Hermes Agent credential pools rotate same-provider keys past rate limits

Teknium · x · 2026-09-13

Teknium points out that multi-subscription switching for ChatGPT/Codex already exists in Hermes Agent as "credential pools":

Register multiple API keys or OAuth tokens per provider; on 429 rate limits, plan usage caps (e.g. ChatGPT/Codex "usage limit reached"), or 402 billing errors, Hermes rotates to the next healthy key without killing the session.
Distinct from fallback providers: pools rotate within one provider; only when all pool keys are exhausted does cross-provider failover kick in.
Four key-selection strategies (roundrobin / leastused / fillfirst / random); expired OAuth tokens get refreshed first.
Key gotcha: provider prompt caches are scoped per API key, so every rotation re-reads the full history at undiscounted input price on long conversations. A single Nous Portal OAuth covers 300+ models, so most Portal users don't need a pool.

More from coding & agent