fix: harden native ollama stream routing + options (openclaw#11853) thanks @BrokenFinger98

CHANGELOG.md

@@ -69,6 +69,7 @@ Docs: https://docs.openclaw.ai
 - Config: keep legacy audio transcription migration strict by rejecting non-string/unsafe command tokens while still migrating valid custom script executables. (#5042) Thanks @shayan919293.
 - Status/Sessions: stop clamping derived `totalTokens` to context-window size, keep prompt-token snapshots wired through session accounting, and surface context usage as unknown when fresh snapshot data is missing to avoid false 100% reports. (#15114) Thanks @echoVic.
 - Providers/MiniMax: switch implicit MiniMax API-key provider from `openai-completions` to `anthropic-messages` with the correct Anthropic-compatible base URL, fixing `invalid role: developer (2013)` errors on MiniMax M2.5. (#15275) Thanks @lailoo.
+- Ollama/Agents: use resolved model/provider base URLs for native `/api/chat` streaming (including aliased providers), normalize `/v1` endpoints, and forward abort + `maxTokens` stream options for reliable cancellation and token caps. (#11853) Thanks @BrokenFinger98.
 
 ## 2026.2.12
 

src/agents/ollama-stream.test.ts

@@ -1,5 +1,6 @@
-import { describe, expect, it } from "vitest";
+import { describe, expect, it, vi } from "vitest";
 import {
+  createOllamaStreamFn,
   convertToOllamaMessages,
   buildAssistantMessage,
   parseNdjsonStream,
@@ -228,3 +229,59 @@ describe("parseNdjsonStream", () => {
     expect(chunks[2].message.tool_calls).toBeUndefined();
   });
 });
+
+describe("createOllamaStreamFn", () => {
+  it("normalizes /v1 baseUrl and maps maxTokens + signal", async () => {
+    const originalFetch = globalThis.fetch;
+    const fetchMock = vi.fn(async () => {
+      const payload = [
+        '{"model":"m","created_at":"t","message":{"role":"assistant","content":"ok"},"done":false}',
+        '{"model":"m","created_at":"t","message":{"role":"assistant","content":""},"done":true,"prompt_eval_count":1,"eval_count":1}',
+      ].join("\n");
+      return new Response(`${payload}\n`, {
+        status: 200,
+        headers: { "Content-Type": "application/x-ndjson" },
+      });
+    });
+    globalThis.fetch = fetchMock as unknown as typeof fetch;
+
+    try {
+      const streamFn = createOllamaStreamFn("http://ollama-host:11434/v1/");
+      const signal = new AbortController().signal;
+      const stream = streamFn(
+        {
+          id: "qwen3:32b",
+          api: "ollama",
+          provider: "custom-ollama",
+          contextWindow: 131072,
+        } as unknown as Parameters<typeof streamFn>[0],
+        {
+          messages: [{ role: "user", content: "hello" }],
+        } as unknown as Parameters<typeof streamFn>[1],
+        {
+          maxTokens: 123,
+          signal,
+        } as unknown as Parameters<typeof streamFn>[2],
+      );
+
+      const events = [];
+      for await (const event of stream) {
+        events.push(event);
+      }
+      expect(events.at(-1)?.type).toBe("done");
+
+      expect(fetchMock).toHaveBeenCalledTimes(1);
+      const [url, requestInit] = fetchMock.mock.calls[0] as [string, RequestInit];
+      expect(url).toBe("http://ollama-host:11434/api/chat");
+      expect(requestInit.signal).toBe(signal);
+
+      const requestBody = JSON.parse(String(requestInit.body)) as {
+        options: { num_ctx?: number; num_predict?: number };
+      };
+      expect(requestBody.options.num_ctx).toBe(131072);
+      expect(requestBody.options.num_predict).toBe(123);
+    } finally {
+      globalThis.fetch = originalFetch;
+    }
+  });
+});

src/agents/ollama-stream.ts

@@ -10,6 +10,8 @@ import type {
 import { createAssistantMessageEventStream } from "@mariozechner/pi-ai";
 import { randomUUID } from "node:crypto";
 
+export const OLLAMA_NATIVE_BASE_URL = "http://127.0.0.1:11434";
+
 // ── Ollama /api/chat request types ──────────────────────────────────────────
 
 interface OllamaChatRequest {
@@ -273,8 +275,15 @@ export async function* parseNdjsonStream(
 
 // ── Main StreamFn factory ───────────────────────────────────────────────────
 
+function resolveOllamaChatUrl(baseUrl: string): string {
+  const trimmed = baseUrl.trim().replace(/\/+$/, "");
+  const normalizedBase = trimmed.replace(/\/v1$/i, "");
+  const apiBase = normalizedBase || OLLAMA_NATIVE_BASE_URL;
+  return `${apiBase}/api/chat`;
+}
+
 export function createOllamaStreamFn(baseUrl: string): StreamFn {
-  const chatUrl = `${baseUrl.replace(/\/+$/, "")}/api/chat`;
+  const chatUrl = resolveOllamaChatUrl(baseUrl);
 
   return (model, context, options) => {
     const stream = createAssistantMessageEventStream();
@@ -294,6 +303,9 @@ export function createOllamaStreamFn(baseUrl: string): StreamFn {
         if (typeof options?.temperature === "number") {
           ollamaOptions.temperature = options.temperature;
         }
+        if (typeof options?.maxTokens === "number") {
+          ollamaOptions.num_predict = options.maxTokens;
+        }
 
         const body: OllamaChatRequest = {
           model: model.id,
@@ -315,6 +327,7 @@ export function createOllamaStreamFn(baseUrl: string): StreamFn {
           method: "POST",
           headers,
           body: JSON.stringify(body),
+          signal: options?.signal,
         });
 
         if (!response.ok) {
@@ -404,5 +417,3 @@ export function createOllamaStreamFn(baseUrl: string): StreamFn {
     return stream;
   };
 }
-
-export const OLLAMA_NATIVE_BASE_URL = "http://127.0.0.1:11434";

src/agents/pi-embedded-runner/run/attempt.ts

@@ -588,8 +588,13 @@ export async function runEmbeddedAttempt(
       // Ollama native API: bypass SDK's streamSimple and use direct /api/chat calls
       // for reliable streaming + tool calling support (#11828).
       if (params.model.api === "ollama") {
-        const providerConfig = params.config?.models?.providers?.ollama;
-        const ollamaBaseUrl = providerConfig?.baseUrl ?? OLLAMA_NATIVE_BASE_URL;
+        // Use the resolved model baseUrl first so custom provider aliases work.
+        const providerConfig = params.config?.models?.providers?.[params.model.provider];
+        const modelBaseUrl =
+          typeof params.model.baseUrl === "string" ? params.model.baseUrl.trim() : "";
+        const providerBaseUrl =
+          typeof providerConfig?.baseUrl === "string" ? providerConfig.baseUrl.trim() : "";
+        const ollamaBaseUrl = modelBaseUrl || providerBaseUrl || OLLAMA_NATIVE_BASE_URL;
         activeSession.agent.streamFn = createOllamaStreamFn(ollamaBaseUrl);
       } else {
         // Force a stable streamFn reference so vitest can reliably mock @mariozechner/pi-ai.