openclaw-openclaw/docs/help/faq-models.md

summary, read_when, title, sidebarTitle

summary	read_when	title	sidebarTitle
FAQ: model defaults, selection, aliases, switching, failover, and auth profiles	Choosing or switching models, configuring aliases Debugging model failover / "All models failed" Understanding auth profiles and how to manage them	FAQ: models and auth	Models FAQ

Model- and auth-profile Q&A. For setup, sessions, gateway, channels, and troubleshooting, see the main FAQ.

Models: defaults, selection, aliases, switching

Set with:

```text
agents.defaults.model.primary
```

Models are `provider/model` refs (example: `openai/gpt-5.5`,
`anthropic/claude-sonnet-4-6`). Always set `provider/model` explicitly. If
you omit the provider, OpenClaw tries an alias match first, then a unique
configured-provider match for that model id, then falls back to the
configured default provider (deprecated compatibility path). If that
provider no longer has the configured default model, OpenClaw falls back
to the first configured provider/model instead of a stale default.

Use the strongest latest-generation model your provider stack offers, especially for tool-enabled or untrusted-input agents — weaker or over-quantized models are more vulnerable to prompt injection and unsafe behavior (see [Security](/gateway/security)). Route cheaper models to routine/low-stakes chat by agent role.

Route models per agent and use sub-agents to parallelize long tasks (each
sub-agent consumes its own tokens). See [Models](/concepts/models),
[Sub-agents](/tools/subagents), [MiniMax](/providers/minimax), and
[Local models](/gateway/local-models).

Change only the model fields — avoid full config replaces.

- `/model` in chat (per-session, see [Slash commands](/tools/slash-commands))
- `openclaw models set ...` (updates just model config)
- `openclaw configure --section model` (interactive)
- edit `agents.defaults.model` in `~/.openclaw/openclaw.json` directly

For RPC edits, inspect with `config.schema.lookup` first (normalized
path, shallow schema docs, child summaries), then prefer `config.patch`
over `config.apply` with a partial object. If you did overwrite config,
restore from backup or run `openclaw doctor` to repair.

Docs: [Models](/concepts/models), [Configure](/cli/configure),
[Config](/cli/config), [Doctor](/gateway/doctor).

Yes — Ollama is the easiest path. Quick setup:

1. Install Ollama from `https://ollama.com/download`
2. Pull a local model, e.g. `ollama pull gemma4`
3. For cloud models too, run `ollama signin`
4. Run `openclaw onboard`, choose `Ollama`, then `Local` or `Cloud + Local`

`Cloud + Local` gives you cloud models plus your local Ollama models;
cloud models such as `kimi-k2.5:cloud` need no local pull. To switch
manually: `openclaw models list`, then `openclaw models set ollama/<model>`.

Smaller/heavily quantized models are more vulnerable to prompt injection.
Use large models for any bot with tool access; if you use small models
anyway, enable sandboxing and strict tool allowlists.

Docs: [Ollama](/providers/ollama), [Local models](/gateway/local-models),
[Model providers](/concepts/model-providers), [Security](/gateway/security),
[Sandboxing](/gateway/sandboxing).

Send `/model ` as a standalone message. See [Slash commands](/tools/slash-commands) for the full command list, including the numbered picker (`/model`, `/model list`, `/model 3`), `/model default` to clear a session override, and `/model status` for endpoint/API-mode detail.

Force a specific auth profile per session with `@profile`:

```text
/model opus@anthropic:default
/model opus@anthropic:work
```

To unpin a profile set with `@profile`, re-run `/model` without the
suffix (e.g. `/model anthropic/claude-opus-4-6`), or pick the default from
`/model`. Use `/model status` to confirm the active auth profile.

`/model provider/model` selects that exact provider route. For example, `qianfan/deepseek-v4-flash` and `deepseek/deepseek-v4-flash` are different refs even though the model id matches — OpenClaw does not silently switch providers on a bare id match.

A user-selected `/model` ref is strict for fallback: if that
provider/model becomes unavailable, the reply fails visibly instead of
falling back to `agents.defaults.model.fallbacks`. Configured fallback
chains still apply to configured defaults, cron job primaries, and
auto-selected fallback state. When a non-session-override run is allowed
to use fallback, OpenClaw tries the requested provider/model first, then
configured fallbacks, then the configured primary — so duplicate bare
model ids never jump straight back to the default provider.

See [Models](/concepts/models) and [Model failover](/concepts/model-failover).

Yes — model choice and runtime choice are separate:

- **Native Codex coding agent:** set `agents.defaults.model.primary` to
  `openai/gpt-5.5`. Sign in with `openclaw models auth login --provider
  openai` for ChatGPT/Codex subscription auth.
- **Direct OpenAI API tasks outside the agent loop:** configure
  `OPENAI_API_KEY` for images, embeddings, speech, realtime, and other
  non-agent OpenAI API surfaces.
- **OpenAI agent API-key auth:** `/model openai/gpt-5.5` with an ordered
  `openai` API-key profile.
- **Sub-agents:** route coding tasks to a Codex-focused agent with its
  own `openai/gpt-5.5` model.

See [Models](/concepts/models) and [Slash commands](/tools/slash-commands).

- **Per session:** send `/fast on` while using `openai/gpt-5.5`. - **Per model default:** set `agents.defaults.models["openai/gpt-5.5"].params.fastMode` to `true`. - **Automatic cutoff:** `/fast auto` or `params.fastMode: "auto"` runs new model calls fast until the cutoff, then runs later retry, fallback, tool-result, or continuation calls without fast mode. Cutoff defaults to 60 seconds; override with `params.fastAutoOnSeconds` on the model.

```json5
{
  agents: {
    defaults: {
      models: {
        "openai/gpt-5.5": {
          params: {
            fastMode: "auto",
            fastAutoOnSeconds: 30,
          },
        },
      },
    },
  },
}
```

Fast mode maps to `service_tier = "priority"` on native OpenAI Responses
requests; existing `service_tier` values are preserved and fast mode does
not rewrite `reasoning` or `text.verbosity`. Session `/fast` overrides beat
config defaults.

See [Thinking and fast mode](/tools/thinking) and the Fast mode section
under Advanced configuration on the [OpenAI](/providers/openai) provider
page.

If `agents.defaults.models` is set, it becomes the **allowlist** for `/model` and session overrides. Picking a model outside that list returns this instead of a normal reply:

```text
Model "provider/model" is not allowed. Use /models to list providers, or /models <provider> to list models.
Add it with: openclaw config set agents.defaults.models '{"provider/model":{}}' --strict-json --merge
```

Fix: add the exact model to `agents.defaults.models`, add a provider
wildcard such as `"provider/*": {}` for dynamic catalogs, remove the
allowlist, or pick a model from `/model list`. If the command also
included `--runtime codex`, update the allowlist first, then retry the
same `/model provider/model --runtime codex` command.

If you're on an older OpenClaw release, upgrade first (or run from source `main`) and restart the gateway — `MiniMax-M3` may not be in your installed release's catalog yet. Otherwise the MiniMax provider is not configured (no provider entry or auth profile found), so the model can't resolve. See the Troubleshooting section on the [MiniMax](/providers/minimax) provider page for the full fix checklist, provider/model id table, and config-block example. Yes. Use MiniMax as the default and switch models per session — fallbacks are for errors, not "hard tasks", so use `/model` or a separate agent.

**Option A: switch per session**

```json5
{
  env: { MINIMAX_API_KEY: "sk-...", OPENAI_API_KEY: "sk-..." },
  agents: {
    defaults: {
      model: { primary: "minimax/MiniMax-M3" },
      models: {
        "minimax/MiniMax-M3": { alias: "minimax" },
        "openai/gpt-5.5": { alias: "gpt" },
      },
    },
  },
}
```

Then `/model gpt`.

**Option B: separate agents** — Agent A defaults to MiniMax, Agent B
defaults to OpenAI; route by agent or use `/agent` to switch.

Docs: [Models](/concepts/models), [Multi-Agent Routing](/concepts/multi-agent),
[MiniMax](/providers/minimax), [OpenAI](/providers/openai).

Yes — built-in shorthands, applied only when the target model exists in `agents.defaults.models`:

| Alias | Resolves to |
| --- | --- |
| `opus` | `anthropic/claude-opus-4-8` |
| `sonnet` | `anthropic/claude-sonnet-4-6` |
| `gpt` | `openai/gpt-5.4` |
| `gpt-mini` | `openai/gpt-5.4-mini` |
| `gpt-nano` | `openai/gpt-5.4-nano` |
| `gemini` | `google/gemini-3.1-pro-preview` |
| `gemini-flash` | `google/gemini-3-flash-preview` |
| `gemini-flash-lite` | `google/gemini-3.1-flash-lite` |

Your own alias with the same name overrides the built-in one.

Aliases live at `agents.defaults.models..alias`:

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

Then `/model sonnet` (or `/<alias>` when supported) resolves to that
model id.

OpenRouter (pay-per-token; many models):

```json5
{
  agents: {
    defaults: {
      model: { primary: "openrouter/anthropic/claude-sonnet-4-6" },
      models: { "openrouter/anthropic/claude-sonnet-4-6": {} },
    },
  },
  env: { OPENROUTER_API_KEY: "sk-or-..." },
}
```

Z.AI (GLM models):

```json5
{
  agents: {
    defaults: {
      model: { primary: "zai/glm-5.1" },
      models: { "zai/glm-5.1": {} },
    },
  },
  env: { ZAI_API_KEY: "..." },
}
```

Missing provider key for a referenced provider/model raises a runtime
auth error (e.g. `No API key found for provider "zai"`).

**No API key found for provider after adding a new agent**

A new agent has an empty auth store — auth is per-agent, stored at:

```text
~/.openclaw/agents/<agentId>/agent/auth-profiles.json
```

Fix: run `openclaw agents add <id>` and configure auth in the wizard, or
copy only portable static `api_key`/`token` profiles from the main
agent's store. For OAuth, sign in from the new agent when it needs its
own account. See [Multi-Agent Routing](/concepts/multi-agent) for the
full `agentDir` reuse and credential-sharing rules — never reuse
`agentDir` across agents.

Model failover and "All models failed"

Two stages:

1. **Auth profile rotation** within the same provider.
2. **Model fallback** to the next model in `agents.defaults.model.fallbacks`.

Cooldowns apply to failing profiles (exponential backoff), so OpenClaw
keeps responding when a provider is rate-limited or temporarily failing.

The rate-limit bucket covers more than plain `429`: `Too many concurrent
requests`, `ThrottlingException`, `concurrency limit reached`, `workers_ai
... quota limit exceeded`, `resource exhausted`, and periodic
usage-window limits (`weekly/monthly limit reached`) all count as
failover-worthy rate limits.

Billing responses aren't always `402`, and some `402`s stay in the
transient/rate-limit bucket rather than the billing lane. Explicit
billing text on `401`/`403` can still route to billing; provider-specific
text matchers (e.g. OpenRouter `Key limit exceeded`) stay scoped to their
own provider. A `402` that reads like a retryable usage-window or
org/workspace spend limit (`daily limit reached, resets tomorrow`,
`organization spending limit exceeded`) is treated as `rate_limit`, not a
long billing disable.

Context-overflow errors stay off the fallback path entirely — signatures
like `request_too_large`, `input exceeds the maximum number of tokens`,
`input token count exceeds the maximum number of input tokens`, `input is
too long for the model`, or `ollama error: context length exceeded` go to
compaction/retry instead of advancing model fallback.

Generic server-error text is narrower than "anything with unknown/error
in it". Provider-scoped transient shapes that do count as failover
signals: Anthropic bare `An unknown error occurred`, OpenRouter bare
`Provider returned error`, stop-reason errors like `Unhandled stop reason:
error`, JSON `api_error` payloads with transient server text (`internal
server error`, `unknown error, 520`, `upstream error`, `backend error`),
and provider-busy errors like `ModelNotReadyException` when the provider
context matches. Generic internal fallback text like `LLM request failed
with an unknown error.` stays conservative and does not trigger fallback
by itself.

The auth profile id `anthropic:default` has no credentials in the expected auth store.

**Fix checklist:**

- Confirm where profiles live — current:
  `~/.openclaw/agents/<agentId>/agent/auth-profiles.json`; legacy:
  `~/.openclaw/agent/*` (migrated by `openclaw doctor`).
- Confirm the Gateway loads your env var. `ANTHROPIC_API_KEY` set only in
  your shell won't reach a Gateway run via systemd/launchd — put it in
  `~/.openclaw/.env` or enable `env.shellEnv`.
- Confirm you're editing the right agent — multi-agent setups have
  multiple `auth-profiles.json` files.
- Run `openclaw models status` to see configured models and provider
  auth state.

**For "No credentials found for profile anthropic" (no email suffix):**

The run is pinned to an Anthropic profile the Gateway can't find.

- Use Claude CLI: run `openclaw models auth login --provider anthropic
  --method cli --set-default` on the gateway host.
- Prefer an API key instead: put `ANTHROPIC_API_KEY` in
  `~/.openclaw/.env` on the gateway host, then clear any pinned order
  that forces the missing profile:

  ```bash
  openclaw models auth order clear --provider anthropic
  ```

- Remote mode: auth profiles live on the gateway machine, not your
  laptop — confirm you're running commands there.

If your model config includes Google Gemini as a fallback (or you switched to a Gemini shorthand), OpenClaw tries it during fallback. No Google credentials configured gives `No API key found for provider "google"`. Fix: add Google auth, or remove Google models from `agents.defaults.model.fallbacks`/aliases.

**LLM request rejected: thinking signature required (Google Antigravity)**

Cause: session history has thinking blocks without signatures (often
from an aborted/partial stream); Google Antigravity requires signatures
on thinking blocks. OpenClaw strips unsigned thinking blocks for Google
Antigravity Claude; if it still appears, start a new session or set
`/thinking off` for that agent.

Auth profiles: what they are and how to manage them

Related: /concepts/oauth (OAuth flows, token storage, multi-account patterns)

A named credential record (OAuth or API key) tied to a provider, stored at:

```text
~/.openclaw/agents/<agentId>/agent/auth-profiles.json
```

Inspect saved profiles without dumping secrets: `openclaw models auth
list` (optionally `--provider <id>` or `--json`). See
[Models CLI](/cli/models#auth-profiles).

Provider-prefixed: `anthropic:default` (common when no email identity exists), `anthropic:` for OAuth identities, or a custom id you choose (e.g. `anthropic:work`). Yes. `auth.order.` config sets rotation order per provider (metadata only — no secrets stored).

OpenClaw may skip a profile in a short **cooldown** (rate limits,
timeouts, auth failures) or a longer **disabled** state
(billing/insufficient credits). Inspect with `openclaw models status
--json` and check `auth.unusableProfiles`. Tune with
`auth.cooldowns.billingBackoffHours*`. Rate-limit cooldowns can be
model-scoped — a profile cooling down for one model can still serve a
sibling model on the same provider; billing/disabled windows block the
whole profile.

Set a per-agent order override (stored in that agent's `auth-state.json`):

```bash
# Defaults to the configured default agent (omit --agent)
openclaw models auth order get --provider anthropic

# Lock rotation to a single profile
openclaw models auth order set --provider anthropic anthropic:default

# Or set an explicit order (fallback within provider)
openclaw models auth order set --provider anthropic anthropic:work anthropic:default

# Clear override (fall back to config auth.order / round-robin)
openclaw models auth order clear --provider anthropic

# Target a specific agent
openclaw models auth order set --provider anthropic --agent main anthropic:default
```

Verify what will actually be tried: `openclaw models status --probe`. A
stored profile omitted from an explicit order reports
`excluded_by_auth_order` instead of being tried silently.

- **OAuth / CLI login** often uses subscription access where the provider supports it. For Anthropic, OpenClaw's Claude CLI backend uses Claude Code `claude -p`, which Anthropic currently treats as Agent SDK/programmatic usage drawing from subscription usage limits — see [Anthropic](/providers/anthropic) for the current billing-pause status and source links. - **API keys** use pay-per-token billing.

The wizard supports Anthropic Claude CLI, OpenAI Codex OAuth, and API
keys.

Related

• FAQ — the main FAQ
• FAQ — quick start and first-run setup
• Model selection
• Model failover