feat: add fast mode toggle for OpenAI models

2026-03-12 23:30:58 +00:00 · 2026-03-12 23:30:58 +00:00 · d5bffcdeab
parent ddcaec89e9
commit d5bffcdeab
66 changed files with 990 additions and 36 deletions
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@ -10,6 +10,7 @@ Docs: https://docs.openclaw.ai
 - Docs/Kubernetes: Add a starter K8s install path with raw manifests, Kind setup, and deployment docs. Thanks @sallyom @dzianisv @egkristi
 - Control UI/dashboard-v2: refresh the gateway dashboard with modular overview, chat, config, agent, and session views, plus a command palette, mobile bottom tabs, and richer chat tools like slash commands, search, export, and pinned messages. (#41503) Thanks @BunsDev.
 - Models/plugins: move Ollama, vLLM, and SGLang onto the provider-plugin architecture, with provider-owned onboarding, discovery, model-picker setup, and post-selection hooks so core provider wiring is more modular.
+- OpenAI/GPT-5.4 fast mode: add configurable session-level fast toggles across `/fast`, TUI, Control UI, and ACP, with per-model config defaults and OpenAI/Codex request shaping.

 ### Fixes

--- a/docs/concepts/model-providers.md
+++ b/docs/concepts/model-providers.md
@ -47,6 +47,7 @@ OpenClaw ships with the pi‑ai catalog. These providers require **no**
 - Override per model via `agents.defaults.models["openai/<model>"].params.transport` (`"sse"`, `"websocket"`, or `"auto"`)
 - OpenAI Responses WebSocket warm-up defaults to enabled via `params.openaiWsWarmup` (`true`/`false`)
 - OpenAI priority processing can be enabled via `agents.defaults.models["openai/<model>"].params.serviceTier`
+- OpenAI fast mode can be enabled per model via `agents.defaults.models["<provider>/<model>"].params.fastMode`

 ```json5
 {
@ -78,6 +79,7 @@ OpenClaw ships with the pi‑ai catalog. These providers require **no**
 - CLI: `openclaw onboard --auth-choice openai-codex` or `openclaw models auth login --provider openai-codex`
 - Default transport is `auto` (WebSocket-first, SSE fallback)
 - Override per model via `agents.defaults.models["openai-codex/<model>"].params.transport` (`"sse"`, `"websocket"`, or `"auto"`)
+- Shares the same `/fast` toggle and `params.fastMode` config as direct `openai/*`
 - Policy note: OpenAI Codex OAuth is explicitly supported for external tools/workflows like OpenClaw.

 ```json5
--- a/docs/concepts/session.md
+++ b/docs/concepts/session.md
@ -281,7 +281,7 @@ Runtime override (owner only):
 - `openclaw status` — shows store path and recent sessions.
 - `openclaw sessions --json` — dumps every entry (filter with `--active <minutes>`).
 - `openclaw gateway call sessions.list --params '{}'` — fetch sessions from the running gateway (use `--url`/`--token` for remote gateway access).
- Send `/status` as a standalone message in chat to see whether the agent is reachable, how much of the session context is used, current thinking/verbose toggles, and when your WhatsApp web creds were last refreshed (helps spot relink needs).
+- Send `/status` as a standalone message in chat to see whether the agent is reachable, how much of the session context is used, current thinking/fast/verbose toggles, and when your WhatsApp web creds were last refreshed (helps spot relink needs).
 - Send `/context list` or `/context detail` to see what’s in the system prompt and injected workspace files (and the biggest context contributors).
 - Send `/stop` (or standalone abort phrases like `stop`, `stop action`, `stop run`, `stop openclaw`) to abort the current run, clear queued followups for that session, and stop any sub-agent runs spawned from it (the reply includes the stopped count).
 - Send `/compact` (optional instructions) as a standalone message to summarize older context and free up window space. See [/concepts/compaction](/concepts/compaction).
--- a/docs/providers/openai.md
+++ b/docs/providers/openai.md
@ -165,6 +165,46 @@ pass that field through on direct `openai/*` Responses requests.

 Supported values are `auto`, `default`, `flex`, and `priority`.

+### OpenAI fast mode
+
+OpenClaw exposes a shared fast-mode toggle for both `openai/*` and
+`openai-codex/*` sessions:
+
+- Chat/UI: `/fast status|on|off`
+- Config: `agents.defaults.models["<provider>/<model>"].params.fastMode`
+
+When fast mode is enabled, OpenClaw applies a low-latency OpenAI profile:
+
+- `reasoning.effort = "low"` when the payload does not already specify reasoning
+- `text.verbosity = "low"` when the payload does not already specify verbosity
+- `service_tier = "priority"` for direct `openai/*` Responses calls to `api.openai.com`
+
+Example:
+
+```json5
+{
+  agents: {
+    defaults: {
+      models: {
+        "openai/gpt-5.4": {
+          params: {
+            fastMode: true,
+          },
+        },
+        "openai-codex/gpt-5.4": {
+          params: {
+            fastMode: true,
+          },
+        },
+      },
+    },
+  },
+}
+```
+
+Session overrides win over config. Clearing the session override in the Sessions UI
+returns the session to the configured default.
+
 ### OpenAI Responses server-side compaction

 For direct OpenAI Responses models (`openai/*` using `api: "openai-responses"` with
--- a/docs/tools/slash-commands.md
+++ b/docs/tools/slash-commands.md
@ -14,7 +14,7 @@ The host-only bash chat command uses `! <cmd>` (with `/bash <cmd>` as an alias).
 There are two related systems:

 - **Commands**: standalone `/...` messages.
- **Directives**: `/think`, `/verbose`, `/reasoning`, `/elevated`, `/exec`, `/model`, `/queue`.
+- **Directives**: `/think`, `/fast`, `/verbose`, `/reasoning`, `/elevated`, `/exec`, `/model`, `/queue`.
  - Directives are stripped from the message before the model sees it.
  - In normal chat messages (not directive-only), they are treated as “inline hints” and do **not** persist session settings.
  - In directive-only messages (the message contains only directives), they persist to the session and reply with an acknowledgement.
@ -102,6 +102,7 @@ Text + native (when enabled):
 - `/send on|off|inherit` (owner-only)
 - `/reset` or `/new [model]` (optional model hint; remainder is passed through)
 - `/think <off|minimal|low|medium|high|xhigh>` (dynamic choices by model/provider; aliases: `/thinking`, `/t`)
+- `/fast status|on|off` (omitting the arg shows the current effective fast-mode state)
 - `/verbose on|full|off` (alias: `/v`)
 - `/reasoning on|off|stream` (alias: `/reason`; when on, sends a separate message prefixed `Reasoning:`; `stream` = Telegram draft only)
 - `/elevated on|off|ask|full` (alias: `/elev`; `full` skips exec approvals)
@ -130,6 +131,7 @@ Notes:
 - Discord thread-binding commands (`/focus`, `/unfocus`, `/agents`, `/session idle`, `/session max-age`) require effective thread bindings to be enabled (`session.threadBindings.enabled` and/or `channels.discord.threadBindings.enabled`).
 - ACP command reference and runtime behavior: [ACP Agents](/tools/acp-agents).
 - `/verbose` is meant for debugging and extra visibility; keep it **off** in normal use.
+- `/fast on|off` persists a session override. Use the Sessions UI `inherit` option to clear it and fall back to config defaults.
 - Tool failure summaries are still shown when relevant, but detailed failure text is only included when `/verbose` is `on` or `full`.
 - `/reasoning` (and `/verbose`) are risky in group settings: they may reveal internal reasoning or tool output you did not intend to expose. Prefer leaving them off, especially in group chats.
 - **Fast path:** command-only messages from allowlisted senders are handled immediately (bypass queue + model).
--- a/docs/tools/thinking.md
+++ b/docs/tools/thinking.md
@ -1,7 +1,7 @@
 ---
-summary: "Directive syntax for /think + /verbose and how they affect model reasoning"
+summary: "Directive syntax for /think, /fast, /verbose, and reasoning visibility"
 read_when:
-  - Adjusting thinking or verbose directive parsing or defaults
+  - Adjusting thinking, fast-mode, or verbose directive parsing or defaults
 title: "Thinking Levels"
 ---

@ -42,6 +42,19 @@ title: "Thinking Levels"

 - **Embedded Pi**: the resolved level is passed to the in-process Pi agent runtime.

+## Fast mode (/fast)
+
+- Levels: `on|off`.
+- Directive-only message toggles a session fast-mode override and replies `Fast mode enabled.` / `Fast mode disabled.`.
+- Send `/fast` (or `/fast status`) with no mode to see the current effective fast-mode state.
+- OpenClaw resolves fast mode in this order:
+  1. Inline/directive-only `/fast on|off`
+  2. Session override
+  3. Per-model config: `agents.defaults.models["<provider>/<model>"].params.fastMode`
+  4. Fallback: `off`
+- For `openai/*`, fast mode applies the OpenAI fast profile: `service_tier=priority` when supported, plus low reasoning effort and low text verbosity.
+- For `openai-codex/*`, fast mode applies the same low-latency profile on Codex Responses. OpenClaw keeps one shared `/fast` toggle across both auth paths.
+
 ## Verbose directives (/verbose or /v)

 - Levels: `on` (minimal) | `full` | `off` (default).
--- a/docs/web/control-ui.md
+++ b/docs/web/control-ui.md
@ -75,7 +75,7 @@ The Control UI can localize itself on first load based on your browser locale, a
 - Stream tool calls + live tool output cards in Chat (agent events)
 - Channels: WhatsApp/Telegram/Discord/Slack + plugin channels (Mattermost, etc.) status + QR login + per-channel config (`channels.status`, `web.login.*`, `config.patch`)
 - Instances: presence list + refresh (`system-presence`)
- Sessions: list + per-session thinking/verbose overrides (`sessions.list`, `sessions.patch`)
+- Sessions: list + per-session thinking/fast/verbose/reasoning overrides (`sessions.list`, `sessions.patch`)
 - Cron jobs: list/add/edit/run/enable/disable + run history (`cron.*`)
 - Skills: status, enable/disable, install, API key updates (`skills.*`)
 - Nodes: list + caps (`node.list`)
--- a/docs/web/tui.md
+++ b/docs/web/tui.md
@ -37,7 +37,7 @@ Use `--password` if your Gateway uses password auth.
 - Header: connection URL, current agent, current session.
 - Chat log: user messages, assistant replies, system notices, tool cards.
 - Status line: connection/run state (connecting, running, streaming, idle, error).
- Footer: connection state + agent + session + model + think/verbose/reasoning + token counts + deliver.
+- Footer: connection state + agent + session + model + think/fast/verbose/reasoning + token counts + deliver.
 - Input: text editor with autocomplete.

 ## Mental model: agents + sessions
@ -92,6 +92,7 @@ Core:
 Session controls:

 - `/think <off|minimal|low|medium|high>`
+- `/fast <status|on|off>`
 - `/verbose <on|full|off>`
 - `/reasoning <on|off|stream>`
 - `/usage <off|tokens|full>`
--- a/src/acp/translator.session-rate-limit.test.ts
+++ b/src/acp/translator.session-rate-limit.test.ts
@ -645,6 +645,77 @@ describe("acp setSessionConfigOption bridge behavior", () => {
    sessionStore.clearAllSessionsForTest();
  });

+  it("updates fast mode ACP config options through gateway session patches", async () => {
+    const sessionStore = createInMemorySessionStore();
+    const connection = createAcpConnection();
+    const sessionUpdate = connection.__sessionUpdateMock;
+    const request = vi.fn(async (method: string, params?: unknown) => {
+      if (method === "sessions.list") {
+        return {
+          ts: Date.now(),
+          path: "/tmp/sessions.json",
+          count: 1,
+          defaults: {
+            modelProvider: null,
+            model: null,
+            contextTokens: null,
+          },
+          sessions: [
+            {
+              key: "fast-session",
+              kind: "direct",
+              updatedAt: Date.now(),
+              thinkingLevel: "minimal",
+              modelProvider: "openai",
+              model: "gpt-5.4",
+              fastMode: true,
+            },
+          ],
+        };
+      }
+      if (method === "sessions.patch") {
+        expect(params).toEqual({
+          key: "fast-session",
+          fastMode: true,
+        });
+      }
+      return { ok: true };
+    }) as GatewayClient["request"];
+    const agent = new AcpGatewayAgent(connection, createAcpGateway(request), {
+      sessionStore,
+    });
+
+    await agent.loadSession(createLoadSessionRequest("fast-session"));
+    sessionUpdate.mockClear();
+
+    const result = await agent.setSessionConfigOption(
+      createSetSessionConfigOptionRequest("fast-session", "fast_mode", "on"),
+    );
+
+    expect(result.configOptions).toEqual(
+      expect.arrayContaining([
+        expect.objectContaining({
+          id: "fast_mode",
+          currentValue: "on",
+        }),
+      ]),
+    );
+    expect(sessionUpdate).toHaveBeenCalledWith({
+      sessionId: "fast-session",
+      update: {
+        sessionUpdate: "config_option_update",
+        configOptions: expect.arrayContaining([
+          expect.objectContaining({
+            id: "fast_mode",
+            currentValue: "on",
+          }),
+        ]),
+      },
+    });
+
+    sessionStore.clearAllSessionsForTest();
+  });
+
  it("rejects non-string ACP config option values", async () => {
    const sessionStore = createInMemorySessionStore();
    const connection = createAcpConnection();
--- a/src/acp/translator.ts
+++ b/src/acp/translator.ts
@ -53,6 +53,7 @@ import { ACP_AGENT_INFO, type AcpServerOptions } from "./types.js";
 // Maximum allowed prompt size (2MB) to prevent DoS via memory exhaustion (CWE-400, GHSA-cxpw-2g23-2vgw)
 const MAX_PROMPT_BYTES = 2 * 1024 * 1024;
 const ACP_THOUGHT_LEVEL_CONFIG_ID = "thought_level";
+const ACP_FAST_MODE_CONFIG_ID = "fast_mode";
 const ACP_VERBOSE_LEVEL_CONFIG_ID = "verbose_level";
 const ACP_REASONING_LEVEL_CONFIG_ID = "reasoning_level";
 const ACP_RESPONSE_USAGE_CONFIG_ID = "response_usage";
@ -88,6 +89,7 @@ type GatewaySessionPresentationRow = Pick<
  | "derivedTitle"
  | "updatedAt"
  | "thinkingLevel"
+  | "fastMode"
  | "modelProvider"
  | "model"
  | "verboseLevel"
@ -209,6 +211,13 @@ function buildSessionPresentation(params: {
      currentValue: currentModeId,
      values: availableLevelIds,
    }),
+    buildSelectConfigOption({
+      id: ACP_FAST_MODE_CONFIG_ID,
+      name: "Fast mode",
+      description: "Controls whether OpenAI sessions use the Gateway fast-mode profile.",
+      currentValue: row.fastMode ? "on" : "off",
+      values: ["off", "on"],
+    }),
    buildSelectConfigOption({
      id: ACP_VERBOSE_LEVEL_CONFIG_ID,
      name: "Tool verbosity",
@ -925,6 +934,7 @@ export class AcpGatewayAgent implements Agent {
      thinkingLevel: session.thinkingLevel,
      modelProvider: session.modelProvider,
      model: session.model,
+      fastMode: session.fastMode,
      verboseLevel: session.verboseLevel,
      reasoningLevel: session.reasoningLevel,
      responseUsage: session.responseUsage,
@ -940,7 +950,7 @@ export class AcpGatewayAgent implements Agent {
    value: string | boolean,
  ): {
    overrides: Partial<GatewaySessionPresentationRow>;
-    patch: Record<string, string>;
+    patch: Record<string, string | boolean>;
  } {
    if (typeof value !== "string") {
      throw new Error(
@ -953,6 +963,11 @@ export class AcpGatewayAgent implements Agent {
          patch: { thinkingLevel: value },
          overrides: { thinkingLevel: value },
        };
+      case ACP_FAST_MODE_CONFIG_ID:
+        return {
+          patch: { fastMode: value === "on" },
+          overrides: { fastMode: value === "on" },
+        };
      case ACP_VERBOSE_LEVEL_CONFIG_ID:
        return {
          patch: { verboseLevel: value },
--- a/src/agents/fast-mode.ts
+++ b/src/agents/fast-mode.ts
@ -0,0 +1,50 @@
+import { normalizeFastMode } from "../auto-reply/thinking.js";
+import type { OpenClawConfig } from "../config/config.js";
+import type { SessionEntry } from "../config/sessions.js";
+
+export type FastModeState = {
+  enabled: boolean;
+  source: "session" | "config" | "default";
+};
+
+function resolveConfiguredFastModeRaw(params: {
+  cfg: OpenClawConfig | undefined;
+  provider: string;
+  model: string;
+}): unknown {
+  const modelKey = `${params.provider}/${params.model}`;
+  const modelConfig = params.cfg?.agents?.defaults?.models?.[modelKey];
+  return modelConfig?.params?.fastMode ?? modelConfig?.params?.fast_mode;
+}
+
+export function resolveConfiguredFastMode(params: {
+  cfg: OpenClawConfig | undefined;
+  provider: string;
+  model: string;
+}): boolean {
+  return (
+    normalizeFastMode(
+      resolveConfiguredFastModeRaw(params) as string | boolean | null | undefined,
+    ) ?? false
+  );
+}
+
+export function resolveFastModeState(params: {
+  cfg: OpenClawConfig | undefined;
+  provider: string;
+  model: string;
+  sessionEntry?: Pick<SessionEntry, "fastMode"> | undefined;
+}): FastModeState {
+  const sessionOverride = normalizeFastMode(params.sessionEntry?.fastMode);
+  if (sessionOverride !== undefined) {
+    return { enabled: sessionOverride, source: "session" };
+  }
+
+  const configuredRaw = resolveConfiguredFastModeRaw(params);
+  const configured = normalizeFastMode(configuredRaw as string | boolean | null | undefined);
+  if (configured !== undefined) {
+    return { enabled: configured, source: "config" };
+  }
+
+  return { enabled: false, source: "default" };
+}
--- a/src/agents/pi-embedded-runner-extraparams.test.ts
+++ b/src/agents/pi-embedded-runner-extraparams.test.ts
@ -204,6 +204,7 @@ describe("applyExtraParamsToAgent", () => {
      | Model<"openai-completions">;
    options?: SimpleStreamOptions;
    cfg?: Record<string, unknown>;
+    extraParamsOverride?: Record<string, unknown>;
    payload?: Record<string, unknown>;
  }) {
    const payload = params.payload ?? { store: false };
@ -217,6 +218,7 @@ describe("applyExtraParamsToAgent", () => {
      params.cfg as Parameters<typeof applyExtraParamsToAgent>[1],
      params.applyProvider,
      params.applyModelId,
+      params.extraParamsOverride,
    );
    const context: Context = { messages: [] };
    void agent.streamFn?.(params.model, context, params.options ?? {});
@ -1627,6 +1629,80 @@ describe("applyExtraParamsToAgent", () => {
    expect(payload.service_tier).toBe("default");
  });

+  it("injects fast-mode payload defaults for direct OpenAI Responses", () => {
+    const payload = runResponsesPayloadMutationCase({
+      applyProvider: "openai",
+      applyModelId: "gpt-5.4",
+      cfg: {
+        agents: {
+          defaults: {
+            models: {
+              "openai/gpt-5.4": {
+                params: {
+                  fastMode: true,
+                },
+              },
+            },
+          },
+        },
+      },
+      model: {
+        api: "openai-responses",
+        provider: "openai",
+        id: "gpt-5.4",
+        baseUrl: "https://api.openai.com/v1",
+      } as unknown as Model<"openai-responses">,
+      payload: {
+        store: false,
+      },
+    });
+    expect(payload.reasoning).toEqual({ effort: "low" });
+    expect(payload.text).toEqual({ verbosity: "low" });
+    expect(payload.service_tier).toBe("priority");
+  });
+
+  it("preserves caller-provided OpenAI payload fields when fast mode is enabled", () => {
+    const payload = runResponsesPayloadMutationCase({
+      applyProvider: "openai",
+      applyModelId: "gpt-5.4",
+      extraParamsOverride: { fastMode: true },
+      model: {
+        api: "openai-responses",
+        provider: "openai",
+        id: "gpt-5.4",
+        baseUrl: "https://api.openai.com/v1",
+      } as unknown as Model<"openai-responses">,
+      payload: {
+        reasoning: { effort: "medium" },
+        text: { verbosity: "high" },
+        service_tier: "default",
+      },
+    });
+    expect(payload.reasoning).toEqual({ effort: "medium" });
+    expect(payload.text).toEqual({ verbosity: "high" });
+    expect(payload.service_tier).toBe("default");
+  });
+
+  it("applies fast-mode defaults for openai-codex responses without service_tier", () => {
+    const payload = runResponsesPayloadMutationCase({
+      applyProvider: "openai-codex",
+      applyModelId: "gpt-5.4",
+      extraParamsOverride: { fastMode: true },
+      model: {
+        api: "openai-codex-responses",
+        provider: "openai-codex",
+        id: "gpt-5.4",
+        baseUrl: "https://chatgpt.com/backend-api",
+      } as unknown as Model<"openai-codex-responses">,
+      payload: {
+        store: false,
+      },
+    });
+    expect(payload.reasoning).toEqual({ effort: "low" });
+    expect(payload.text).toEqual({ verbosity: "low" });
+    expect(payload).not.toHaveProperty("service_tier");
+  });
+
  it("does not inject service_tier for non-openai providers", () => {
    const payload = runResponsesPayloadMutationCase({
      applyProvider: "azure-openai-responses",
--- a/src/agents/pi-embedded-runner/extra-params.ts
+++ b/src/agents/pi-embedded-runner/extra-params.ts
@ -22,8 +22,10 @@ import {
 import {
  createCodexDefaultTransportWrapper,
  createOpenAIDefaultTransportWrapper,
+  createOpenAIFastModeWrapper,
  createOpenAIResponsesContextManagementWrapper,
  createOpenAIServiceTierWrapper,
+  resolveOpenAIFastMode,
  resolveOpenAIServiceTier,
 } from "./openai-stream-wrappers.js";
 import {
@ -437,6 +439,12 @@ export function applyExtraParamsToAgent(
  // upstream model-ID heuristics for Gemini 3.1 variants.
  agent.streamFn = createGoogleThinkingPayloadWrapper(agent.streamFn, thinkingLevel);

+  const openAIFastMode = resolveOpenAIFastMode(merged);
+  if (openAIFastMode) {
+    log.debug(`applying OpenAI fast mode for ${provider}/${modelId}`);
+    agent.streamFn = createOpenAIFastModeWrapper(agent.streamFn);
+  }
+
  const openAIServiceTier = resolveOpenAIServiceTier(merged);
  if (openAIServiceTier) {
    log.debug(`applying OpenAI service_tier=${openAIServiceTier} for ${provider}/${modelId}`);
--- a/src/agents/pi-embedded-runner/openai-stream-wrappers.ts
+++ b/src/agents/pi-embedded-runner/openai-stream-wrappers.ts
@ -4,6 +4,7 @@ import { streamSimple } from "@mariozechner/pi-ai";
 import { log } from "./logger.js";

 type OpenAIServiceTier = "auto" | "default" | "flex" | "priority";
+type OpenAIReasoningEffort = "low" | "medium" | "high";

 const OPENAI_RESPONSES_APIS = new Set(["openai-responses"]);
 const OPENAI_RESPONSES_PROVIDERS = new Set(["openai", "azure-openai", "azure-openai-responses"]);
@ -168,6 +169,89 @@ export function resolveOpenAIServiceTier(
  return normalized;
 }

+function normalizeOpenAIFastMode(value: unknown): boolean | undefined {
+  if (typeof value === "boolean") {
+    return value;
+  }
+  if (typeof value !== "string") {
+    return undefined;
+  }
+  const normalized = value.trim().toLowerCase();
+  if (
+    normalized === "on" ||
+    normalized === "true" ||
+    normalized === "yes" ||
+    normalized === "1" ||
+    normalized === "fast"
+  ) {
+    return true;
+  }
+  if (
+    normalized === "off" ||
+    normalized === "false" ||
+    normalized === "no" ||
+    normalized === "0" ||
+    normalized === "normal"
+  ) {
+    return false;
+  }
+  return undefined;
+}
+
+export function resolveOpenAIFastMode(
+  extraParams: Record<string, unknown> | undefined,
+): boolean | undefined {
+  const raw = extraParams?.fastMode ?? extraParams?.fast_mode;
+  const normalized = normalizeOpenAIFastMode(raw);
+  if (raw !== undefined && normalized === undefined) {
+    const rawSummary = typeof raw === "string" ? raw : typeof raw;
+    log.warn(`ignoring invalid OpenAI fast mode param: ${rawSummary}`);
+  }
+  return normalized;
+}
+
+function resolveFastModeReasoningEffort(modelId: unknown): OpenAIReasoningEffort {
+  if (typeof modelId !== "string") {
+    return "low";
+  }
+  const normalized = modelId.trim().toLowerCase();
+  // Keep fast mode broadly compatible across GPT-5 family variants by using
+  // the lowest shared non-disabled effort that current transports accept.
+  if (normalized.startsWith("gpt-5")) {
+    return "low";
+  }
+  return "low";
+}
+
+function applyOpenAIFastModePayloadOverrides(params: {
+  payloadObj: Record<string, unknown>;
+  model: { provider?: unknown; id?: unknown; baseUrl?: unknown; api?: unknown };
+}): void {
+  if (params.payloadObj.reasoning === undefined) {
+    params.payloadObj.reasoning = {
+      effort: resolveFastModeReasoningEffort(params.model.id),
+    };
+  }
+
+  const existingText = params.payloadObj.text;
+  if (existingText === undefined) {
+    params.payloadObj.text = { verbosity: "low" };
+  } else if (existingText && typeof existingText === "object" && !Array.isArray(existingText)) {
+    const textObj = existingText as Record<string, unknown>;
+    if (textObj.verbosity === undefined) {
+      textObj.verbosity = "low";
+    }
+  }
+
+  if (
+    params.model.provider === "openai" &&
+    params.payloadObj.service_tier === undefined &&
+    isOpenAIPublicApiBaseUrl(params.model.baseUrl)
+  ) {
+    params.payloadObj.service_tier = "priority";
+  }
+}
+
 export function createOpenAIResponsesContextManagementWrapper(
  baseStreamFn: StreamFn | undefined,
  extraParams: Record<string, unknown> | undefined,
@ -203,6 +287,31 @@ export function createOpenAIResponsesContextManagementWrapper(
  };
 }

+export function createOpenAIFastModeWrapper(baseStreamFn: StreamFn | undefined): StreamFn {
+  const underlying = baseStreamFn ?? streamSimple;
+  return (model, context, options) => {
+    if (
+      (model.api !== "openai-responses" && model.api !== "openai-codex-responses") ||
+      (model.provider !== "openai" && model.provider !== "openai-codex")
+    ) {
+      return underlying(model, context, options);
+    }
+    const originalOnPayload = options?.onPayload;
+    return underlying(model, context, {
+      ...options,
+      onPayload: (payload) => {
+        if (payload && typeof payload === "object") {
+          applyOpenAIFastModePayloadOverrides({
+            payloadObj: payload as Record<string, unknown>,
+            model,
+          });
+        }
+        return originalOnPayload?.(payload, model);
+      },
+    });
+  };
+}
+
 export function createOpenAIServiceTierWrapper(
  baseStreamFn: StreamFn | undefined,
  serviceTier: OpenAIServiceTier,
--- a/src/agents/pi-embedded-runner/run.ts
+++ b/src/agents/pi-embedded-runner/run.ts
@ -892,6 +892,7 @@ export async function runEmbeddedPiAgent(
            agentId: workspaceResolution.agentId,
            legacyBeforeAgentStartResult,
            thinkLevel,
+            fastMode: params.fastMode,
            verboseLevel: params.verboseLevel,
            reasoningLevel: params.reasoningLevel,
            toolResultFormat: resolvedToolResultFormat,
--- a/src/agents/pi-embedded-runner/run/attempt.ts
+++ b/src/agents/pi-embedded-runner/run/attempt.ts
@ -1930,7 +1930,10 @@ export async function runEmbeddedAttempt(
        params.config,
        params.provider,
        params.modelId,
-        params.streamParams,
+        {
+          ...params.streamParams,
+          fastMode: params.fastMode,
+        },
        params.thinkLevel,
        sessionAgentId,
      );
--- a/src/agents/pi-embedded-runner/run/params.ts
+++ b/src/agents/pi-embedded-runner/run/params.ts
@ -79,6 +79,7 @@ export type RunEmbeddedPiAgentParams = {
  authProfileId?: string;
  authProfileIdSource?: "auto" | "user";
  thinkLevel?: ThinkLevel;
+  fastMode?: boolean;
  verboseLevel?: VerboseLevel;
  reasoningLevel?: ReasoningLevel;
  toolResultFormat?: ToolResultFormat;
--- a/src/auto-reply/commands-registry.data.ts
+++ b/src/auto-reply/commands-registry.data.ts
@ -597,6 +597,22 @@ function buildChatCommands(): ChatCommandDefinition[] {
      ],
      argsMenu: "auto",
    }),
+    defineChatCommand({
+      key: "fast",
+      nativeName: "fast",
+      description: "Toggle fast mode.",
+      textAlias: "/fast",
+      category: "options",
+      args: [
+        {
+          name: "mode",
+          description: "status, on, or off",
+          type: "string",
+          choices: ["status", "on", "off"],
+        },
+      ],
+      argsMenu: "auto",
+    }),
    defineChatCommand({
      key: "reasoning",
      nativeName: "reasoning",
--- a/src/auto-reply/commands-registry.test.ts
+++ b/src/auto-reply/commands-registry.test.ts
@ -198,6 +198,17 @@ describe("commands registry", () => {
    ]);
  });

+  it("registers fast mode as a first-class options command", () => {
+    const fast = listChatCommands().find((command) => command.key === "fast");
+    expect(fast).toMatchObject({
+      nativeName: "fast",
+      textAliases: ["/fast"],
+      category: "options",
+    });
+    const modeArg = fast?.args?.find((arg) => arg.name === "mode");
+    expect(modeArg?.choices).toEqual(["status", "on", "off"]);
+  });
+
  it("detects known text commands", () => {
    const detection = getCommandDetection();
    expect(detection.exact.has("/commands")).toBe(true);
--- a/src/auto-reply/reply.directive.directive-behavior.e2e-harness.ts
+++ b/src/auto-reply/reply.directive.directive-behavior.e2e-harness.ts
@ -4,6 +4,7 @@ import { withTempHome as withTempHomeBase } from "../../test/helpers/temp-home.j
 import { loadModelCatalog } from "../agents/model-catalog.js";
 import { runEmbeddedPiAgent } from "../agents/pi-embedded.js";
 import { loadSessionStore } from "../config/sessions.js";
+import { runEmbeddedPiAgentMock } from "./reply.directive.directive-behavior.e2e-mocks.js";

 export { loadModelCatalog } from "../agents/model-catalog.js";
 export { runEmbeddedPiAgent } from "../agents/pi-embedded.js";
@ -134,7 +135,7 @@ export function assertElevatedOffStatusReply(text: string | undefined) {

 export function installDirectiveBehaviorE2EHooks() {
  beforeEach(() => {
-    vi.mocked(runEmbeddedPiAgent).mockReset();
+    runEmbeddedPiAgentMock.mockReset();
    vi.mocked(loadModelCatalog).mockResolvedValue(DEFAULT_TEST_MODEL_CATALOG);
  });

--- a/src/auto-reply/reply.directive.directive-behavior.shows-current-verbose-level-verbose-has-no.test.ts
+++ b/src/auto-reply/reply.directive.directive-behavior.shows-current-verbose-level-verbose-has-no.test.ts
@ -1,5 +1,5 @@
 import "./reply.directive.directive-behavior.e2e-mocks.js";
-import { describe, expect, it, vi } from "vitest";
+import { describe, expect, it } from "vitest";
 import type { OpenClawConfig } from "../config/config.js";
 import { loadSessionStore } from "../config/sessions.js";
 import {
@ -10,10 +10,10 @@ import {
  makeRestrictedElevatedDisabledConfig,
  makeWhatsAppDirectiveConfig,
  replyText,
-  runEmbeddedPiAgent,
  sessionStorePath,
  withTempHome,
 } from "./reply.directive.directive-behavior.e2e-harness.js";
+import { runEmbeddedPiAgentMock } from "./reply.directive.directive-behavior.e2e-mocks.js";
 import { getReplyFromConfig } from "./reply.js";

 const COMMAND_MESSAGE_BASE = {
@ -126,6 +126,18 @@ describe("directive behavior", () => {

  it("reports current directive defaults when no arguments are provided", async () => {
    await withTempHome(async (home) => {
+      const fastText = await runCommand(home, "/fast", {
+        defaults: {
+          models: {
+            "anthropic/claude-opus-4-5": {
+              params: { fastMode: true },
+            },
+          },
+        },
+      });
+      expect(fastText).toContain("Current fast mode: on (config)");
+      expect(fastText).toContain("Options: on, off.");
+
      const verboseText = await runCommand(home, "/verbose", {
        defaults: { verboseDefault: "on" },
      });
@ -158,7 +170,28 @@ describe("directive behavior", () => {
      expect(execText).toContain(
        "Options: host=sandbox|gateway|node, security=deny|allowlist|full, ask=off|on-miss|always, node=<id>.",
      );
-      expect(runEmbeddedPiAgent).not.toHaveBeenCalled();
+      expect(runEmbeddedPiAgentMock).not.toHaveBeenCalled();
+    });
+  });
+  it("persists fast toggles across /status and /fast", async () => {
+    await withTempHome(async (home) => {
+      const storePath = sessionStorePath(home);
+
+      const onText = await runCommand(home, "/fast on");
+      expect(onText).toContain("Fast mode enabled");
+      expect(loadSessionStore(storePath)["agent:main:main"]?.fastMode).toBe(true);
+
+      const statusText = await runCommand(home, "/status");
+      const optionsLine = statusText?.split("\n").find((line) => line.trim().startsWith("⚙️"));
+      expect(optionsLine).toContain("Fast: on");
+
+      const offText = await runCommand(home, "/fast off");
+      expect(offText).toContain("Fast mode disabled");
+      expect(loadSessionStore(storePath)["agent:main:main"]?.fastMode).toBe(false);
+
+      const fastText = await runCommand(home, "/fast");
+      expect(fastText).toContain("Current fast mode: off");
+      expect(runEmbeddedPiAgentMock).not.toHaveBeenCalled();
    });
  });
  it("persists elevated toggles across /status and /elevated", async () => {
@ -181,7 +214,7 @@ describe("directive behavior", () => {

      const store = loadSessionStore(storePath);
      expect(store["agent:main:main"]?.elevatedLevel).toBe("on");
-      expect(runEmbeddedPiAgent).not.toHaveBeenCalled();
+      expect(runEmbeddedPiAgentMock).not.toHaveBeenCalled();
    });
  });
  it("enforces per-agent elevated restrictions and status visibility", async () => {
@ -217,7 +250,7 @@ describe("directive behavior", () => {
      );
      const statusText = replyText(statusRes);
      expect(statusText).not.toContain("elevated");
-      expect(runEmbeddedPiAgent).not.toHaveBeenCalled();
+      expect(runEmbeddedPiAgentMock).not.toHaveBeenCalled();
    });
  });
  it("applies per-agent allowlist requirements before allowing elevated", async () => {
@ -245,7 +278,7 @@ describe("directive behavior", () => {

      const allowedText = replyText(allowedRes);
      expect(allowedText).toContain("Elevated mode set to ask");
-      expect(runEmbeddedPiAgent).not.toHaveBeenCalled();
+      expect(runEmbeddedPiAgentMock).not.toHaveBeenCalled();
    });
  });
  it("handles runtime warning, invalid level, and multi-directive elevated inputs", async () => {
@ -280,7 +313,7 @@ describe("directive behavior", () => {
          expect(text).toContain(snippet);
        }
      }
-      expect(runEmbeddedPiAgent).not.toHaveBeenCalled();
+      expect(runEmbeddedPiAgentMock).not.toHaveBeenCalled();
    });
  });
  it("persists queue overrides and reset behavior", async () => {
@ -317,12 +350,12 @@ describe("directive behavior", () => {
      expect(entry?.queueDebounceMs).toBeUndefined();
      expect(entry?.queueCap).toBeUndefined();
      expect(entry?.queueDrop).toBeUndefined();
-      expect(runEmbeddedPiAgent).not.toHaveBeenCalled();
+      expect(runEmbeddedPiAgentMock).not.toHaveBeenCalled();
    });
  });
  it("strips inline elevated directives from the user text (does not persist session override)", async () => {
    await withTempHome(async (home) => {
-      vi.mocked(runEmbeddedPiAgent).mockResolvedValue({
+      runEmbeddedPiAgentMock.mockResolvedValue({
        payloads: [{ text: "ok" }],
        meta: {
          durationMs: 1,
@ -346,7 +379,7 @@ describe("directive behavior", () => {
      const store = loadSessionStore(storePath);
      expect(store["agent:main:main"]?.elevatedLevel).toBeUndefined();

-      const calls = vi.mocked(runEmbeddedPiAgent).mock.calls;
+      const calls = runEmbeddedPiAgentMock.mock.calls;
      expect(calls.length).toBeGreaterThan(0);
      const call = calls[0]?.[0];
      expect(call?.prompt).toContain("hello there");
--- a/src/auto-reply/reply.directive.parse.test.ts
+++ b/src/auto-reply/reply.directive.parse.test.ts
@ -8,7 +8,7 @@ import {
  extractThinkDirective,
  extractVerboseDirective,
 } from "./reply.js";
-import { extractStatusDirective } from "./reply/directives.js";
+import { extractFastDirective, extractStatusDirective } from "./reply/directives.js";

 describe("directive parsing", () => {
  it("ignores verbose directive inside URL", () => {
@ -49,6 +49,12 @@ describe("directive parsing", () => {
    expect(res.reasoningLevel).toBe("stream");
  });

+  it("matches fast directive", () => {
+    const res = extractFastDirective("/fast on please");
+    expect(res.hasDirective).toBe(true);
+    expect(res.fastMode).toBe(true);
+  });
+
  it("matches elevated with leading space", () => {
    const res = extractElevatedDirective(" please /elevated on now");
    expect(res.hasDirective).toBe(true);
@ -106,6 +112,14 @@ describe("directive parsing", () => {
    expect(res.cleaned).toBe("");
  });

+  it("matches fast with no argument", () => {
+    const res = extractFastDirective("/fast:");
+    expect(res.hasDirective).toBe(true);
+    expect(res.fastMode).toBeUndefined();
+    expect(res.rawLevel).toBeUndefined();
+    expect(res.cleaned).toBe("");
+  });
+
  it("matches reasoning with no argument", () => {
    const res = extractReasoningDirective("/reasoning:");
    expect(res.hasDirective).toBe(true);
--- a/src/auto-reply/reply/commands-core.ts
+++ b/src/auto-reply/reply/commands-core.ts
@ -26,6 +26,7 @@ import { handlePluginCommand } from "./commands-plugin.js";
 import {
  handleAbortTrigger,
  handleActivationCommand,
+  handleFastCommand,
  handleRestartCommand,
  handleSessionCommand,
  handleSendPolicyCommand,
@ -176,6 +177,7 @@ export async function handleCommands(params: HandleCommandsParams): Promise<Comm
      handleBashCommand,
      handleActivationCommand,
      handleSendPolicyCommand,
+      handleFastCommand,
      handleUsageCommand,
      handleSessionCommand,
      handleRestartCommand,
--- a/src/auto-reply/reply/commands-session.ts
+++ b/src/auto-reply/reply/commands-session.ts
@ -1,3 +1,4 @@
+import { resolveFastModeState } from "../../agents/fast-mode.js";
 import { parseDurationMs } from "../../cli/parse-duration.js";
 import { isRestartEnabled } from "../../config/commands.js";
 import {
@ -22,7 +23,7 @@ import {
 import { formatTokenCount, formatUsd } from "../../utils/usage-format.js";
 import { parseActivationCommand } from "../group-activation.js";
 import { parseSendPolicyCommand } from "../send-policy.js";
-import { normalizeUsageDisplay, resolveResponseUsageMode } from "../thinking.js";
+import { normalizeFastMode, normalizeUsageDisplay, resolveResponseUsageMode } from "../thinking.js";
 import { isDiscordSurface, isTelegramSurface, resolveChannelAccountId } from "./channel-context.js";
 import { handleAbortTrigger, handleStopCommand } from "./commands-session-abort.js";
 import { persistSessionEntry } from "./commands-session-store.js";
@ -291,6 +292,57 @@ export const handleUsageCommand: CommandHandler = async (params, allowTextComman
  };
 };

+export const handleFastCommand: CommandHandler = async (params, allowTextCommands) => {
+  if (!allowTextCommands) {
+    return null;
+  }
+  const normalized = params.command.commandBodyNormalized;
+  if (normalized !== "/fast" && !normalized.startsWith("/fast ")) {
+    return null;
+  }
+  if (!params.command.isAuthorizedSender) {
+    logVerbose(
+      `Ignoring /fast from unauthorized sender: ${params.command.senderId || "<unknown>"}`,
+    );
+    return { shouldContinue: false };
+  }
+
+  const rawArgs = normalized === "/fast" ? "" : normalized.slice("/fast".length).trim();
+  const rawMode = rawArgs.toLowerCase();
+  if (!rawMode || rawMode === "status") {
+    const state = resolveFastModeState({
+      cfg: params.cfg,
+      provider: params.provider,
+      model: params.model,
+      sessionEntry: params.sessionEntry,
+    });
+    const suffix =
+      state.source === "config" ? " (config)" : state.source === "default" ? " (default)" : "";
+    return {
+      shouldContinue: false,
+      reply: { text: `⚙️ Current fast mode: ${state.enabled ? "on" : "off"}${suffix}.` },
+    };
+  }
+
+  const nextMode = normalizeFastMode(rawMode);
+  if (nextMode === undefined) {
+    return {
+      shouldContinue: false,
+      reply: { text: "⚙️ Usage: /fast status|on|off" },
+    };
+  }
+
+  if (params.sessionEntry && params.sessionStore && params.sessionKey) {
+    params.sessionEntry.fastMode = nextMode;
+    await persistSessionEntry(params);
+  }
+
+  return {
+    shouldContinue: false,
+    reply: { text: `⚙️ Fast mode ${nextMode ? "enabled" : "disabled"}.` },
+  };
+};
+
 export const handleSessionCommand: CommandHandler = async (params, allowTextCommands) => {
  if (!allowTextCommands) {
    return null;
--- a/src/auto-reply/reply/commands-status.ts
+++ b/src/auto-reply/reply/commands-status.ts
@ -3,6 +3,7 @@ import {
  resolveDefaultAgentId,
  resolveSessionAgentId,
 } from "../../agents/agent-scope.js";
+import { resolveFastModeState } from "../../agents/fast-mode.js";
 import { resolveModelAuthLabel } from "../../agents/model-auth-label.js";
 import { listSubagentRunsForRequester } from "../../agents/subagent-registry.js";
 import {
@ -40,6 +41,7 @@ export async function buildStatusReply(params: {
  model: string;
  contextTokens: number;
  resolvedThinkLevel?: ThinkLevel;
+  resolvedFastMode?: boolean;
  resolvedVerboseLevel: VerboseLevel;
  resolvedReasoningLevel: ReasoningLevel;
  resolvedElevatedLevel?: ElevatedLevel;
@ -60,6 +62,7 @@ export async function buildStatusReply(params: {
    model,
    contextTokens,
    resolvedThinkLevel,
+    resolvedFastMode,
    resolvedVerboseLevel,
    resolvedReasoningLevel,
    resolvedElevatedLevel,
@ -160,6 +163,14 @@ export async function buildStatusReply(params: {
      })
    : selectedModelAuth;
  const agentDefaults = cfg.agents?.defaults ?? {};
+  const effectiveFastMode =
+    resolvedFastMode ??
+    resolveFastModeState({
+      cfg,
+      provider,
+      model,
+      sessionEntry,
+    }).enabled;
  const statusText = buildStatusMessage({
    config: cfg,
    agent: {
@ -181,6 +192,7 @@ export async function buildStatusReply(params: {
    sessionStorePath: storePath,
    groupActivation,
    resolvedThink: resolvedThinkLevel ?? (await resolveDefaultThinkingLevel()),
+    resolvedFast: effectiveFastMode,
    resolvedVerbose: resolvedVerboseLevel,
    resolvedReasoning: resolvedReasoningLevel,
    resolvedElevated: resolvedElevatedLevel,
--- a/src/auto-reply/reply/directive-handling.fast-lane.ts
+++ b/src/auto-reply/reply/directive-handling.fast-lane.ts
@ -48,8 +48,13 @@ export async function applyInlineDirectivesFastLane(
  }

  const agentCfg = params.agentCfg;
-  const { currentThinkLevel, currentVerboseLevel, currentReasoningLevel, currentElevatedLevel } =
-    await resolveCurrentDirectiveLevels({
+  const {
+    currentThinkLevel,
+    currentFastMode,
+    currentVerboseLevel,
+    currentReasoningLevel,
+    currentElevatedLevel,
+  } = await resolveCurrentDirectiveLevels({
    sessionEntry,
    agentCfg,
    resolveDefaultThinkingLevel: () => modelState.resolveDefaultThinkingLevel(),
@ -77,6 +82,7 @@ export async function applyInlineDirectivesFastLane(
    initialModelLabel: params.initialModelLabel,
    formatModelSwitchEvent,
    currentThinkLevel,
+    currentFastMode,
    currentVerboseLevel,
    currentReasoningLevel,
    currentElevatedLevel,
--- a/src/auto-reply/reply/directive-handling.impl.ts
+++ b/src/auto-reply/reply/directive-handling.impl.ts
@ -3,6 +3,7 @@ import {
  resolveAgentDir,
  resolveSessionAgentId,
 } from "../../agents/agent-scope.js";
+import { resolveFastModeState } from "../../agents/fast-mode.js";
 import { resolveSandboxRuntimeStatus } from "../../agents/sandbox.js";
 import type { OpenClawConfig } from "../../config/config.js";
 import { type SessionEntry, updateSessionStore } from "../../config/sessions.js";
@ -78,6 +79,7 @@ export async function handleDirectiveOnly(
    initialModelLabel,
    formatModelSwitchEvent,
    currentThinkLevel,
+    currentFastMode,
    currentVerboseLevel,
    currentReasoningLevel,
    currentElevatedLevel,
@ -131,6 +133,15 @@ export async function handleDirectiveOnly(

  const resolvedProvider = modelSelection?.provider ?? provider;
  const resolvedModel = modelSelection?.model ?? model;
+  const fastModeState = resolveFastModeState({
+    cfg: params.cfg,
+    provider: resolvedProvider,
+    model: resolvedModel,
+    sessionEntry,
+  });
+  const effectiveFastMode = directives.fastMode ?? currentFastMode ?? fastModeState.enabled;
+  const effectiveFastModeSource =
+    directives.fastMode !== undefined ? "session" : fastModeState.source;

  if (directives.hasThinkDirective && !directives.thinkLevel) {
    // If no argument was provided, show the current level
@ -158,6 +169,25 @@ export async function handleDirectiveOnly(
      text: `Unrecognized verbose level "${directives.rawVerboseLevel}". Valid levels: off, on, full.`,
    };
  }
+  if (directives.hasFastDirective && directives.fastMode === undefined) {
+    if (!directives.rawFastMode) {
+      const sourceSuffix =
+        effectiveFastModeSource === "config"
+          ? " (config)"
+          : effectiveFastModeSource === "default"
+            ? " (default)"
+            : "";
+      return {
+        text: withOptions(
+          `Current fast mode: ${effectiveFastMode ? "on" : "off"}${sourceSuffix}.`,
+          "on, off",
+        ),
+      };
+    }
+    return {
+      text: `Unrecognized fast mode "${directives.rawFastMode}". Valid levels: on, off.`,
+    };
+  }
  if (directives.hasReasoningDirective && !directives.reasoningLevel) {
    if (!directives.rawReasoningLevel) {
      const level = currentReasoningLevel ?? "off";
@ -279,11 +309,18 @@ export async function handleDirectiveOnly(
    directives.elevatedLevel !== undefined &&
    elevatedEnabled &&
    elevatedAllowed;
+  const fastModeChanged =
+    directives.hasFastDirective &&
+    directives.fastMode !== undefined &&
+    directives.fastMode !== currentFastMode;
  let reasoningChanged =
    directives.hasReasoningDirective && directives.reasoningLevel !== undefined;
  if (directives.hasThinkDirective && directives.thinkLevel) {
    sessionEntry.thinkingLevel = directives.thinkLevel;
  }
+  if (directives.hasFastDirective && directives.fastMode !== undefined) {
+    sessionEntry.fastMode = directives.fastMode;
+  }
  if (shouldDowngradeXHigh) {
    sessionEntry.thinkingLevel = "high";
  }
@ -380,6 +417,13 @@ export async function handleDirectiveOnly(
        : `Thinking level set to ${directives.thinkLevel}.`,
    );
  }
+  if (directives.hasFastDirective && directives.fastMode !== undefined) {
+    parts.push(
+      directives.fastMode
+        ? formatDirectiveAck("Fast mode enabled.")
+        : formatDirectiveAck("Fast mode disabled."),
+    );
+  }
  if (directives.hasVerboseDirective && directives.verboseLevel) {
    parts.push(
      directives.verboseLevel === "off"
@ -459,6 +503,12 @@ export async function handleDirectiveOnly(
  if (directives.hasQueueDirective && directives.dropPolicy) {
    parts.push(formatDirectiveAck(`Queue drop set to ${directives.dropPolicy}.`));
  }
+  if (fastModeChanged) {
+    enqueueSystemEvent(`Fast mode ${sessionEntry.fastMode ? "enabled" : "disabled"}.`, {
+      sessionKey,
+      contextKey: `fast:${sessionEntry.fastMode ? "on" : "off"}`,
+    });
+  }
  const ack = parts.join(" ").trim();
  if (!ack && directives.hasStatusDirective) {
    return undefined;
--- a/src/auto-reply/reply/directive-handling.levels.ts
+++ b/src/auto-reply/reply/directive-handling.levels.ts
@ -3,6 +3,7 @@ import type { ElevatedLevel, ReasoningLevel, ThinkLevel, VerboseLevel } from "..
 export async function resolveCurrentDirectiveLevels(params: {
  sessionEntry?: {
    thinkingLevel?: unknown;
+    fastMode?: unknown;
    verboseLevel?: unknown;
    reasoningLevel?: unknown;
    elevatedLevel?: unknown;
@ -15,6 +16,7 @@ export async function resolveCurrentDirectiveLevels(params: {
  resolveDefaultThinkingLevel: () => Promise<ThinkLevel | undefined>;
 }): Promise<{
  currentThinkLevel: ThinkLevel | undefined;
+  currentFastMode: boolean | undefined;
  currentVerboseLevel: VerboseLevel | undefined;
  currentReasoningLevel: ReasoningLevel;
  currentElevatedLevel: ElevatedLevel | undefined;
@ -24,6 +26,8 @@ export async function resolveCurrentDirectiveLevels(params: {
    (await params.resolveDefaultThinkingLevel()) ??
    (params.agentCfg?.thinkingDefault as ThinkLevel | undefined);
  const currentThinkLevel = resolvedDefaultThinkLevel;
+  const currentFastMode =
+    typeof params.sessionEntry?.fastMode === "boolean" ? params.sessionEntry.fastMode : undefined;
  const currentVerboseLevel =
    (params.sessionEntry?.verboseLevel as VerboseLevel | undefined) ??
    (params.agentCfg?.verboseDefault as VerboseLevel | undefined);
@ -34,6 +38,7 @@ export async function resolveCurrentDirectiveLevels(params: {
    (params.agentCfg?.elevatedDefault as ElevatedLevel | undefined);
  return {
    currentThinkLevel,
+    currentFastMode,
    currentVerboseLevel,
    currentReasoningLevel,
    currentElevatedLevel,
--- a/src/auto-reply/reply/directive-handling.params.ts
+++ b/src/auto-reply/reply/directive-handling.params.ts
@ -32,6 +32,7 @@ export type HandleDirectiveOnlyCoreParams = {

 export type HandleDirectiveOnlyParams = HandleDirectiveOnlyCoreParams & {
  currentThinkLevel?: ThinkLevel;
+  currentFastMode?: boolean;
  currentVerboseLevel?: VerboseLevel;
  currentReasoningLevel?: ReasoningLevel;
  currentElevatedLevel?: ElevatedLevel;
--- a/src/auto-reply/reply/directive-handling.parse.ts
+++ b/src/auto-reply/reply/directive-handling.parse.ts
@ -6,6 +6,7 @@ import type { ElevatedLevel, ReasoningLevel, ThinkLevel, VerboseLevel } from "./
 import {
  extractElevatedDirective,
  extractExecDirective,
+  extractFastDirective,
  extractReasoningDirective,
  extractStatusDirective,
  extractThinkDirective,
@ -23,6 +24,9 @@ export type InlineDirectives = {
  hasVerboseDirective: boolean;
  verboseLevel?: VerboseLevel;
  rawVerboseLevel?: string;
+  hasFastDirective: boolean;
+  fastMode?: boolean;
+  rawFastMode?: string;
  hasReasoningDirective: boolean;
  reasoningLevel?: ReasoningLevel;
  rawReasoningLevel?: string;
@ -80,12 +84,18 @@ export function parseInlineDirectives(
    rawLevel: rawVerboseLevel,
    hasDirective: hasVerboseDirective,
  } = extractVerboseDirective(thinkCleaned);
+  const {
+    cleaned: fastCleaned,
+    fastMode,
+    rawLevel: rawFastMode,
+    hasDirective: hasFastDirective,
+  } = extractFastDirective(verboseCleaned);
  const {
    cleaned: reasoningCleaned,
    reasoningLevel,
    rawLevel: rawReasoningLevel,
    hasDirective: hasReasoningDirective,
-  } = extractReasoningDirective(verboseCleaned);
+  } = extractReasoningDirective(fastCleaned);
  const {
    cleaned: elevatedCleaned,
    elevatedLevel,
@ -151,6 +161,9 @@ export function parseInlineDirectives(
    hasVerboseDirective,
    verboseLevel,
    rawVerboseLevel,
+    hasFastDirective,
+    fastMode,
+    rawFastMode,
    hasReasoningDirective,
    reasoningLevel,
    rawReasoningLevel,
@ -201,6 +214,7 @@ export function isDirectiveOnly(params: {
  if (
    !directives.hasThinkDirective &&
    !directives.hasVerboseDirective &&
+    !directives.hasFastDirective &&
    !directives.hasReasoningDirective &&
    !directives.hasElevatedDirective &&
    !directives.hasExecDirective &&
--- a/src/auto-reply/reply/directives.ts
+++ b/src/auto-reply/reply/directives.ts
@ -2,6 +2,7 @@ import { escapeRegExp } from "../../utils.js";
 import type { NoticeLevel, ReasoningLevel } from "../thinking.js";
 import {
  type ElevatedLevel,
+  normalizeFastMode,
  normalizeElevatedLevel,
  normalizeNoticeLevel,
  normalizeReasoningLevel,
@ -124,6 +125,24 @@ export function extractVerboseDirective(body?: string): {
  };
 }

+export function extractFastDirective(body?: string): {
+  cleaned: string;
+  fastMode?: boolean;
+  rawLevel?: string;
+  hasDirective: boolean;
+} {
+  if (!body) {
+    return { cleaned: "", hasDirective: false };
+  }
+  const extracted = extractLevelDirective(body, ["fast"], normalizeFastMode);
+  return {
+    cleaned: extracted.cleaned,
+    fastMode: extracted.level,
+    rawLevel: extracted.rawLevel,
+    hasDirective: extracted.hasDirective,
+  };
+}
+
 export function extractNoticeDirective(body?: string): {
  cleaned: string;
  noticeLevel?: NoticeLevel;
--- a/src/auto-reply/reply/get-reply-directives-apply.ts
+++ b/src/auto-reply/reply/get-reply-directives-apply.ts
@ -150,6 +150,7 @@ export async function applyInlineDirectiveOverrides(params: {
    }
    const {
      currentThinkLevel: resolvedDefaultThinkLevel,
+      currentFastMode,
      currentVerboseLevel,
      currentReasoningLevel,
      currentElevatedLevel,
@ -162,6 +163,7 @@ export async function applyInlineDirectiveOverrides(params: {
    const directiveReply = await handleDirectiveOnly({
      ...createDirectiveHandlingBase(),
      currentThinkLevel,
+      currentFastMode,
      currentVerboseLevel,
      currentReasoningLevel,
      currentElevatedLevel,
@ -201,6 +203,7 @@ export async function applyInlineDirectiveOverrides(params: {

  const hasAnyDirective =
    directives.hasThinkDirective ||
+    directives.hasFastDirective ||
    directives.hasVerboseDirective ||
    directives.hasReasoningDirective ||
    directives.hasElevatedDirective ||
--- a/src/auto-reply/reply/get-reply-directives-utils.ts
+++ b/src/auto-reply/reply/get-reply-directives-utils.ts
@ -26,6 +26,9 @@ export function clearInlineDirectives(cleaned: string): InlineDirectives {
    hasVerboseDirective: false,
    verboseLevel: undefined,
    rawVerboseLevel: undefined,
+    hasFastDirective: false,
+    fastMode: undefined,
+    rawFastMode: undefined,
    hasReasoningDirective: false,
    reasoningLevel: undefined,
    rawReasoningLevel: undefined,
--- a/src/auto-reply/reply/get-reply-directives.ts
+++ b/src/auto-reply/reply/get-reply-directives.ts
@ -1,4 +1,5 @@
 import type { ExecToolDefaults } from "../../agents/bash-tools.js";
+import { resolveFastModeState } from "../../agents/fast-mode.js";
 import type { ModelAliasIndex } from "../../agents/model-selection.js";
 import { resolveSandboxRuntimeStatus } from "../../agents/sandbox.js";
 import type { SkillCommandSpec } from "../../agents/skills.js";
@ -37,6 +38,7 @@ export type ReplyDirectiveContinuation = {
  elevatedFailures: Array<{ gate: string; key: string }>;
  defaultActivation: ReturnType<typeof defaultGroupActivation>;
  resolvedThinkLevel: ThinkLevel | undefined;
+  resolvedFastMode: boolean;
  resolvedVerboseLevel: VerboseLevel | undefined;
  resolvedReasoningLevel: ReasoningLevel;
  resolvedElevatedLevel: ElevatedLevel;
@ -228,6 +230,7 @@ export async function resolveReplyDirectives(params: {
  const hasInlineDirective =
    parsedDirectives.hasThinkDirective ||
    parsedDirectives.hasVerboseDirective ||
+    parsedDirectives.hasFastDirective ||
    parsedDirectives.hasReasoningDirective ||
    parsedDirectives.hasElevatedDirective ||
    parsedDirectives.hasExecDirective ||
@ -260,6 +263,7 @@ export async function resolveReplyDirectives(params: {
        ...parsedDirectives,
        hasThinkDirective: false,
        hasVerboseDirective: false,
+        hasFastDirective: false,
        hasReasoningDirective: false,
        hasStatusDirective: false,
        hasModelDirective: false,
@ -340,6 +344,14 @@ export async function resolveReplyDirectives(params: {
  const defaultActivation = defaultGroupActivation(requireMention);
  const resolvedThinkLevel =
    directives.thinkLevel ?? (sessionEntry?.thinkingLevel as ThinkLevel | undefined);
+  const resolvedFastMode =
+    directives.fastMode ??
+    resolveFastModeState({
+      cfg,
+      provider,
+      model,
+      sessionEntry,
+    }).enabled;

  const resolvedVerboseLevel =
    directives.verboseLevel ??
@ -479,6 +491,7 @@ export async function resolveReplyDirectives(params: {
      elevatedFailures,
      defaultActivation,
      resolvedThinkLevel: resolvedThinkLevelWithDefault,
+      resolvedFastMode,
      resolvedVerboseLevel,
      resolvedReasoningLevel,
      resolvedElevatedLevel,
--- a/src/auto-reply/reply/get-reply-inline-actions.ts
+++ b/src/auto-reply/reply/get-reply-inline-actions.ts
@ -30,8 +30,13 @@ import type { createModelSelectionState } from "./model-selection.js";
 import { extractInlineSimpleCommand } from "./reply-inline.js";
 import type { TypingController } from "./typing.js";

-const builtinSlashCommands = (() => {
-  return listReservedChatSlashCommandNames([
+let builtinSlashCommands: Set<string> | null = null;
+
+function getBuiltinSlashCommands(): Set<string> {
+  if (builtinSlashCommands) {
+    return builtinSlashCommands;
+  }
+  builtinSlashCommands = listReservedChatSlashCommandNames([
    "think",
    "verbose",
    "reasoning",
@ -41,7 +46,8 @@ const builtinSlashCommands = (() => {
    "status",
    "queue",
  ]);
-})();
+  return builtinSlashCommands;
+}

 function resolveSlashCommandName(commandBodyNormalized: string): string | null {
  const trimmed = commandBodyNormalized.trim();
@ -163,7 +169,7 @@ export async function handleInlineActions(params: {
    allowTextCommands &&
    slashCommandName !== null &&
    // `/skill …` needs the full skill command list.
-    (slashCommandName === "skill" || !builtinSlashCommands.has(slashCommandName));
+    (slashCommandName === "skill" || !getBuiltinSlashCommands().has(slashCommandName));
  const skillCommands =
    shouldLoadSkillCommands && params.skillCommands
      ? params.skillCommands
--- a/src/auto-reply/reply/get-reply-run.ts
+++ b/src/auto-reply/reply/get-reply-run.ts
@ -1,6 +1,7 @@
 import crypto from "node:crypto";
 import { resolveSessionAuthProfileOverride } from "../../agents/auth-profiles/session-override.js";
 import type { ExecToolDefaults } from "../../agents/bash-tools.js";
+import { resolveFastModeState } from "../../agents/fast-mode.js";
 import {
  abortEmbeddedPiRun,
  isEmbeddedPiRunActive,
@ -509,6 +510,12 @@ export async function runPreparedReply(
      authProfileId,
      authProfileIdSource,
      thinkLevel: resolvedThinkLevel,
+      fastMode: resolveFastModeState({
+        cfg,
+        provider,
+        model,
+        sessionEntry,
+      }).enabled,
      verboseLevel: resolvedVerboseLevel,
      reasoningLevel: resolvedReasoningLevel,
      elevatedLevel: resolvedElevatedLevel,
--- a/src/auto-reply/status.test.ts
+++ b/src/auto-reply/status.test.ts
@ -113,6 +113,23 @@ describe("buildStatusMessage", () => {
    expect(normalized).toContain("Reasoning: on");
  });

+  it("shows fast mode when enabled", () => {
+    const text = buildStatusMessage({
+      agent: {
+        model: "openai/gpt-5.4",
+      },
+      sessionEntry: {
+        sessionId: "fast",
+        updatedAt: 0,
+        fastMode: true,
+      },
+      sessionKey: "agent:main:main",
+      queue: { mode: "collect", depth: 0 },
+    });
+
+    expect(normalizeTestText(text)).toContain("Fast: on");
+  });
+
  it("notes channel model overrides in status output", () => {
    const text = buildStatusMessage({
      config: {
@ -708,6 +725,10 @@ describe("buildHelpMessage", () => {
    expect(text).not.toContain("/config");
    expect(text).not.toContain("/debug");
  });
+
+  it("includes /fast in help output", () => {
+    expect(buildHelpMessage()).toContain("/fast on|off");
+  });
 });

 describe("buildCommandsMessagePaginated", () => {
--- a/src/auto-reply/status.ts
+++ b/src/auto-reply/status.ts
@ -77,6 +77,7 @@ type StatusArgs = {
  sessionStorePath?: string;
  groupActivation?: "mention" | "always";
  resolvedThink?: ThinkLevel;
+  resolvedFast?: boolean;
  resolvedVerbose?: VerboseLevel;
  resolvedReasoning?: ReasoningLevel;
  resolvedElevated?: ElevatedLevel;
@ -510,6 +511,7 @@ export function buildStatusMessage(args: StatusArgs): string {
    args.resolvedThink ?? args.sessionEntry?.thinkingLevel ?? args.agent?.thinkingDefault ?? "off";
  const verboseLevel =
    args.resolvedVerbose ?? args.sessionEntry?.verboseLevel ?? args.agent?.verboseDefault ?? "off";
+  const fastMode = args.resolvedFast ?? args.sessionEntry?.fastMode ?? false;
  const reasoningLevel = args.resolvedReasoning ?? args.sessionEntry?.reasoningLevel ?? "off";
  const elevatedLevel =
    args.resolvedElevated ??
@ -556,6 +558,7 @@ export function buildStatusMessage(args: StatusArgs): string {
  const optionParts = [
    `Runtime: ${runtime.label}`,
    `Think: ${thinkLevel}`,
+    fastMode ? "Fast: on" : null,
    verboseLabel,
    reasoningLevel !== "off" ? `Reasoning: ${reasoningLevel}` : null,
    elevatedLabel,
@ -728,7 +731,7 @@ export function buildHelpMessage(cfg?: OpenClawConfig): string {
  lines.push("  /new  |  /reset  |  /compact [instructions]  |  /stop");
  lines.push("");

-  const optionParts = ["/think <level>", "/model <id>", "/verbose on|off"];
+  const optionParts = ["/think <level>", "/model <id>", "/fast on|off", "/verbose on|off"];
  if (isCommandFlagEnabled(cfg, "config")) {
    optionParts.push("/config");
  }
--- a/src/auto-reply/thinking.ts
+++ b/src/auto-reply/thinking.ts
@ -218,6 +218,24 @@ export function resolveResponseUsageMode(raw?: string | null): UsageDisplayLevel
  return normalizeUsageDisplay(raw) ?? "off";
 }

+// Normalize fast-mode flags used to toggle low-latency model behavior.
+export function normalizeFastMode(raw?: string | boolean | null): boolean | undefined {
+  if (typeof raw === "boolean") {
+    return raw;
+  }
+  if (!raw) {
+    return undefined;
+  }
+  const key = raw.toLowerCase();
+  if (["off", "false", "no", "0", "disable", "disabled", "normal"].includes(key)) {
+    return false;
+  }
+  if (["on", "true", "yes", "1", "enable", "enabled", "fast"].includes(key)) {
+    return true;
+  }
+  return undefined;
+}
+
 // Normalize elevated flags used to toggle elevated bash permissions.
 export function normalizeElevatedLevel(raw?: string | null): ElevatedLevel | undefined {
  if (!raw) {
--- a/src/commands/agent/types.ts
+++ b/src/commands/agent/types.ts
@ -15,6 +15,8 @@ export type AgentStreamParams = {
  /** Provider stream params override (best-effort). */
  temperature?: number;
  maxTokens?: number;
+  /** Provider fast-mode override (best-effort). */
+  fastMode?: boolean;
 };

 export type AgentRunContext = {
--- a/src/commands/status.summary.ts
+++ b/src/commands/status.summary.ts
@ -36,6 +36,9 @@ const buildFlags = (entry?: SessionEntry): string[] => {
  if (typeof verbose === "string" && verbose.length > 0) {
    flags.push(`verbose:${verbose}`);
  }
+  if (typeof entry?.fastMode === "boolean") {
+    flags.push(entry.fastMode ? "fast" : "fast:off");
+  }
  const reasoning = entry?.reasoningLevel;
  if (typeof reasoning === "string" && reasoning.length > 0) {
    flags.push(`reasoning:${reasoning}`);
@ -170,6 +173,7 @@ export async function getStatusSummary(
          updatedAt,
          age,
          thinkingLevel: entry?.thinkingLevel,
+          fastMode: entry?.fastMode,
          verboseLevel: entry?.verboseLevel,
          reasoningLevel: entry?.reasoningLevel,
          elevatedLevel: entry?.elevatedLevel,
--- a/src/commands/status.types.ts
+++ b/src/commands/status.types.ts
@ -8,6 +8,7 @@ export type SessionStatus = {
  updatedAt: number | null;
  age: number | null;
  thinkingLevel?: string;
+  fastMode?: boolean;
  verboseLevel?: string;
  reasoningLevel?: string;
  elevatedLevel?: string;
--- a/src/config/sessions/types.ts
+++ b/src/config/sessions/types.ts
@ -100,6 +100,7 @@ export type SessionEntry = {
  abortCutoffTimestamp?: number;
  chatType?: SessionChatType;
  thinkingLevel?: string;
+  fastMode?: boolean;
  verboseLevel?: string;
  reasoningLevel?: string;
  elevatedLevel?: string;
--- a/src/gateway/protocol/schema/sessions.ts
+++ b/src/gateway/protocol/schema/sessions.ts
@ -52,6 +52,7 @@ export const SessionsPatchParamsSchema = Type.Object(
    key: NonEmptyString,
    label: Type.Optional(Type.Union([SessionLabelString, Type.Null()])),
    thinkingLevel: Type.Optional(Type.Union([NonEmptyString, Type.Null()])),
+    fastMode: Type.Optional(Type.Union([Type.Boolean(), Type.Null()])),
    verboseLevel: Type.Optional(Type.Union([NonEmptyString, Type.Null()])),
    reasoningLevel: Type.Optional(Type.Union([NonEmptyString, Type.Null()])),
    responseUsage: Type.Optional(
--- a/src/gateway/server-methods/agent.ts
+++ b/src/gateway/server-methods/agent.ts
@ -379,6 +379,7 @@ export const agentHandlers: GatewayRequestHandlers = {
        sessionId,
        updatedAt: now,
        thinkingLevel: entry?.thinkingLevel,
+        fastMode: entry?.fastMode,
        verboseLevel: entry?.verboseLevel,
        reasoningLevel: entry?.reasoningLevel,
        systemSent: entry?.systemSent,
--- a/src/gateway/server-methods/chat.ts
+++ b/src/gateway/server-methods/chat.ts
@ -980,6 +980,7 @@ export const chatHandlers: GatewayRequestHandlers = {
      sessionId,
      messages: bounded.messages,
      thinkingLevel,
+      fastMode: entry?.fastMode,
      verboseLevel,
    });
  },
--- a/src/gateway/server-node-events.ts
+++ b/src/gateway/server-node-events.ts
@ -166,6 +166,7 @@ async function touchSessionStore(params: {
      sessionId: params.sessionId,
      updatedAt: params.now,
      thinkingLevel: params.entry?.thinkingLevel,
+      fastMode: params.entry?.fastMode,
      verboseLevel: params.entry?.verboseLevel,
      reasoningLevel: params.entry?.reasoningLevel,
      systemSent: params.entry?.systemSent,
--- a/src/gateway/session-reset-service.ts
+++ b/src/gateway/session-reset-service.ts
@ -326,6 +326,7 @@ export async function performGatewaySessionReset(params: {
      systemSent: false,
      abortedLastRun: false,
      thinkingLevel: currentEntry?.thinkingLevel,
+      fastMode: currentEntry?.fastMode,
      verboseLevel: currentEntry?.verboseLevel,
      reasoningLevel: currentEntry?.reasoningLevel,
      responseUsage: currentEntry?.responseUsage,
--- a/src/gateway/session-utils.ts
+++ b/src/gateway/session-utils.ts
@ -929,6 +929,7 @@ export function listSessionsFromStore(params: {
        systemSent: entry?.systemSent,
        abortedLastRun: entry?.abortedLastRun,
        thinkingLevel: entry?.thinkingLevel,
+        fastMode: entry?.fastMode,
        verboseLevel: entry?.verboseLevel,
        reasoningLevel: entry?.reasoningLevel,
        elevatedLevel: entry?.elevatedLevel,
--- a/src/gateway/session-utils.types.ts
+++ b/src/gateway/session-utils.types.ts
@ -32,6 +32,7 @@ export type GatewaySessionRow = {
  systemSent?: boolean;
  abortedLastRun?: boolean;
  thinkingLevel?: string;
+  fastMode?: boolean;
  verboseLevel?: string;
  reasoningLevel?: string;
  elevatedLevel?: string;
--- a/src/gateway/sessions-patch.test.ts
+++ b/src/gateway/sessions-patch.test.ts
@ -149,6 +149,37 @@ describe("gateway sessions patch", () => {
    expect(entry.reasoningLevel).toBeUndefined();
  });

+  test("persists fastMode=false (does not clear)", async () => {
+    const entry = expectPatchOk(
+      await runPatch({
+        patch: { key: MAIN_SESSION_KEY, fastMode: false },
+      }),
+    );
+    expect(entry.fastMode).toBe(false);
+  });
+
+  test("persists fastMode=true", async () => {
+    const entry = expectPatchOk(
+      await runPatch({
+        patch: { key: MAIN_SESSION_KEY, fastMode: true },
+      }),
+    );
+    expect(entry.fastMode).toBe(true);
+  });
+
+  test("clears fastMode when patch sets null", async () => {
+    const store: Record<string, SessionEntry> = {
+      [MAIN_SESSION_KEY]: { fastMode: true } as SessionEntry,
+    };
+    const entry = expectPatchOk(
+      await runPatch({
+        store,
+        patch: { key: MAIN_SESSION_KEY, fastMode: null },
+      }),
+    );
+    expect(entry.fastMode).toBeUndefined();
+  });
+
  test("persists elevatedLevel=off (does not clear)", async () => {
    const entry = expectPatchOk(
      await runPatch({
--- a/src/gateway/sessions-patch.ts
+++ b/src/gateway/sessions-patch.ts
@ -11,6 +11,7 @@ import {
  formatThinkingLevels,
  formatXHighModelHint,
  normalizeElevatedLevel,
+  normalizeFastMode,
  normalizeReasoningLevel,
  normalizeThinkLevel,
  normalizeUsageDisplay,
@ -252,6 +253,19 @@ export async function applySessionsPatchToStore(params: {
    }
  }

+  if ("fastMode" in patch) {
+    const raw = patch.fastMode;
+    if (raw === null) {
+      delete next.fastMode;
+    } else if (raw !== undefined) {
+      const normalized = normalizeFastMode(raw);
+      if (normalized === undefined) {
+        return invalid("invalid fastMode (use true or false)");
+      }
+      next.fastMode = normalized;
+    }
+  }
+
  if ("verboseLevel" in patch) {
    const raw = patch.verboseLevel;
    const parsed = parseVerboseOverride(raw);
--- a/src/tui/commands.ts
+++ b/src/tui/commands.ts
@ -4,6 +4,7 @@ import { formatThinkingLevels, listThinkingLevelLabels } from "../auto-reply/thi
 import type { OpenClawConfig } from "../config/types.js";

 const VERBOSE_LEVELS = ["on", "off"];
+const FAST_LEVELS = ["status", "on", "off"];
 const REASONING_LEVELS = ["on", "off"];
 const ELEVATED_LEVELS = ["on", "off", "ask", "full"];
 const ACTIVATION_LEVELS = ["mention", "always"];
@ -52,6 +53,7 @@ export function parseCommand(input: string): ParsedCommand {
 export function getSlashCommands(options: SlashCommandOptions = {}): SlashCommand[] {
  const thinkLevels = listThinkingLevelLabels(options.provider, options.model);
  const verboseCompletions = createLevelCompletion(VERBOSE_LEVELS);
+  const fastCompletions = createLevelCompletion(FAST_LEVELS);
  const reasoningCompletions = createLevelCompletion(REASONING_LEVELS);
  const usageCompletions = createLevelCompletion(USAGE_FOOTER_LEVELS);
  const elevatedCompletions = createLevelCompletion(ELEVATED_LEVELS);
@ -76,6 +78,11 @@ export function getSlashCommands(options: SlashCommandOptions = {}): SlashComman
          .filter((v) => v.startsWith(prefix.toLowerCase()))
          .map((value) => ({ value, label: value })),
    },
+    {
+      name: "fast",
+      description: "Set fast mode on/off",
+      getArgumentCompletions: fastCompletions,
+    },
    {
      name: "verbose",
      description: "Set verbose on/off",
@ -142,6 +149,7 @@ export function helpText(options: SlashCommandOptions = {}): string {
    "/session <key> (or /sessions)",
    "/model <provider/model> (or /models)",
    `/think <${thinkLevels}>`,
+    "/fast <status|on|off>",
    "/verbose <on|off>",
    "/reasoning <on|off>",
    "/usage <off|tokens|full>",
--- a/src/tui/gateway-chat.ts
+++ b/src/tui/gateway-chat.ts
@ -79,6 +79,7 @@ export type GatewaySessionList = {
    Pick<
      SessionInfo,
      | "thinkingLevel"
+      | "fastMode"
      | "verboseLevel"
      | "reasoningLevel"
      | "model"
@ -92,6 +93,7 @@ export type GatewaySessionList = {
      key: string;
      sessionId?: string;
      updatedAt?: number | null;
+      fastMode?: boolean;
      sendPolicy?: string;
      responseUsage?: ResponseUsageMode;
      label?: string;
--- a/src/tui/tui-command-handlers.ts
+++ b/src/tui/tui-command-handlers.ts
@ -345,6 +345,27 @@ export function createCommandHandlers(context: CommandHandlerContext) {
          chatLog.addSystem(`verbose failed: ${String(err)}`);
        }
        break;
+      case "fast":
+        if (!args || args === "status") {
+          chatLog.addSystem(`fast mode: ${state.sessionInfo.fastMode ? "on" : "off"}`);
+          break;
+        }
+        if (args !== "on" && args !== "off") {
+          chatLog.addSystem("usage: /fast <status|on|off>");
+          break;
+        }
+        try {
+          const result = await client.patchSession({
+            key: state.currentSessionKey,
+            fastMode: args === "on",
+          });
+          chatLog.addSystem(`fast mode ${args === "on" ? "enabled" : "disabled"}`);
+          applySessionInfoFromPatch(result);
+          await refreshSessionInfo();
+        } catch (err) {
+          chatLog.addSystem(`fast failed: ${String(err)}`);
+        }
+        break;
      case "reasoning":
        if (!args) {
          chatLog.addSystem("usage: /reasoning <on|off>");
--- a/src/tui/tui-session-actions.ts
+++ b/src/tui/tui-session-actions.ts
@ -165,6 +165,9 @@ export function createSessionActions(context: SessionActionContext) {
    if (entry?.thinkingLevel !== undefined) {
      next.thinkingLevel = entry.thinkingLevel;
    }
+    if (entry?.fastMode !== undefined) {
+      next.fastMode = entry.fastMode;
+    }
    if (entry?.verboseLevel !== undefined) {
      next.verboseLevel = entry.verboseLevel;
    }
@ -286,10 +289,12 @@ export function createSessionActions(context: SessionActionContext) {
        messages?: unknown[];
        sessionId?: string;
        thinkingLevel?: string;
+        fastMode?: boolean;
        verboseLevel?: string;
      };
      state.currentSessionId = typeof record.sessionId === "string" ? record.sessionId : null;
      state.sessionInfo.thinkingLevel = record.thinkingLevel ?? state.sessionInfo.thinkingLevel;
+      state.sessionInfo.fastMode = record.fastMode ?? state.sessionInfo.fastMode;
      state.sessionInfo.verboseLevel = record.verboseLevel ?? state.sessionInfo.verboseLevel;
      const showTools = (state.sessionInfo.verboseLevel ?? "off") !== "off";
      chatLog.clearAll();
--- a/src/tui/tui-types.ts
+++ b/src/tui/tui-types.ts
@ -28,6 +28,7 @@ export type ResponseUsageMode = "on" | "off" | "tokens" | "full";

 export type SessionInfo = {
  thinkingLevel?: string;
+  fastMode?: boolean;
  verboseLevel?: string;
  reasoningLevel?: string;
  model?: string;
--- a/src/tui/tui.ts
+++ b/src/tui/tui.ts
@ -752,6 +752,7 @@ export async function runTui(opts: TuiOptions) {
      : "unknown";
    const tokens = formatTokens(sessionInfo.totalTokens ?? null, sessionInfo.contextTokens ?? null);
    const think = sessionInfo.thinkingLevel ?? "off";
+    const fast = sessionInfo.fastMode === true;
    const verbose = sessionInfo.verboseLevel ?? "off";
    const reasoning = sessionInfo.reasoningLevel ?? "off";
    const reasoningLabel =
@ -761,6 +762,7 @@ export async function runTui(opts: TuiOptions) {
      `session ${sessionLabel}`,
      modelLabel,
      think !== "off" ? `think ${think}` : null,
+      fast ? "fast" : null,
      verbose !== "off" ? `verbose ${verbose}` : null,
      reasoningLabel,
      tokens,
--- a/ui/src/ui/chat/slash-command-executor.node.test.ts
+++ b/ui/src/ui/chat/slash-command-executor.node.test.ts
@ -378,4 +378,42 @@ describe("executeSlashCommand directives", () => {
    expect(result.content).toBe("Current verbose level: full.\nOptions: on, full, off.");
    expect(request).toHaveBeenNthCalledWith(1, "sessions.list", {});
  });
+
+  it("reports the current fast mode for bare /fast", async () => {
+    const request = vi.fn(async (method: string, _payload?: unknown) => {
+      if (method === "sessions.list") {
+        return {
+          sessions: [row("agent:main:main", { fastMode: true })],
+        };
+      }
+      throw new Error(`unexpected method: ${method}`);
+    });
+
+    const result = await executeSlashCommand(
+      { request } as unknown as GatewayBrowserClient,
+      "agent:main:main",
+      "fast",
+      "",
+    );
+
+    expect(result.content).toBe("Current fast mode: on.\nOptions: status, on, off.");
+    expect(request).toHaveBeenNthCalledWith(1, "sessions.list", {});
+  });
+
+  it("patches fast mode for /fast on", async () => {
+    const request = vi.fn().mockResolvedValue({ ok: true });
+
+    const result = await executeSlashCommand(
+      { request } as unknown as GatewayBrowserClient,
+      "agent:main:main",
+      "fast",
+      "on",
+    );
+
+    expect(result.content).toBe("Fast mode enabled.");
+    expect(request).toHaveBeenCalledWith("sessions.patch", {
+      key: "agent:main:main",
+      fastMode: true,
+    });
+  });
 });
--- a/ui/src/ui/chat/slash-command-executor.ts
+++ b/ui/src/ui/chat/slash-command-executor.ts
@ -63,6 +63,8 @@ export async function executeSlashCommand(
      return await executeModel(client, sessionKey, args);
    case "think":
      return await executeThink(client, sessionKey, args);
+    case "fast":
+      return await executeFast(client, sessionKey, args);
    case "verbose":
      return await executeVerbose(client, sessionKey, args);
    case "export":
@ -252,6 +254,44 @@ async function executeVerbose(
  }
 }

+async function executeFast(
+  client: GatewayBrowserClient,
+  sessionKey: string,
+  args: string,
+): Promise<SlashCommandResult> {
+  const rawMode = args.trim().toLowerCase();
+
+  if (!rawMode || rawMode === "status") {
+    try {
+      const session = await loadCurrentSession(client, sessionKey);
+      return {
+        content: formatDirectiveOptions(
+          `Current fast mode: ${resolveCurrentFastMode(session)}.`,
+          "status, on, off",
+        ),
+      };
+    } catch (err) {
+      return { content: `Failed to get fast mode: ${String(err)}` };
+    }
+  }
+
+  if (rawMode !== "on" && rawMode !== "off") {
+    return {
+      content: `Unrecognized fast mode "${args.trim()}". Valid levels: status, on, off.`,
+    };
+  }
+
+  try {
+    await client.request("sessions.patch", { key: sessionKey, fastMode: rawMode === "on" });
+    return {
+      content: `Fast mode ${rawMode === "on" ? "enabled" : "disabled"}.`,
+      action: "refresh",
+    };
+  } catch (err) {
+    return { content: `Failed to set fast mode: ${String(err)}` };
+  }
+}
+
 async function executeUsage(
  client: GatewayBrowserClient,
  sessionKey: string,
@ -534,6 +574,10 @@ function resolveCurrentThinkingLevel(
  });
 }

+function resolveCurrentFastMode(session: GatewaySessionRow | undefined): "on" | "off" {
+  return session?.fastMode === true ? "on" : "off";
+}
+
 function fmtTokens(n: number): string {
  if (n >= 1_000_000) {
    return `${(n / 1_000_000).toFixed(1).replace(/\.0$/, "")}M`;
--- a/ui/src/ui/chat/slash-commands.node.test.ts
+++ b/ui/src/ui/chat/slash-commands.node.test.ts
@ -23,4 +23,11 @@ describe("parseSlashCommand", () => {
      args: "full",
    });
  });
+
+  it("parses fast commands", () => {
+    expect(parseSlashCommand("/fast:on")).toMatchObject({
+      command: { name: "fast" },
+      args: "on",
+    });
+  });
 });
--- a/ui/src/ui/chat/slash-commands.ts
+++ b/ui/src/ui/chat/slash-commands.ts
@ -88,6 +88,15 @@ export const SLASH_COMMANDS: SlashCommandDef[] = [
    executeLocal: true,
    argOptions: ["on", "off", "full"],
  },
+  {
+    name: "fast",
+    description: "Toggle fast mode",
+    args: "<status|on|off>",
+    icon: "zap",
+    category: "model",
+    executeLocal: true,
+    argOptions: ["status", "on", "off"],
+  },

  // ── Tools ──
  {
--- a/ui/src/ui/controllers/sessions.ts
+++ b/ui/src/ui/controllers/sessions.ts
@ -63,6 +63,7 @@ export async function patchSession(
  patch: {
    label?: string | null;
    thinkingLevel?: string | null;
+    fastMode?: boolean | null;
    verboseLevel?: string | null;
    reasoningLevel?: string | null;
  },
@ -77,6 +78,9 @@ export async function patchSession(
  if ("thinkingLevel" in patch) {
    params.thinkingLevel = patch.thinkingLevel;
  }
+  if ("fastMode" in patch) {
+    params.fastMode = patch.fastMode;
+  }
  if ("verboseLevel" in patch) {
    params.verboseLevel = patch.verboseLevel;
  }
--- a/ui/src/ui/types.ts
+++ b/ui/src/ui/types.ts
@ -379,6 +379,7 @@ export type GatewaySessionRow = {
  systemSent?: boolean;
  abortedLastRun?: boolean;
  thinkingLevel?: string;
+  fastMode?: boolean;
  verboseLevel?: string;
  reasoningLevel?: string;
  elevatedLevel?: string;
@ -396,6 +397,7 @@ export type SessionsPatchResult = SessionsPatchResultBase<{
  sessionId: string;
  updatedAt?: number;
  thinkingLevel?: string;
+  fastMode?: boolean;
  verboseLevel?: string;
  reasoningLevel?: string;
  elevatedLevel?: string;
--- a/ui/src/ui/views/sessions.test.ts
+++ b/ui/src/ui/views/sessions.test.ts
@ -60,7 +60,7 @@ describe("sessions view", () => {
    await Promise.resolve();

    const selects = container.querySelectorAll("select");
-    const verbose = selects[1] as HTMLSelectElement | undefined;
+    const verbose = selects[2] as HTMLSelectElement | undefined;
    expect(verbose?.value).toBe("full");
    expect(Array.from(verbose?.options ?? []).some((option) => option.value === "full")).toBe(true);
  });
@ -83,10 +83,32 @@ describe("sessions view", () => {
    await Promise.resolve();

    const selects = container.querySelectorAll("select");
-    const reasoning = selects[2] as HTMLSelectElement | undefined;
+    const reasoning = selects[3] as HTMLSelectElement | undefined;
    expect(reasoning?.value).toBe("custom-mode");
    expect(
      Array.from(reasoning?.options ?? []).some((option) => option.value === "custom-mode"),
    ).toBe(true);
  });
+
+  it("renders explicit fast mode without falling back to inherit", async () => {
+    const container = document.createElement("div");
+    render(
+      renderSessions(
+        buildProps(
+          buildResult({
+            key: "agent:main:main",
+            kind: "direct",
+            updatedAt: Date.now(),
+            fastMode: true,
+          }),
+        ),
+      ),
+      container,
+    );
+    await Promise.resolve();
+
+    const selects = container.querySelectorAll("select");
+    const fast = selects[1] as HTMLSelectElement | undefined;
+    expect(fast?.value).toBe("on");
+  });
 });
--- a/ui/src/ui/views/sessions.ts
+++ b/ui/src/ui/views/sessions.ts
@ -37,6 +37,7 @@ export type SessionsProps = {
    patch: {
      label?: string | null;
      thinkingLevel?: string | null;
+      fastMode?: boolean | null;
      verboseLevel?: string | null;
      reasoningLevel?: string | null;
    },
@ -52,6 +53,11 @@ const VERBOSE_LEVELS = [
  { value: "on", label: "on" },
  { value: "full", label: "full" },
 ] as const;
+const FAST_LEVELS = [
+  { value: "", label: "inherit" },
+  { value: "on", label: "on" },
+  { value: "off", label: "off" },
+] as const;
 const REASONING_LEVELS = ["", "off", "on", "stream"] as const;
 const PAGE_SIZES = [10, 25, 50, 100] as const;

@ -306,6 +312,7 @@ export function renderSessions(props: SessionsProps) {
                ${sortHeader("updated", "Updated")}
                ${sortHeader("tokens", "Tokens")}
                <th>Thinking</th>
+                <th>Fast</th>
                <th>Verbose</th>
                <th>Reasoning</th>
                <th style="width: 60px;"></th>
@ -316,7 +323,7 @@ export function renderSessions(props: SessionsProps) {
                paginated.length === 0
                  ? html`
                      <tr>
-                        <td colspan="9" style="text-align: center; padding: 48px 16px; color: var(--muted)">
+                        <td colspan="10" style="text-align: center; padding: 48px 16px; color: var(--muted)">
                          No sessions found.
                        </td>
                      </tr>
@ -390,6 +397,8 @@ function renderRow(
  const isBinaryThinking = isBinaryThinkingProvider(row.modelProvider);
  const thinking = resolveThinkLevelDisplay(rawThinking, isBinaryThinking);
  const thinkLevels = withCurrentOption(resolveThinkLevelOptions(row.modelProvider), thinking);
+  const fastMode = row.fastMode === true ? "on" : row.fastMode === false ? "off" : "";
+  const fastLevels = withCurrentLabeledOption(FAST_LEVELS, fastMode);
  const verbose = row.verboseLevel ?? "";
  const verboseLevels = withCurrentLabeledOption(VERBOSE_LEVELS, verbose);
  const reasoning = row.reasoningLevel ?? "";
@ -465,6 +474,23 @@ function renderRow(
          )}
        </select>
      </td>
+      <td>
+        <select
+          ?disabled=${disabled}
+          style="padding: 6px 10px; font-size: 13px; border: 1px solid var(--border); border-radius: var(--radius-sm); min-width: 90px;"
+          @change=${(e: Event) => {
+            const value = (e.target as HTMLSelectElement).value;
+            onPatch(row.key, { fastMode: value === "" ? null : value === "on" });
+          }}
+        >
+          ${fastLevels.map(
+            (level) =>
+              html`<option value=${level.value} ?selected=${fastMode === level.value}>
+                ${level.label}
+              </option>`,
+          )}
+        </select>
+      </td>
      <td>
        <select
          ?disabled=${disabled}