openclaw/docs/gateway/config-tools.md

summary, read_when, title, sidebarTitle

summary	read_when	title	sidebarTitle
Tools config (policy, experimental toggles, provider-backed tools) and custom provider/base-URL setup	Configuring `tools.*` policy, allowlists, or experimental features Registering custom providers or overriding base URLs Setting up OpenAI-compatible self-hosted endpoints	Configuration — tools and custom providers	Tools and custom providers

tools.* config keys and custom provider / base-URL setup. For agents, channels, and other top-level config keys, see Configuration reference.

Tools

Tool profiles

tools.profile sets a base allowlist before tools.allow/tools.deny:

Local onboarding defaults new local configs to `tools.profile: "coding"` when unset (existing explicit profiles are preserved).

Profile	Includes
minimal	session_status only
coding	group:fs, group:runtime, group:web, group:sessions, group:memory, cron, get_goal, create_goal, update_goal, progress_card, ask_user, skill_workshop, image, image_generate, music_generate, video_generate
messaging	group:messaging, sessions, sessions_list, sessions_history, sessions_search, conversations_list, conversations_send, conversations_turn, sessions_send, sessions_spawn, sessions_yield, subagents, session_status, ask_user
full	No restriction (same as unset)

coding and messaging also implicitly allow bundle-mcp (configured MCP servers).

Tool groups

Group	Tools
group:runtime	exec, process, code_execution (bash is accepted as an alias for exec)
group:fs	read, write, edit, apply_patch
group:sessions	sessions, sessions_list, sessions_history, sessions_search, conversations_list, conversations_send, conversations_turn, sessions_send, sessions_spawn, sessions_yield, subagents, session_status, suggest_task, dismiss_task
group:memory	memory_search, memory_get
group:web	web_search, x_search, web_fetch
group:ui	browser, screen, dashboard, terminal, portal, canvas, show_widget
group:automation	heartbeat_respond, cron, gateway
group:messaging	message
group:nodes	nodes, computer
group:agents	agents_list, get_goal, create_goal, update_goal, progress_card, ask_user, skill_workshop
group:media	image, image_generate, music_generate, video_generate, tts
group:openclaw	All built-in tools above except read/write/edit/apply_patch/exec/process/canvas (excludes plugin tools)
group:plugins	Tools owned by loaded plugins, including configured MCP servers exposed through bundle-mcp

suggest_task lets a coding agent propose confirmed follow-up work without starting it. The suggestion's project directory must be a git checkout; invalid suggestions, including a non-git directory or blank prompt, are rejected when the tool records them. The Control UI shows the title and summary as an actionable chip; a Gateway-backed TUI shows an equivalent interactive prompt. Accepting a suggestion can start it in a fresh managed worktree (the default), start it locally in a new session in the suggested checkout, send it to a cloud worker profile when one is configured, or deliver it into the source session. OpenClaw sends the full prompt to the selected destination while the current turn continues. dismiss_task withdraws a still-pending suggestion by the ephemeral task_id returned from suggest_task.

The tools are offered only when the initiating operator surface can receive and action Gateway task-suggestion events. Channel sessions and local/embedded TUI sessions do not receive them; channel transports need a portable typed task action before they can safely expose this flow. Suggestions are process-local and disappear when the Gateway restarts. Both tools remain in the coding profile and group:sessions, so normal tools.allow and tools.deny policy configures them automatically when the surface supports them.

MCP and plugin tools inside sandbox tool policy

Configured MCP servers are exposed as plugin-owned tools under the bundle-mcp plugin id. Normal tool profiles can allow them, but tools.sandbox.tools is an additional gate for sandboxed sessions. If sandbox mode is "all" or "non-main", include one of these entries in the sandbox tool allowlist when MCP/plugin tools should be visible:

• bundle-mcp for OpenClaw-managed MCP servers from mcp.servers
• the plugin id for a specific native plugin
• group:plugins for all loaded plugin-owned tools
• exact MCP server tool names or server globs such as outlook__send_mail or outlook__* when you only want one server

Server globs use the provider-safe MCP server prefix, not necessarily the raw mcp.servers key. Non-[A-Za-z0-9_-] characters become -, names that do not start with a letter get an mcp- prefix, and long or duplicate prefixes may be truncated or suffixed; for example, mcp.servers["Outlook Graph"] uses a glob like outlook-graph__*.

{
  agents: { defaults: { sandbox: { mode: "all" } } },
  mcp: {
    servers: {
      outlook: { command: "node", args: ["./outlook-mcp.js"] },
    },
  },
  tools: {
    sandbox: {
      tools: {
        alsoAllow: ["web_search", "web_fetch", "memory_search", "memory_get", "bundle-mcp"],
      },
    },
  },
}

Without that sandbox-layer entry, the MCP server can still load successfully while its tools are filtered before the provider request. Use openclaw doctor to catch this shape for OpenClaw-managed servers in mcp.servers. MCP servers loaded from bundled plugin manifests or Claude .mcp.json use the same sandbox gate, but this diagnostic does not enumerate those sources yet; use the same allowlist entries if their tools disappear in sandboxed turns.

tools.codeMode

tools.codeMode gates the generic OpenClaw code-mode surface. When engaged for a run with tools, normal OpenClaw tools move behind the in-sandbox tools.* catalog bridge, and MCP tools are available through the generated MCP namespace. The model normally sees exec and wait; tools such as computer whose structured results cannot cross the JSON-only bridge stay direct.

enabled defaults to "auto", which engages code mode only for models whose catalog entry flags compat.codeMode: "preferred". See Code Mode - automatic per-model activation.

To opt out for every run:

{
  tools: {
    codeMode: {
      enabled: false,
    },
  },
}

The shorthand is also accepted:

{
  tools: { codeMode: false },
}

enabled: true forces code mode on for every tool-capable run, regardless of model.

MCP declarations are exposed through the read-only virtual API file surface in code mode. Guest code can call API.list("mcp") and API.read("mcp/<server>.d.ts") to inspect TypeScript-style signatures before calling MCP.<server>.<tool>(). See Code Mode for the runtime contract, limits, and debugging steps.

tools.allow / tools.deny

Global tool allow/deny policy (deny wins). Case-insensitive, supports * wildcards. Applied even when Docker sandbox is off.

{
  tools: { deny: ["browser", "canvas"] },
}

write and apply_patch are separate tool ids. allow: ["write"] also enables apply_patch for compatible models, but deny: ["write"] does not deny apply_patch. To block all file mutation, deny group:fs or list each mutating tool explicitly:

{
  tools: { deny: ["write", "edit", "apply_patch"] },
}

`allow` and `alsoAllow` cannot both be set in the same scope (`tools`, `tools.byProvider.`, `agents.entries.*.tools`) — config validation rejects it. Merge `alsoAllow` entries into `allow`, or drop `allow` and use `profile` + `alsoAllow` instead.

The image inspection tool is view_image. If an older config still names image in an allow, alsoAllow, or deny list, run openclaw doctor --fix to rewrite supported global, per-agent, provider, sandbox, sender, channel, and Gateway policy surfaces. Doctor preserves patterns such as image* that may still match other tools and adds view_image when the pattern no longer covers inspection. Patterns that already cover both names, such as * or *image*, remain unchanged.

tools.byProvider

Further restrict tools for specific providers or models. Order: base profile → provider profile → allow/deny.

{
  tools: {
    profile: "coding",
    byProvider: {
      anthropic: { profile: "minimal" },
      "openai/gpt-5.4": { allow: ["group:fs", "sessions_list"] },
    },
  },
}

tools.toolsBySender

Restricts tools for the current turn's originating requester. This is defense-in-depth on top of channel access control; sender values must come from the channel adapter, not message text. It does not authenticate other content in the model prompt; see Requester-scoped controls and prompt context.

{
  tools: {
    toolsBySender: {
      "channel:discord:1234567890123": { alsoAllow: ["group:fs"] },
      "id:guest-user-id": { deny: ["group:runtime", "group:fs"] },
      "*": { deny: ["exec", "process", "write", "edit", "apply_patch"] },
    },
  },
}

Keys use explicit prefixes: channel:<channelId>:<senderId>, id:<senderId>, e164:<phone>, username:<handle>, name:<displayName>, or "*". Channel ids are canonical OpenClaw ids; aliases such as teams normalize to msteams. Legacy unprefixed keys are accepted as id: only. Matching order is channel+id, id, e164, username, name, then wildcard.

Per-agent agents.entries.*.tools.toolsBySender overrides the global sender match when it matches, even with an empty {} policy.

tools.elevated

Controls elevated exec access outside the sandbox:

{
  tools: {
    elevated: {
      enabled: true,
      allowFrom: {
        whatsapp: ["+15555550123"],
        discord: ["1234567890123", "987654321098765432"],
      },
    },
  },
}

• Per-agent override (agents.entries.*.tools.elevated) can only further restrict.
• /elevated on|off|ask|full stores state per session; inline directives apply to single message.
• Elevated exec bypasses sandboxing and uses the configured escape path (gateway by default, or node when the exec target is node).

tools.github

GitHub CLI identity is native by default. When tools.github is omitted, local agent tools, the Codex harness, and Agent Settings follow normal gh resolution: GH_TOKEN or GITHUB_TOKEN from the Gateway process takes precedence, followed by the runtime user's gh keyring/config. The Git author comes from the selected agent's workspace.

Use Agents → Tools → GitHub Identity to configure a managed fine-grained personal access token. The browser places the pasted token in the secret store as a one-use handoff. The Gateway hard-deletes that handoff before passing its value to gh auth login on stdin, verifies the account, publishes an account-owned managed gh profile, and stores only this secret-free config:

{
  tools: {
    github: {
      profileId: "ghp_0123456789abcdef0123456789abcdef",
      gitAuthor: { name: "Automation User", email: "automation@example.com" },
    },
  },
  agents: {
    entries: {
      reviewer: {
        tools: {
          github: {
            profileId: "ghp_fedcba9876543210fedcba9876543210",
            gitAuthor: { name: "Review Agent" },
          },
        },
      },
    },
  },
}

Omitting agents.entries.<id>.tools.github inherits the system identity. An agent object is a complete managed override. If a configured managed profile is missing or unusable, GitHub status reports configured_unavailable; it never falls back to the native profile.

Managed identity applies to the gh CLI/API account and optional Git author/committer metadata in local OpenClaw exec and the local Codex harness. OpenClaw supplies a private GH_CONFIG_DIR, clears ambient GH_TOKEN and GITHUB_TOKEN precedence, and applies configured author fields through process-local environment and Git config overlays. It does not install a credential helper, rewrite SSH remotes, add HTTP authorization headers, or otherwise override an existing repository's Git network credentials. New runs switch to a replacement or inherited identity immediately. Existing local processes keep their immutable profile generation until they close; retired profile files are cleaned on the next Gateway restart, so changing this setting is not immediate credential revocation.

Managed profiles provide execution and coordination identity; they are not an OS-user security sandbox. A process with unrestricted host execution under the same OS account can access account-owned files, including managed gh profiles. Use an OpenClaw sandbox, a dedicated host, or a dedicated OS user when adversarial isolation is required.

The profile is not forwarded to node hosts, OpenClaw sandboxes, remote-exec placements, or cloud workers; cloud workers remain credential-free. Direct branch or pull request publication is not part of this identity foundation and belongs to PR3 in this stacked series. Until that separate Gateway broker lands, repository remotes continue to use their existing credentials and publication workflow.

Verification proves which account answered the GitHub API request. Status distinguishes missing or invalid managed credentials, unverified transport failures, and GitHub rate limiting without returning gh diagnostics. It does not claim that fine-grained write permissions or Git transport access were remotely verified. GitHub App and brokered publication support remain future work.

Control UI repository previews and project discovery use the separate optional gateway.controlUi.github.token service credential. They never consume an agent tool identity. When this SecretRef is explicit, OpenClaw excludes its exact environment or store name from agent execution. A custom name does not clear unrelated GH_TOKEN or GITHUB_TOKEN values used by native identity; a ref named GH_TOKEN or GITHUB_TOKEN excludes that exact variable.

tools.exec

{
  tools: {
    exec: {
      backgroundMs: 10000,
      timeoutSeconds: 1800,
      cleanupMs: 1800000,
      approvalRunningNoticeMs: 10000,
      notifyOnExit: true,
      notifyOnExitEmptySuccess: false,
      commandHighlighting: false,
      applyPatch: {
        enabled: true,
        allowModels: ["gpt-5.6-sol"],
      },
    },
  },
}

Values shown are defaults except applyPatch.allowModels (empty/unset by default, meaning any compatible model may use apply_patch). approvalRunningNoticeMs emits a running notice when approval-backed exec runs long; 0 disables it.

tools.loopDetection

Tool-loop safety checks are disabled by default. Set enabled: true to activate detection. Settings can be defined globally in tools.loopDetection and overridden per-agent at agents.entries.*.tools.loopDetection.

{
  tools: {
    loopDetection: {
      enabled: true,
    },
  },
}

tools.web

{
  tools: {
    web: {
      search: {
        enabled: true,
        apiKey: "brave_api_key", // or BRAVE_API_KEY env (Brave provider)
        maxResults: 5,
        timeoutSeconds: 30,
        cacheTtlMinutes: 15,
      },
      fetch: {
        enabled: true,
        provider: "firecrawl", // optional; omit for auto-detect
        maxChars: 20000,
        maxCharsCap: 20000,
        maxResponseBytes: 750000,
        timeoutSeconds: 30,
        cacheTtlMinutes: 15,
        maxRedirects: 3,
        readability: true,
        userAgent: "custom-ua",
      },
    },
  },
}

Values shown are defaults except provider and userAgent. maxResponseBytes clamps to 32000–10000000; maxChars clamps to maxCharsCap (raise maxCharsCap to allow larger responses).

tools.media

Configures inbound media understanding (image/audio/video):

{
  tools: {
    media: {
      concurrency: 2,
      models: [
        { provider: "openai", model: "gpt-4o-mini-transcribe", capabilities: ["audio"] },
        {
          type: "cli",
          command: "whisper",
          args: ["--model", "base", "{{AttachmentPath}}"],
          capabilities: ["audio"],
        },
        { provider: "ollama", model: "gemma4:26b", capabilities: ["image"] },
        { provider: "google", model: "gemini-3-flash-preview", capabilities: ["video"] },
      ],
      audio: { enabled: true, preferredModel: "openai/gpt-4o-mini-transcribe" },
      image: { enabled: true, preferredModel: "ollama/gemma4:26b" },
      video: { enabled: true },
    },
  },
}

tools.media.models is the only configured model list. Every entry declares the capabilities it handles. The optional preferredModel selector accepts provider/model, a model id, provider:<id> for provider-default entries, or cli:command; matching entries move to the front of that capability's fallback order. Per-capability prompts, limits, request settings, scope, attachment policy, and audio transcript echo remain defaults for configured and auto-detected models; a model entry can override model-specific fields.

**Provider entry** (`type: "provider"` or omitted):

- `provider`: API provider id (`openai`, `anthropic`, `google`/`gemini`, `groq`, etc.)
- `model`: model id override
- `profile` / `preferredProfile`: `auth-profiles.json` profile selection

**CLI entry** (`type: "cli"`):

- `command`: executable to run
- `args`: templated args (supports `{{AttachmentPath}}`, `{{AttachmentUrl}}`, `{{AttachmentContentType}}`, `{{AttachmentDir}}`, `{{AttachmentIndex}}`, `{{Prompt}}`, `{{MaxChars}}`, etc.; `openclaw doctor --fix` migrates deprecated `{input}` placeholders to `{{AttachmentPath}}`). The older `{{MediaPath}}`, `{{MediaUrl}}`, `{{MediaType}}`, and `{{MediaDir}}` aliases remain available during their compatibility window but are deprecated.

**Common fields:**

- `capabilities`: list containing one or more of `image`, `audio`, and `video`.
- `prompt`, `maxChars`, `maxBytes`, `timeoutSeconds`, `language`: per-entry overrides.
- Matching image model `timeoutSeconds` entries also apply when the agent calls the explicit `view_image` tool. For image understanding, this timeout applies to the request itself and is not reduced by earlier preparation work.
- Failures fall back to the next entry.

Provider auth follows standard order: `auth-profiles.json` → env vars → `models.providers.*.apiKey`.

tools.agentToAgent

{
  tools: {
    agentToAgent: {
      enabled: false,
      allow: ["home", "work"],
    },
  },
}

tools.sessions

Controls which sessions can be targeted by the session tools (sessions_list, sessions_history, sessions_send).

Default: tree (current session + sessions spawned by it, such as subagents; the main session can reach every session of the same agent).

{
  tools: {
    sessions: {
      // "self" | "tree" | "agent" | "all"
      visibility: "tree",
    },
  },
}

- `self`: only the current session key. - `tree`: current session + sessions spawned by the current session (subagents). When the caller is the canonical main session, it includes every same-agent session for list, history, search, send, and status. - `agent`: any session belonging to the current agent id (can include other users if you run per-sender sessions under the same agent id). - `all`: any session. Cross-agent targeting still requires `tools.agentToAgent`. - `self` remains strict for main. Incognito denial remains absolute, and cross-agent access still requires `all` plus `tools.agentToAgent` policy. - Sandbox clamp: when the current session is sandboxed and `agents.defaults.sandbox.sessionToolsVisibility="spawned"` (the default), access stays limited to spawned sessions even if the caller is main or `tools.sessions.visibility="all"`. - When not `all`, `sessions_list` includes a compact `visibility` field describing the effective mode and a warning that some sessions may be omitted outside the current scope.

Ambient group watches still queue activity notices and tell the main session where something happened. They do not grant access: main's same-agent access is built into tree. In a multi-user setup, session.dmScope: "main" shares that main session across users; use a per-peer DM scope for isolation, or set tools.sessions.visibility: "self" for strict current-session access.

tools.sessions_spawn

Controls inline attachment support for sessions_spawn.

{
  tools: {
    sessions_spawn: {
      attachments: {
        enabled: false, // opt-in: set true to allow inline file attachments
        maxTotalBytes: 5242880, // 5 MB total across all files
        maxFiles: 50,
        maxFileBytes: 1048576, // 1 MB per file
        retainOnSessionKeep: false, // keep attachments when cleanup="keep"
      },
    },
  },
}

- Attachments require `enabled: true`. - Subagent attachments are materialized into the child workspace at `.openclaw/attachments//` with a `.manifest.json`. - ACP attachments are image-only and forwarded inline to the ACP runtime after the same file count, per-file byte, and total byte limits pass. - Attachment content is automatically redacted from transcript persistence. - Base64 inputs are validated with strict alphabet/padding checks and a pre-decode size guard. - Subagent attachment file permissions are `0700` for directories and `0600` for files. - Subagent cleanup follows the `cleanup` policy: `delete` always removes attachments; `keep` retains them only when `retainOnSessionKeep: true`.

tools.updatePlan

Kill switch for progress_card, the durable plan and status note used for non-trivial multi-step work tracking.

{
  tools: {
    updatePlan: false, // hide progress_card from every run
  },
}

• Default: true for every provider and model. Set false to keep the tool off; there is no model-specific auto-enable rule.
• The tool description tells the model to keep the plan current, use at most one in_progress step, and add Markdown only when it contributes information beyond the steps.
• Use progress_card in new tools.allow and tools.deny policies. Existing policies that name update_plan map to progress_card, so shipped allowlists and denylists keep their meaning.

Older configs used tools.experimental.planTool. Run openclaw doctor --fix to move the value to tools.updatePlan.

agents.defaults.subagents

{
  agents: {
    defaults: {
      subagents: {
        allowAgents: ["research"],
        model: "minimax/MiniMax-M2.7",
        maxConcurrent: 8,
        runTimeoutSeconds: 900,
        announceTimeoutMs: 120000,
        archiveAfterMinutes: 60,
      },
    },
  },
}

• model: default model for spawned sub-agents. If omitted, sub-agents inherit the caller's model.
• allowAgents: default allowlist of configured target agent ids for sessions_spawn when the requester agent does not set its own subagents.allowAgents (["*"] = any configured target; default: same agent only). Stale entries whose agent config was deleted are rejected by sessions_spawn and omitted from agents_list; run openclaw doctor --fix to clean them up.
• maxConcurrent: max concurrent sub-agent runs. Default: 8.
• runTimeoutSeconds: timeout (seconds) for sessions_spawn when the caller does not pass its own override. Default: 0 (no timeout); the 900 shown above is a common opt-in value, not the built-in default.
• announceTimeoutMs: per-call timeout (milliseconds) for gateway agent announce delivery attempts. Default: 120000. Transient retries can make the total announce wait longer than one configured timeout.
• archiveAfterMinutes: minutes after a sub-agent session completes before it is auto-archived. Default: 60; 0 disables auto-archive.
• Per-subagent tool policy: tools.subagents.tools.allow / tools.subagents.tools.deny.

---

Custom providers and base URLs

Provider plugins publish their own model catalog rows. Add custom providers via models.providers in config or ~/.openclaw/agents/<agentId>/agent/models.json.

Configuring a custom/local provider baseUrl is also the narrow network trust decision for model HTTP requests: OpenClaw allows that exact scheme://host:port origin through the guarded fetch path, without adding a separate config option or trusting other private origins.

{
  models: {
    mode: "merge", // merge (default) | replace
    providers: {
      "custom-proxy": {
        baseUrl: "http://localhost:4000/v1",
        apiKey: "LITELLM_KEY",
        api: "openai-completions", // openai-completions | openai-responses | anthropic-messages | google-generative-ai | etc.
        models: [
          {
            id: "llama-3.1-8b",
            name: "Llama 3.1 8B",
            reasoning: false,
            input: ["text"],
            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
            contextWindow: 128000,
            contextTokens: 96000,
            maxTokens: 32000,
          },
        ],
      },
    },
  },
}

- Use `authHeader: true` + `headers` for custom auth needs. - Override agent config root with `OPENCLAW_AGENT_DIR`. - Merge precedence for matching provider IDs: - Non-empty agent `models.json` `baseUrl` values win. - Non-empty agent `apiKey` values win only when that provider is not SecretRef-managed in current config/auth-profile context. - SecretRef-managed provider `apiKey` values are refreshed from source markers (`ENV_VAR_NAME` for env refs, `secretref-managed` for file/exec/store refs) instead of persisting resolved secrets. - SecretRef-managed provider header values are refreshed from source markers (`secretref-env:ENV_VAR_NAME` for env refs, `secretref-managed` for file/exec/store refs). - Empty or missing agent `apiKey`/`baseUrl` fall back to `models.providers` in config. - Matching model `contextWindow`/`maxTokens`: the explicit config value wins when present and valid (a positive finite number); otherwise the implicit/generated catalog value is used. - Matching model `contextTokens` follows the same explicit-wins-else-implicit rule; use it to limit effective context without changing native model metadata. - Provider-plugin catalogs are stored as generated plugin-owned catalog shards under the agent's plugin state. - Use `models.mode: "replace"` when you want config to fully rewrite `models.json` and skip merging in plugin-owned catalog shards. - Marker persistence is source-authoritative: markers are written from the active source config snapshot (pre-resolution), not from resolved runtime secret values.

Provider field details

- `models.mode`: provider catalog behavior (`merge` or `replace`). - `models.providers`: custom provider map keyed by provider id. - Safe edits: use `openclaw config set models.providers. '' --strict-json --merge` or `openclaw config set models.providers..models '' --strict-json --merge` for additive updates. `config set` refuses destructive replacements unless you pass `--replace`. - `models.providers.*.api`: request adapter (`openai-completions`, `openai-responses`, `openai-chatgpt-responses`, `anthropic-messages`, `google-generative-ai`, `google-vertex`, `github-copilot`, `bedrock-converse-stream`, `ollama`, `azure-openai-responses`). For self-hosted `/v1/chat/completions` backends such as MLX, vLLM, SGLang, and most OpenAI-compatible local servers, use `openai-completions`. A custom provider with `baseUrl` but no `api` defaults to `openai-completions`; set `openai-responses` only when the backend supports `/v1/responses`. - `models.providers.*.apiKey`: provider credential (prefer SecretRef/env substitution). - `models.providers.*.auth`: auth strategy (`api-key`, `token`, `oauth`, `aws-sdk`). - `models.providers.*.maxTokens`: default output-token cap for models under this provider when the model entry does not set `maxTokens`. - `models.providers.*.timeoutSeconds`: optional per-provider model HTTP request timeout in seconds, including connect, headers, body, and total request abort handling. - `models.providers.*.injectNumCtxForOpenAICompat`: for Ollama + `openai-completions`, inject `options.num_ctx` into requests (default: `true`). - `models.providers.*.authHeader`: force credential transport in the `Authorization` header when required. - `models.providers.*.baseUrl`: upstream API base URL. - `models.providers.*.headers`: extra static headers for proxy/tenant routing. `models.providers.*.request`: transport overrides for model-provider HTTP requests.

- `request.headers`: extra headers (merged with provider defaults). Values accept SecretRef.
- `request.auth`: auth strategy override. Modes: `"provider-default"` (use provider's built-in auth), `"authorization-bearer"` (with `token`), `"header"` (with `headerName`, `value`, optional `prefix`).
- `request.proxy`: HTTP proxy override. Modes: `"env-proxy"` (use `HTTP_PROXY`/`HTTPS_PROXY` env vars), `"explicit-proxy"` (with `url`). Both modes accept an optional `tls` sub-object.
- `request.tls`: TLS override for direct connections. Fields: `ca`, `cert`, `key`, `passphrase` (all accept SecretRef), `serverName`, `insecureSkipVerify`.
- `request.allowPrivateNetwork`: when `true`, allow model-provider HTTP requests to private, CGNAT, or similar ranges through the provider HTTP fetch guard. Custom/local provider base URLs already trust the exact configured origin, except metadata/link-local origins, which remain blocked without explicit opt-in. Set this to `false` to opt out of exact-origin trust. WebSocket uses the same `request` for headers/TLS but not that fetch SSRF gate. Default `false`.

- `models.providers.*.models`: explicit provider model catalog entries. - `models.providers.*.models.*.input`: model input modalities. Use `["text"]` for text-only models and `["text", "image"]` for native image/vision models. Image attachments are only injected into agent turns when the selected model is marked image-capable. - `models.providers.*.models.*.contextWindow`: native context-window metadata for that model. - `models.providers.*.models.*.contextTokens`: optional active-input cap for that model; use it when you want an effective budget distinct from the model's native `contextWindow`; `openclaw models list` shows both when they differ.

#### Custom provider capability declarations

Provider catalogs own `compat` for bundled and catalog-known model routes. Do not copy those flags into config: OpenClaw uses the catalog row when the configured `api` and `baseUrl` still identify that route. `openclaw doctor --fix` removes matching legacy overrides and reports divergent values for review.

A `compat` block remains supported for a genuinely custom provider, custom model, or catalog model routed to a different endpoint. Set only capabilities verified against that endpoint:

| Custom-route key | Runtime contract |
| --- | --- |
| `supportsStore` | Accepts the OpenAI `store` request field. |
| `supportsPromptCacheKey` | Accepts OpenAI prompt-cache/session-affinity keys. |
| `supportsDeveloperRole` | Accepts `developer` messages instead of requiring `system`. |
| `supportsReasoningEffort` | Accepts a reasoning-effort control. |
| `supportsTemperature` | Accepts `temperature` for this model and adapter. |
| `supportsUsageInStreaming` | Emits usage metadata in streaming responses. |
| `supportsTools` | Supports structured tool/function calling. Set `false` to disable tools. |
| `supportsStrictMode` | Accepts strict tool schemas. |
| `requiresStringContent` | Requires plain-string Chat Completions message content. |
| `strictMessageKeys` | Requires outgoing messages to contain only accepted keys. |
| `visibleReasoningDetailTypes` | Names reasoning detail block types safe to show in transcripts. |
| `supportedReasoningEfforts` | Lists the endpoint's accepted reasoning labels. |
| `reasoningEffortMap` | Maps OpenClaw thinking labels to endpoint-specific labels. |
| `maxTokensField` | Selects `max_tokens` or `max_completion_tokens`. |
| `thinkingFormat` | Selects the endpoint's reasoning payload dialect. |
| `requiresToolResultName` | Requires a tool name on tool-result messages. |
| `requiresAssistantAfterToolResult` | Requires an assistant message after tool results. |
| `requiresThinkingAsText` | Replays reasoning as text rather than structured content. |
| `requiresReasoningContentOnAssistantMessages` | Preserves DeepSeek-style `reasoning_content` during replay. |
| `toolSchemaProfile` | Selects a tool-schema normalization profile. Custom model entries recognize `llamacpp` and `gemini`. The `llamacpp` profile removes `pattern` and `maxLength` values at or above 2000; built-in `llama-cpp`, `ollama`, and `lmstudio` providers apply the same cleaner automatically. Custom `llama-server` models must select it explicitly. See the llama.cpp example below. |
| `unsupportedToolSchemaKeywords` | Removes named JSON Schema keywords rejected by the endpoint before tool schemas are sent. Use this for endpoint-specific gaps beyond a profile's targeted transformations. |
| `toolCallArgumentsEncoding` | Selects the endpoint's tool-call argument encoding. |
| `requiresOpenAiAnthropicToolPayload` | Converts OpenAI-shaped tool calls to Anthropic-family payloads. |

- `plugins.entries.amazon-bedrock.config.discovery`: Bedrock auto-discovery settings root. - `plugins.entries.amazon-bedrock.config.discovery.enabled`: turn implicit discovery on/off. - `plugins.entries.amazon-bedrock.config.discovery.region`: AWS region for discovery. - `plugins.entries.amazon-bedrock.config.discovery.providerFilter`: optional provider-id filter for targeted discovery. - `plugins.entries.amazon-bedrock.config.discovery.refreshInterval`: polling interval for discovery refresh. - `plugins.entries.amazon-bedrock.config.discovery.defaultContextWindow`: fallback context window for discovered models. - `plugins.entries.amazon-bedrock.config.discovery.defaultMaxTokens`: fallback max output tokens for discovered models.

Interactive custom-provider onboarding infers image input for known vision-model-id patterns, including GPT-4o/GPT-4.1/GPT-5+, the o1/o3/o4 reasoning families, Claude, Gemini, any -vl-suffixed id (Qwen-VL and similar), and named families such as LLaVA, Pixtral, InternVL, Mllama, MiniCPM-V, and GLM-4V; it skips the extra question for known text-only families (Llama, DeepSeek, Mistral/Mixtral, Kimi/Moonshot, Codestral, Devstral, Phi, QwQ, CodeLlama, and bare Qwen ids without a vl/vision suffix). Unknown model IDs still prompt for image support. Non-interactive onboarding uses the same inference; pass --custom-image-input to force image-capable metadata or --custom-text-input to force text-only metadata.

Provider examples

The official external `cerebras` provider plugin can configure this via `openclaw onboard --auth-choice cerebras-api-key`. Use explicit provider config only when overriding defaults.

```json5
{
  env: { vars: { CEREBRAS_API_KEY: "sk-..." } },
  agents: {
    defaults: {
      model: {
        primary: "cerebras/zai-glm-4.7",
        fallbacks: ["cerebras/gpt-oss-120b"],
      },
      models: {
        "cerebras/zai-glm-4.7": { alias: "GLM 4.7 (Cerebras)" },
        "cerebras/gpt-oss-120b": { alias: "GPT OSS 120B (Cerebras)" },
      },
    },
  },
  models: {
    mode: "merge",
    providers: {
      cerebras: {
        baseUrl: "https://api.cerebras.ai/v1",
        apiKey: "${CEREBRAS_API_KEY}",
        api: "openai-completions",
        models: [
          { id: "zai-glm-4.7", name: "GLM 4.7 (Cerebras)" },
          { id: "gpt-oss-120b", name: "GPT OSS 120B (Cerebras)" },
        ],
      },
    },
  },
}
```

Use `cerebras/zai-glm-4.7` for Cerebras; `zai/glm-4.7` for Z.AI direct.

```json5 { env: { vars: { KIMI_API_KEY: "sk-..." } }, agents: { defaults: { model: { primary: "kimi/kimi-for-coding" }, models: { "kimi/kimi-for-coding": { alias: "Kimi Code" } }, }, }, } ```

Anthropic-compatible, built-in provider. Shortcut: `openclaw onboard --auth-choice kimi-code-api-key`.

Point a **custom** `openai-completions` provider at a remote `llama-server` (or another OpenAI-compatible llama.cpp endpoint). The built-in `llama-cpp`, `ollama`, and `lmstudio` providers apply the llama.cpp schema cleaner automatically; a custom endpoint does not. Set `compat.toolSchemaProfile: "llamacpp"` on each model whose llama-server chat template compiles tool arguments into GBNF. The profile removes `pattern` and `maxLength` values at or above 2000, covering the `cron` tool's `trigger.script` limit of 65536. It is a targeted mitigation, not complete compatibility for every JSON Schema constraint or `minLength`.

```json5
{
  agents: {
    defaults: {
      model: { primary: "my-llamacpp/qwen35" },
    },
  },
  models: {
    mode: "merge",
    providers: {
      "my-llamacpp": {
        baseUrl: "http://127.0.0.1:8080/v1",
        apiKey: "llamacpp-no-key",
        api: "openai-completions",
        models: [
          {
            id: "qwen35",
            name: "Qwen3.5 (llama-server)",
            contextWindow: 8192,
            maxTokens: 2048,
            compat: {
              supportsTools: true,
              toolSchemaProfile: "llamacpp",
            },
          },
        ],
      },
    },
  },
}
```

On older builds without `toolSchemaProfile`, the broader fallback is `compat.unsupportedToolSchemaKeywords: ["pattern", "patternProperties", "format", "propertyNames", "uniqueItems", "contains", "minContains", "maxContains", "minLength", "maxLength"]`. Unlike the profile, this removes every listed keyword unconditionally.

See [Local Models](/gateway/local-models). TL;DR: run a large local model via LM Studio Responses API on serious hardware; keep hosted models merged for fallback. ```json5 { agents: { defaults: { model: { primary: "minimax/MiniMax-M3" }, models: { "minimax/MiniMax-M3": { alias: "Minimax" }, }, }, }, models: { mode: "merge", providers: { minimax: { baseUrl: "https://api.minimax.io/anthropic", apiKey: "${MINIMAX_API_KEY}", api: "anthropic-messages", models: [ { id: "MiniMax-M3", name: "MiniMax M3", reasoning: true, input: ["text", "image"], cost: { input: 0.6, output: 2.4, cacheRead: 0.12, cacheWrite: 0 }, contextWindow: 1000000, maxTokens: 131072, }, ], }, }, }, } ```

Set `MINIMAX_API_KEY`. Shortcuts: `openclaw onboard --auth-choice minimax-global-api` or `openclaw onboard --auth-choice minimax-cn-api`. The model catalog defaults to M3 and also includes the M2.7 variants. On the Anthropic-compatible streaming path, OpenClaw disables MiniMax M2.x thinking by default unless you explicitly set `thinking` yourself; MiniMax-M3 (and M3.x) stays on the provider's omitted/adaptive thinking path by default. `/fast on` or `params.fastMode: true` rewrites `MiniMax-M2.7` to `MiniMax-M2.7-highspeed`.

```json5 { env: { vars: { MOONSHOT_API_KEY: "sk-..." } }, agents: { defaults: { model: { primary: "moonshot/kimi-k2.6" }, models: { "moonshot/kimi-k2.6": { alias: "Kimi K2.6" } }, }, }, models: { mode: "merge", providers: { moonshot: { baseUrl: "https://api.moonshot.ai/v1", apiKey: "${MOONSHOT_API_KEY}", api: "openai-completions", models: [ { id: "kimi-k2.6", name: "Kimi K2.6", reasoning: false, input: ["text", "image"], cost: { input: 0.95, output: 4, cacheRead: 0.16, cacheWrite: 0 }, contextWindow: 262144, maxTokens: 262144, }, ], }, }, }, } ```

For the China endpoint: `baseUrl: "https://api.moonshot.cn/v1"` or `openclaw onboard --auth-choice moonshot-api-key-cn`.

Native Moonshot endpoints advertise streaming usage compatibility on the shared `openai-completions` transport, and OpenClaw keys that off endpoint capabilities rather than the built-in provider id alone.

```json5 { agents: { defaults: { model: { primary: "opencode/claude-opus-4-6" }, models: { "opencode/claude-opus-4-6": { alias: "Opus" } }, }, }, } ```

Set `OPENCODE_API_KEY` (or `OPENCODE_ZEN_API_KEY`). Use `opencode/...` refs for the Zen catalog or `opencode-go/...` refs for the Go catalog. Shortcut: `openclaw onboard --auth-choice opencode-zen` or `openclaw onboard --auth-choice opencode-go`.

```json5 { env: { vars: { SYNTHETIC_API_KEY: "sk-..." } }, agents: { defaults: { model: { primary: "synthetic/hf:MiniMaxAI/MiniMax-M3" }, models: { "synthetic/hf:MiniMaxAI/MiniMax-M3": { alias: "MiniMax M3" } }, }, }, models: { mode: "merge", providers: { synthetic: { baseUrl: "https://api.synthetic.new/anthropic", apiKey: "${SYNTHETIC_API_KEY}", api: "anthropic-messages", models: [ { id: "hf:MiniMaxAI/MiniMax-M3", name: "MiniMax M3", reasoning: true, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262144, maxTokens: 65536, }, ], }, }, }, } ```

Base URL should omit `/v1` (Anthropic client appends it). Shortcut: `openclaw onboard --auth-choice synthetic-api-key`.

```json5 { agents: { defaults: { model: { primary: "zai/glm-4.7" }, models: { "zai/glm-4.7": {} }, }, }, } ```

Set `ZAI_API_KEY`. Model refs use the canonical `zai/*` provider ID. Shortcut: `openclaw onboard --auth-choice zai-api-key`.

- General endpoint: `https://api.z.ai/api/paas/v4`
- Coding endpoint: `https://api.z.ai/api/coding/paas/v4`
- The default `zai-api-key` auth choice probes your key and auto-detects which endpoint it belongs to (falling back to a prompt, defaulting to Global, if detection is inconclusive). Dedicated CN and Coding-Plan auth choices are also available for explicit selection.
- For the general endpoint, define a custom provider with the base URL override.

---

Related

• Configuration — agents
• Configuration — channels
• Configuration reference — other top-level keys
• Tools and plugins