solution model medium macos linux windows

/model says 'model not allowed'

Fix 'model not allowed' by updating or clearing the agents.defaults.models allowlist, then selecting an allowed model.

By CoClaw Team • • Updated June 8, 2026

Symptoms

  • In chat, /model <something> returns: Model "provider/model" is not allowed.
  • It can feel like the bot “didn’t respond” because the selection fails before a normal reply.

Cause

agents.defaults.models is configured as an allowlist. When it’s non-empty, only those models can be selected.

This is separate from provider availability: the provider/model ref may exist, but selection still fails if the ref is not present in the allowlist.

Fix

Check the allowlist:

openclaw config get agents.defaults.models

Then do one of:

  • Add the model you want to the allowlist (recommended for safety), or
  • Clear/remove agents.defaults.models to disable allowlisting, or
  • Pick a model that is already in the allowlist (/model list).

If you are also changing providers or fallback models, rebuild the starter model block with OpenClaw Config Generator instead of hand-building the list. Custom mode can emit agents.defaults.models from the selected primary and fallback model refs, which keeps the allowlist aligned with the provider/model paths you are actually exporting.

Use Providers when you need to choose or compare model/provider refs, and /guides/openclaw-configuration for the broader config mental model.

Verify

  • Probe provider/model resolution:
openclaw models status --probe
  • Re-run /model list and confirm the target model is selectable.
  • Select the target model and send one real message; it uses the expected model.

Verification & references

  • Reviewed by:CoClaw Code Team
  • Last reviewed:March 14, 2026
  • Verified on: macOS · Linux · Windows

References

Official docs

  1. OpenClaw docs: /concepts/models
  2. OpenClaw docs: /help/troubleshooting
Want to explore more? Browse all solutions or ask in the Community Forum .
Report a problem

Related Resources

Model/auth failures: rate limit, billing, or 'all models failed'
Fix
Debug OpenClaw model failures by checking provider auth status, probing profiles, switching models/fallbacks, and verifying provider/model refs.
Ollama configured, but OpenClaw still uses Anthropic (or model discovery keeps failing)
Fix
Fix local Ollama setups where gateway logs show Anthropic fallback or repeated Ollama model-discovery failures by pinning provider config, verifying connectivity from the gateway runtime, and separating model selection problems from OpenAI-compatible payload problems.
Model outputs '[Historical context]' / tool-call JSON instead of a normal reply
Fix
Fix chat replies that leak internal tool metadata (e.g. '[Historical context: ... Do not mimic ...]') by switching to a tool-capable model/provider and ensuring function calling is enabled.
Onboarding skips Model/Auth setup (agent unresponsive after install)
Fix
Fix a broken onboarding flow where OpenClaw defaults to a model without credentials, causing the agent to hang, by configuring provider auth manually and rerunning current configuration.
OpenClaw Not Responding: Fix 'no output', Incorrect API, Rate Limits, and Silent Failures
Guide
A high-signal checklist for when OpenClaw stops replying (TUI shows '(no output)', channels go quiet, or logs show 401/403/429). Covers config precedence, provider auth, model allowlists, relay API-mode mismatch, and rate-limit/billing traps.
OpenClaw on Windows: Native vs WSL2, Install Paths, and When to Switch
Guide
A practical decision guide for Windows users: choose between native Windows and WSL2 based on service model, tool compatibility, and debugging cost, then switch paths cleanly when the platform is the problem.