Background

OpenClaw implements automatic model failover:

• Define multiple auth profiles per provider (API keys, OAuth tokens)
• Define a fallback chain: primary model → backup model → backup provider
• On API errors (rate limit, auth failure, model outage), automatically rotate to next in chain
• No user-visible interruption; transparent failover

Hermes Agent's hermes_cli/models.py has model selection but no automatic failover infrastructure.

Proposal

Add model failover infrastructure:

1. Auth profiles: Allow multiple credentials per provider (key rotation, multiple accounts)
2. Failover chain config: In config.yaml, define models.fallback: ["anthropic/claude-opus-4", "openai/gpt-4.5"]
3. Automatic detection: On 401/403/429/500 errors from provider, rotate to next in chain
4. Retry with backoff: Attempt primary again after cooldown
5. User notification: Inform user when failover occurred ("Using backup model")

References

• OpenClaw: docs/concepts/model-failover, model-fallback.ts