- Gateway/media: route text-only `chat.send` image offloads through media-understanding fields so `agents.defaults.imageModel` can describe WebChat attachments instead of leaving only an opaque `media://inbound` marker. Fixes #72968. Thanks @vorajeeah.

- Gateway/Windows: route no-listener restart handoffs through the Windows supervisor without leaving restart tokens in flight, so failed task scheduling can be retried and successful handoffs do not coalesce later restart requests. (#69056) Thanks @Thatgfsj.

- Gateway/model pricing: skip plugin manifest discovery during background pricing refreshes when `plugins.enabled: false`, so disabled-plugin setups do not keep rebuilding plugin metadata from the Gateway hot path. Fixes #73291. Thanks @slideshow-dingo and @fishgills.

- Ollama/thinking: validate `/think` commands against live Ollama catalog reasoning metadata, so models whose `/api/show` capabilities include `thinking` expose `low`, `medium`, `high`, and `max` instead of being stuck on `off`. Fixes #73366. Thanks @cymise.

- Ollama/thinking: validate `/think` commands against live Ollama catalog reasoning metadata and preserve explicit native `params.think`/`params.thinking`, so models whose `/api/show` capabilities include `thinking` expose `low`, `medium`, `high`, and `max` instead of being stuck on `off`. Fixes #73366. Thanks @cymise.

- Gateway/sessions: remove automatic oversized `sessions.json` rotation backups, deprecate `session.maintenance.rotateBytes`, and teach `openclaw doctor --fix` to remove the ignored key so hot session writes no longer copy multi-MB stores. Refs #72338. Thanks @midhunmonachan and @DougButdorf.

- Channels/Telegram: fail fast when Telegram rejects the startup `getMe` token probe with 401, so invalid or stale BotFather tokens are reported as token auth failures instead of misleading `deleteWebhook` cleanup failures. Fixes #47674. Thanks @samaedan-arch.

- ACPX: keep generated Codex and Claude ACP wrapper startup paths working when remote or special state filesystems reject chmod, since OpenClaw invokes the wrappers through Node instead of executing them directly. Fixes #73333. Thanks @david-garcia-garcia.

}

```

Per-model `params.think` or `params.thinking` can disable or force Ollama API thinking for a specific configured model. Runtime commands such as `/think off` still apply to the active run.

Per-model `params.think` or `params.thinking` can disable or force Ollama API thinking for a specific configured model. OpenClaw preserves those explicit model params when the active run only has the implicit default `off`; non-off runtime commands such as `/think medium` still override the active run.

</Accordion>

<Accordion title="Streaming configuration">

OpenClaw's Ollama integration uses the **native Ollama API** (`/api/chat`) by default, which fully supports streaming and tool calling simultaneously. No special configuration is needed.

For native `/api/chat` requests, OpenClaw also forwards thinking control directly to Ollama: `/think off` and `openclaw agent --thinking off` send top-level `think: false`, while `/think low|medium|high` send the matching top-level `think` effort string. `/think max` maps to Ollama's highest native effort, `think: "high"`.

For native `/api/chat` requests, OpenClaw also forwards thinking control directly to Ollama: `/think off` and `openclaw agent --thinking off` send top-level `think: false` unless an explicit model `params.think`/`params.thinking` value is configured, while `/think low|medium|high` send the matching top-level `think` effort string. `/think max` maps to Ollama's highest native effort, `think: "high"`.

<Tip>

If you need to use the OpenAI-compatible endpoint, see the "Legacy OpenAI-compatible mode" section above. Streaming and tool calling may not work simultaneously in that mode.

);

});

it("does not overwrite configured native Ollama params.thinking with implicit off", async () => {

await withMockNdjsonFetch(

[

'{"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}',

],

async (fetchMock) => {

const baseStreamFn = createOllamaStreamFn("http://ollama-host:11434");

const model = {

api: "ollama",

provider: "ollama",

id: "qwen3:32b",

contextWindow: 131072,

params: { thinking: "medium" },

};

const wrapped = createConfiguredOllamaCompatStreamWrapper({

provider: "ollama",

modelId: "qwen3:32b",

model,

streamFn: baseStreamFn,

thinkingLevel: "off",

} as never);

if (!wrapped) {

throw new Error("Expected wrapped Ollama stream function");

}

const stream = await Promise.resolve(

wrapped(

model as never,

{

messages: [{ role: "user", content: "hello" }],

} as never,

{} as never,

),

);

await collectStreamEvents(stream);

const requestInit = getGuardedFetchCall(fetchMock).init ?? {};

if (typeof requestInit.body !== "string") {

throw new Error("Expected string request body");

}

const requestBody = JSON.parse(requestInit.body) as { think?: string };

expect(requestBody.think).toBe("medium");

},

);

});

it("forwards the native think effort on native Ollama chat requests when thinking is enabled", async () => {

await withMockNdjsonFetch(

[

streamFn = wrapOllamaCompatNumCtx(streamFn, resolveOllamaNumCtx(model));

}

const ollamaThinkValue = isNativeOllamaTransport

const configuredThinkValue = model ? resolveOllamaThinkParamValue(model.params) : undefined;

const runtimeThinkValue = isNativeOllamaTransport

? resolveOllamaThinkValue(ctx.thinkingLevel)

: undefined;

// "off" is also the implicit agent default. Preserve explicit native Ollama

// model config unless the active run requests a non-off thinking level.

const ollamaThinkValue =

runtimeThinkValue === false && configuredThinkValue !== undefined

? undefined

: runtimeThinkValue;

if (ollamaThinkValue !== undefined) {

streamFn = createOllamaThinkingWrapper(streamFn, ollamaThinkValue);

}