adapters/anthropic.js";
 import { streamChat as ollamaStreamChat } from "./adapters/ollama.js";
 import { streamChat as geminiStreamChat } from "./adapters/gemini.js";
+import { streamChat as ollamaLocalStreamChat } from "./adapters/ollamaLocal.js";
 
 const PROVIDER_ADAPTERS = {
   openai: openaiStreamChat,
   anthropic: anthropicStreamChat,
   gemini: geminiStreamChat,
   ollama: ollamaStreamChat,
+  ollama_local: ollamaLocalStreamChat,
   custom: openaiStreamChat,
 };
 
diff --git a/helpers/index.js b/helpers/index.js
index 1e8b979..ce618c7 100644
--- a/helpers/index.js
+++ b/helpers/index.js
@@ -434,7 +434,7 @@ export async function listModels(providerConfig = {}, options = {}) {
     }
   }
 
-  if (provider === "ollama") {
+  if (provider === "ollama" || provider === "ollama_local") {
     const csrf = typeof options?.csrfToken === "string" ? options.csrfToken : "";
     const headers = {
       ...(csrf ? { "x-csrftoken": csrf } : {}),
diff --git a/storage/llmProviderStorage.js b/storage/llmProviderStorage.js
index 04f2ed0..bc9b3bf 100644
--- a/storage/llmProviderStorage.js
+++ b/storage/llmProviderStorage.js
@@ -12,6 +12,7 @@ const PROVIDER_PRESETS = {
   anthropic: { baseUrl: "https://api.anthropic.com/v1", label: "Anthropic", thinkingBudget: 4096 },
   gemini: { baseUrl: "", label: "Google AI Studio" },
   ollama: { baseUrl: "", label: "Ollama Cloud" },
+  ollama_local: { baseUrl: "http://localhost:11434", label: "Ollama (local)" },
   custom: { baseUrl: "", label: "Local / Custom" },
 };