openclaw models

Model discovery, scanning, and configuration (default model, fallbacks, auth profiles).

Related:

• Providers + models: Models
• Model selection concepts + /models slash command: Models concept
• Provider auth setup: Getting started

Common commands

openclaw models --jsonopenclaw models statusopenclaw models listopenclaw models refreshopenclaw models set <model-or-alias>openclaw models set-image <model-or-alias>openclaw models scan

status, list, and auth subcommands accept --agent <id> to target a configured agent. Without it, model inspection uses agents.defaults.systemAgent.agentId when configured, otherwise the sole configured agent. Auth mutations require --agent <id> when multiple agents are configured.

For models status, OPENCLAW_AGENT_DIR overrides the inspected auth directory when --agent is omitted. A matching configured agentDir retains that agent's ownership during credential refresh. An explicit --agent <id> takes precedence over the environment override.

fallbacks/image-fallbacks manage global defaults. set, set-image, scan, refresh, and aliases also operate globally and reject --agent.

models set and models set-image require the provider to be declared by an installed plugin or configured under models.providers. An unknown provider exits nonzero without changing config. If the provider is known but the model is absent from the local catalog, the command saves the selection and prints a warning because newly released and self-hosted models may not be cataloged yet. Writing agents.defaults.model with openclaw config set is stricter than models set: it rejects a model reference it cannot resolve instead of warning. That check is text-model only; config set does not validate agents.defaults.imageModel at all, so it is not the stricter path for the set-image setting. openclaw doctor --json reports configured unknown providers; add --severity-min info to also see active models that the local catalog cannot confirm.

Default-model, alias, and fallback changes resolve provider-owned model aliases using the current plugin configuration. When stored entries resolve to the selected model, their settings move to its canonical key; existing canonical settings take precedence. Adding an alias replaces the model's previous alias. If config changes during that preparation, the command rejects the write; rerun it against the updated config.

Status

Bare openclaw models is equivalent to openclaw models status. openclaw models --json returns the same object as openclaw models status --json.

openclaw models status shows the resolved default/fallbacks plus an auth overview. Active profile cooldowns appear under Unavailable auth profiles with the stored reason and recovery action; JSON output exposes the same data in auth.unusableProfiles. For plugin-owned agent runtimes such as Codex, status also checks whether the owning plugin is enabled and passed startup payload verification. A route with valid credentials but an unavailable runtime reports status: unavailable instead of usable; JSON output includes separate authStatus, runtimeStatus, and bounded runtime diagnostics. When provider usage snapshots are available, the OAuth/API-key status section includes provider usage windows and quota snapshots. Current usage-window providers: Anthropic, GitHub Copilot, OpenAI, MiniMax, SuperGrok via xAI OAuth, Xiaomi, and z.ai. Usage auth comes from provider-specific hooks when available; otherwise OpenClaw falls back to matching OAuth/API-key credentials from auth profiles, env, or config.

Use an explicit agent when diagnosing that agent's selection:

openclaw models list --agent <agentId>openclaw models status --agent <agentId>openclaw models status --agent <agentId> --json --check

The list shows model inventory. Status explains the configured default, fallbacks, and authentication for their routes. It does not inspect a chat session's model override; use /model status in that session.

Read status correctly

These sections answer different questions:

A stored profile or static health entry does not prove that its credential can be used now. For example, a stored file or exec SecretRef (a reference to a secret) can appear in the overview while its route remains indeterminate. That means readiness could not be confirmed in this command path; it does not mean the provider rejected the credential. Inspect the route diagnostic and the configured secret source before replacing credentials.

Status without --probe is not a model-call test. It can still resolve targeted config secrets and inspect provider-owned auth state. Use --probe only when you intend to make the live requests described below.

With --check, exit codes are:

• 0: no configured-route auth or runtime issue was found, and no selected credential is expiring. This does not prove that a model request will succeed.
• 1: a route has missing or expired auth, an incompatible route, an unavailable runtime, or indeterminate readiness.
• 2: a selected credential is expiring, with no condition that requires exit code 1.

Command failures, such as failure to resolve a required config secret, also exit nonzero. In JSON mode, these can return a command error instead of the status object; check the process exit code as well as the output.

Non-secret placeholders can appear as marker(<value>), such as secretref-managed or an environment variable name. These labels describe the credential source; they are not the resolved secret.

Options:

--probe-timeout requires a positive number; --probe-concurrency and --probe-max-tokens require positive integers. Omit these options to use their defaults (8000, 2, and 8, respectively); explicitly empty values are rejected.

Probe rows can come from auth profiles, env credentials, or models.json. Probe status buckets: ok, auth, rate_limit, billing, timeout, format, unknown, no_model.

Direct models status --probe runs create temporary internal sessions in the selected agent's canonical database, so the command requires exclusive ownership of the configured state directory. Stop a running Gateway with openclaw gateway stop before probing. Probe results can be reported before slow cleanup finishes. Temporary auth directories, internal sessions, and the state lock remain held until accepted work and cleanup settle, including after interruption. Cleanup failures are reported; a timeout does not certify that resources have closed.

Probe detail/reason codes to expect when a probe never reaches a model call:

• excluded_by_auth_order: a stored profile exists, but explicit auth.order.<provider> omitted it, so probe reports the exclusion instead of trying it.
• missing_credential, invalid_expires, expired, unresolved_ref: profile is present but not eligible or resolvable.
• ineligible_profile: profile is incompatible with provider config for another reason.
• no_model: provider auth exists, but OpenClaw could not resolve a probeable model candidate for that provider.

For OpenAI ChatGPT/Codex OAuth troubleshooting, openclaw models status, openclaw models auth list --provider openai, and openclaw config get agents.defaults.model --json are the quickest way to confirm whether an agent has a usable openai OAuth profile for openai/* through the native Codex runtime. See OpenAI provider setup.

List

openclaw models list reads published model inventory. It does not start model provider discovery or rewrite models.json. This also applies to --all and --provider <id>.

openclaw models list --agent <agentId>openclaw models list --agent <agentId> --provider <providerId> --jsonopenclaw models list --agent <agentId> --refresh

When a local Gateway is running, or a remote Gateway is selected by configuration or environment, the command reads that Gateway's catalog. --agent selects an agent on that Gateway. Provider filtering, model visibility and availability use the Gateway's captured config and auth facts. The command does not resolve local model-provider secrets for that request.

A selected Gateway must advertise published-model-catalog. If it does not, update or restart it and retry. Connection, authorization and capability errors are reported directly; they do not switch the command to a different local list.

Without a running local Gateway or an explicit Gateway target, the command identifies that it is showing the local cached catalog. This fallback can prepare configured and static facts and resolve its configured authentication, but starts model discovery only when --refresh is supplied.

Use --refresh to acquire provider inventory before listing. A failed refresh warns while showing available published rows. Successful empty acquisition stays empty; it does not restore the old discovered rows. If the published owner is not ready, retry after Gateway startup or the current refresh finishes.

Options: --all (full published catalog), --refresh (provider discovery), --local (local endpoints), --provider <id>, --agent <id>, --json, and --plain. A provider filter reads the full published inventory for that provider, so it does not require --all.

Notes:

• The Auth column uses prepared credential and runtime evidence. A separate API key does not prove a native CLI login. Unknown readiness stays unknown, and catalog metadata does not prove that a model request will succeed. See Read status correctly.
• A provider-level models.providers.<id>.baseUrl outside the plugin’s declared native endpoints excludes its implicit catalog rows, including cached discovery. Add the models supported by your proxy to models.providers.<id>.models; explicitly authored rows, names, defaults, and aliases remain intact. A model-level URL override alone does not exclude the provider catalog.
• Static catalog rows can remain visible without authentication. Listing them does not grant permission to select a restricted model or change modelPolicy.allow.
• In merge mode, refreshing generated catalogs preserves manual root models, including models whose provider also has a generated catalog. Explicit replace mode keeps its existing replacement behavior.
• For plugins that load only when needed (activation.onStartup: false), an eligible provider in the cached catalog needs current profile, environment, or provider-config authentication before the catalog contributes implicit rows. Credentials retained only inside a generated catalog do not qualify. Explicit model declarations remain visible.
• Ctx shows contextTokens/contextWindow when a runtime cap differs from the native context window. JSON retains contextTokens when provided.
• Input and Ctx use the selected physical route plus explicit configured logical overrides. Unresolved route metadata stays unknown instead of borrowing another route's capabilities.
• Configured model IDs retain case. For example, Reader and reader remain distinct. Provider-owned aliases still apply, and configured aliases remain in the table tags and JSON output.
• --provider takes a provider ID, such as moonshot, rather than a picker label such as Moonshot AI.
• Model refs split on the first /. Include the provider prefix when the model ID contains /, for example openrouter/moonshotai/kimi-k2.

Provider discovery through models list --refresh is separate from the hosted metadata download performed by models refresh, described below. See the Gateway catalog request for the wire controls.

Refresh the hosted catalog

openclaw models refresh [--json] checks the hosted metadata catalog. It does not sign in to providers, test credentials, or activate downloaded rows in a running Gateway. It rejects --agent because the hosted catalog is global.

Restart the Gateway to use downloaded updates. The Gateway reports when a checked catalog needs a restart, including an update downloaded by another process. A successful refresh result describes the download, not live activation. If models.catalogRefresh.enabled is false, the command reports that refresh is disabled.

See Hosted catalog updates for the update lifecycle. The public change history is in openclaw/catalog.

Set default / image model

openclaw models set <model-or-alias>openclaw models set-image <model-or-alias>

set writes agents.defaults.model.primary; set-image writes agents.defaults.imageModel.primary. Both accept provider/model or a configured alias. set also repairs Codex/Copilot runtime plugin installs when the newly selected model needs one; set-image does not. Neither command accepts --agent; they always write agent defaults.

If you omit the provider when selecting a model, OpenClaw tries a configured alias, then a unique configured-provider match for that exact model ID, and finally the configured default provider with a deprecation warning. If that provider no longer exposes the configured default, the first configured provider/model is used.

Scan

models scan reads OpenRouter's public :free catalog and ranks candidates for fallback use. The catalog itself is public, so metadata-only scans do not need an OpenRouter key.

By default OpenClaw tries to probe tool and image support with live model calls. If no OpenRouter key is configured, the command falls back to metadata-only output and explains that :free models still require OPENROUTER_API_KEY for probes and inference.

Options:

• --no-probe (metadata only; no config/secrets lookup)
• --min-params <b>
• --max-age-days <days>
• --provider <name>
• --max-candidates <n>
• --timeout <ms> (catalog request and per-probe timeout)
• --concurrency <n>
• --yes
• --no-input
• --set-default
• --set-image
• --json

Numeric scan options reject empty and whitespace-only values. Omit a flag to retain its default behavior.

--set-default and --set-image require live probes; metadata-only scan results are informational and are not applied to config.

Aliases

openclaw models aliases list [--json] [--plain]openclaw models aliases add <alias> <model-or-alias>openclaw models aliases remove <alias>

Aliases are stored per model entry as agents.defaults.models.<key>.alias. add resolves <model-or-alias> to a canonical provider/model key first, so aliasing an alias repoints it rather than chaining. Adding an alias does not change agents.defaults.modelPolicy.allow or restrict model overrides.

Fallbacks

openclaw models fallbacks list [--json] [--plain]openclaw models fallbacks add <model-or-alias>openclaw models fallbacks remove <model-or-alias>openclaw models fallbacks clear

Manages agents.defaults.model.fallbacks. openclaw models image-fallbacks list|add|remove|clear manages the parallel agents.defaults.imageModel.fallbacks list with the same subcommand shape.

Personal model accounts

Use models accounts for accounts owned by your signed-in person on the selected Gateway. The CLI and Settings → Profile → Connected accounts use the same Gateway account store, including when the server is shared.

To configure system/agent credentials for a remote server, run models auth on that server with its OpenClaw state/config. Configuring a remote Gateway URL on your laptop does not make models auth write to the server.

openclaw models accounts listopenclaw models accounts loginopenclaw models accounts login anthropic --method api-keyopenclaw models accounts login openai --method device-codeopenclaw models accounts login xai --method api-keyopenclaw models accounts use <account-id>openclaw models accounts clear-default <provider>

Every account command shows Gateway, Person, and Scope: Personal before accessing accounts or asking for provider credentials. This context goes to stderr so --json output stays pipeable. The person is the Gateway's saved, verified profile, not your operating-system username or an unsaved display-name edit.

Gateway access and provider sign-in are separate. Account ownership follows the profile assigned by the Gateway on this connection. Single-user Control UI connections can use the durable Owner profile, but short-lived CLI connections are not assigned that profile automatically. The personal-account CLI therefore needs an identity-bearing endpoint, even when the local Control UI already shows Owner.

For separate people on a shared server, use its identity-bearing WebSocket endpoint, such as Tailscale Serve or a trusted proxy. Approve device pairing separately if requested: pairing grants device access, not a distinct person's identity. Supplying shared Gateway token/password credentials takes precedence over Tailscale identity authentication. A browser sign-in does not transfer its identity to the CLI. For an identity-aware edge, follow remote edge authentication.

If no person is identified, the command stops before provider sign-in and explains how to use an identity-bearing endpoint. It does not infer ownership or change Gateway authentication. These commands do not accept --agent or an owner id, and they do not modify local shared auth stores or model config. The top-level openclaw connect command enrolls a node; it is not a personal-account login.

list needs operator.read and returns one page of at most 50 saved accounts: id, provider, friendly label, auth type, and whether each is the new-session default. It never returns credentials. Use --json for structured output and list --cursor <nextCursor> for the next page.

login, use, and clear-default need operator.write. login requires an interactive terminal and offers the same provider and sign-in methods as Add account in the Control UI. Omit the provider to choose from the Gateway's catalog; use --method <id> to select a method directly. The catalog includes only methods enabled for personal accounts by their provider plugin, not every system/agent setup method.

Anthropic uses an API key for personal setup, not a Claude subscription token. OpenAI offers API key, browser sign-in, and device-code methods; Grok (xai) offers API key and device sign-in. Follow the steps shown by the selected Gateway. Credentials and authorization codes go into protected inputs, never command arguments or chat. During browser sign-in, the CLI keeps checking for completion even while a redirect prompt is open.

Keep the command running until it reports a terminal result. Ctrl-C cancels that exact sign-in attempt and waits for the Gateway's acknowledgment; a closed connection must start a fresh attempt. Saving an account does not by itself prove that a model request will succeed.

use selects an already-saved account for new sessions without signing in again. clear-default removes only the personal default for that provider: it keeps saved credentials and existing session selections. Neither operation revokes provider access. See Per-person model accounts for collaborator, fork, and failover behavior.

All account subcommands accept --url <url>, --port <port>, --token-file <path>, --password-file <path>, --timeout <ms> (default 30000), and --json. The token/password files authenticate the Gateway, not the provider, and do not establish a personal identity for this CLI. An explicit --url can use an identity-bearing endpoint without supplying a shared token. It does not send ambient or configured shared credentials to the override; configured edge credentials remain bound to their own endpoint. Flags may precede or follow the leaf command:

openclaw models accounts --timeout 45000 list --jsonopenclaw models accounts list --timeout 45000 --json

Auth profiles

These commands manage System / agent credentials, not personal Gateway accounts. Before provider sign-in, models auth login shows the selected agent and that it is operating on the machine running OpenClaw.

Before a models auth command changes the local auth store, OpenClaw compares the selected CLI state/config paths with the local Gateway or its installed service. A proven mismatch stops before the write. A remote Gateway or an authenticated path that cannot be verified produces a warning instead.

openclaw models auth addopenclaw models auth list [--provider <id>] [--json]openclaw models auth login --provider <id> [--agent <agentId>]openclaw models auth login --provider openai --profile-id openai:workopenclaw models auth login-github-copilotopenclaw models auth logout <profileId> [--yes]openclaw models auth paste-api-key --provider <id>openclaw models auth setup-token --provider <id>openclaw models auth paste-token --provider <id>openclaw models auth order get --provider <id>openclaw models auth order set --provider <id> <profileIds...>openclaw models auth order clear --provider <id>

models auth add is the interactive auth helper. It can launch a provider auth flow (OAuth/API key) or guide you into manual token paste, depending on the provider you choose.

models auth list lists saved auth profiles for the selected agent without printing token, API-key, or OAuth secret material. Active cooldown and disable entries include their reason and recovery action. Legacy Gemini CLI OAuth cooldowns direct you to the supported Google AI Studio API-key setup instead of offering an unavailable Gemini CLI login flow. Use --provider <id> to filter to one provider, such as openai, and --json for scripting.

models auth login runs a provider plugin's auth flow (OAuth/API key). Use openclaw plugins list to see which providers are installed. login accepts --profile-id <id> for providers that support named profiles during login (use this to keep multiple logins for the same provider separate), --method <id> to pick a specific auth method, --device-code as a shortcut for --method device-code, --set-default to apply the provider's recommended default model, and --force to remove existing profiles for that provider first (use when a cached OAuth profile is stuck or you want to switch accounts).

After credentials are saved, an existing model restriction can prompt Show all <Provider> models or Keep current restrictions. Only the first choice adds that provider's wildcard to the current restriction. Credentials stay saved either way. The CLI, private-chat login, and Control UI use the same choice. No prompt appears when the provider is already unrestricted. If restrictions change during sign-in, OpenClaw preserves the newer settings and asks you to choose model access again.

The CLI reports saved model access separately from confirmed Gateway application. If application is not confirmed, run openclaw gateway restart to apply the saved policy to the running Gateway. This is required when automatic config reload is disabled.

Without --set-default, login preserves the current default, including an unset default, and keeps unrelated configuration edits made while login is running. If credentials are saved but provider settings cannot be applied, the error reports the saved credentials separately. Auth changes request a refresh from the running local Gateway; a refresh failure does not undo the saved change, and the command reports how to apply it.

With an older Gateway, the CLI tries its legacy auth-status refresh. This cannot confirm that the saved change is active; follow the restart guidance. This fallback applies to auth changes, not to models list.

For the shared-main agent, --force clears the provider's shared credentials and main-agent local overrides, including their order and health state. For another agent it clears only that agent's local profiles, leaving shared credentials unchanged. A busy auth store stops the command before login starts; close other OpenClaw commands using the same state directory and retry. SQLite lock diagnostics can name either the shared state database or an agent database, so checking only the legacy auth file for open handles does not rule out contention.

models auth logout <profileId> removes one saved auth profile from the selected agent auth store. Use the profile id shown by models auth list. It also drops that profile from auth.profiles and from every auth.order list in your config, so no stale reference is left behind, and it deletes an auth.order.<provider> entry that would otherwise be emptied (an authored empty order means "select no profiles" and would disable the provider). It prompts for confirmation on a TTY; pass --yes for scripts and agents. Provider key references are cleared before the credential is removed. Model defaults and connection settings stay unchanged. Logout refuses when the profile is not in the store.

models auth login-github-copilot is a shortcut for models auth login --provider github-copilot --method device (GitHub device flow); it accepts --yes to overwrite an existing profile without prompting.

Use either openclaw models auth --agent <id> <subcommand> or openclaw models auth <subcommand> --agent <id> to target a specific configured agent store. Both forms are supported by add, list, login, logout, paste-api-key, setup-token, paste-token, login-github-copilot, and order get/set/clear.

For OpenAI models, --provider openai defaults to ChatGPT/Codex account login. Use --method api-key only when you want to add an OpenAI API-key profile, usually as a backup for Codex subscription limits. Run openclaw doctor --fix to migrate older legacy OpenAI Codex prefix auth/profile state to openai.

Examples:

openclaw models auth login --provider openai --set-defaultopenclaw models auth login --provider openai --method api-keyopenclaw models auth paste-api-key --provider openaiopenclaw models auth list --provider openaiopenclaw models auth logout openai:manual --yes

Notes:

• paste-api-key accepts API keys generated elsewhere, prompts for the key value, and uses the same credential writer as the Models page. It updates the configured profile or the active saved key, or creates <provider>:manual when neither exists. Use --profile-id to update a named profile or add a backup without changing the active provider connection. A configured provider stores the profile reference, while key material stays in the auth store. Saved changes report any Gateway refresh failure with a recovery step. In automation, pipe the key on stdin, for example printf "%s\n" "$OPENAI_API_KEY" | openclaw models auth paste-api-key --provider openai.
• setup-token and paste-token remain generic token commands for providers that expose token auth methods.
• setup-token requires an interactive TTY and runs the provider's token-auth method (defaulting to that provider's setup-token method when it exposes one).
• paste-token requires --provider, prompts for the token value by default, and writes it to the default profile id <provider>:manual unless you pass --profile-id. In automation, pipe the token on stdin instead of passing it as an argument so provider credentials do not appear in shell history or process lists.
• paste-token --expires-in <duration> stores an absolute token expiry from a relative duration such as 365d or 12h.
• For openai, OpenAI API keys and ChatGPT/OAuth token material are different auth shapes. Use paste-api-key for sk-... OpenAI API keys and paste-token only for token auth material.
• Anthropic: setup-token/paste-token are supported OpenClaw auth paths for anthropic, but OpenClaw prefers reusing the Claude CLI (claude -p) on the host when it is available.
• auth order get/set/clear manages a per-agent auth profile order override for one provider in the SQLite auth store, separate from the auth.order.<provider> config key. set takes one or more profile ids in priority order. The stored order takes precedence over config for profile selection and CLI runtime routing; clear falls back to config/round-robin ordering.

Related

• CLI reference
• Model selection
• Model failover
• openclaw promos — list and claim promotional model offers