Hermes

Hermes "Provider temporarily unavailable, retrying automatically"

Last checked

The error

⏳ Provider temporarily unavailable — retrying automatically in 18s (cycle 1/5); press Esc to stop

⏳ Automatic recovery gave up after 5 cycles — the provider is still unavailable.

From agent/turn_recovery_autorecover.py. On messaging platforms the first line ends with "send /stop to cancel"; on the API server, "cancel the request to stop".

This is Hermes waiting, not crashing. Your provider returned a temporary failure (a 5xx, an overloaded 529 or a timeout), and the normal retries and fallback providers were already used up, so Hermes parks the turn and tries again in up to five cycles. Wait a cycle, check the provider's status page, or switch now with /model.

Why it happens

Hermes has an auto-recovery ladder for turns that fail before any answer text reaches you. It runs only after max retries and the whole fallback chain are spent, and only for transient failures.

  1. The provider returned a 5xx server error or an overloaded (529) response.
  2. The connection or read timed out.
  3. No fallback provider is configured, or every fallback failed too.
  4. Auth, billing, bad-request and content-policy errors never enter this ladder, and a 429 rate limit is not treated as transient either (issue #140255, open).

The fix

  1. 1 Wait for the next cycle. The waits are about 15, 30, 60, 60 and 60 seconds, or the provider's own Retry-After up to 120 seconds.
  2. 2 Check the provider's status page.
  3. 3 Switch models now with /model. In the CLI, TUI and Desktop press Esc to stop the wait; on chat platforms send /stop.
  4. 4 Add fallback_providers in config.yaml so another provider takes over before the ladder starts.
  5. 5 To fail fast instead of waiting, set agent.auto_recovery_cycles to 0.
hermes config set agent.auto_recovery_cycles 0

Still failing?

  • If every cycle fails, read the gateway log (hermes logs) for the provider's real status code.
  • If the provider is up but Hermes still waits, check base_url and network access from the host.

Related errors

Full guideHermes Agent Not Working? 14 Errors and Their Exact Fixes

Hit a different error?

Paste any agent error and get the cause and fix in seconds.

Open the decoder

Frequently asked questions

How many times will Hermes retry?

By default five recovery cycles (agent.auto_recovery_cycles), each after the normal retries and fallbacks. Then it stops with Automatic recovery gave up after 5 cycles.

Will it repeat text I already saw?

No. The ladder only runs when no answer text has reached you yet, so it never replays a partly delivered reply.

Stop firefighting agent errors

Decoding errors one at a time is the manual version of what BetterClaw automates. Run your agents on a no-code AI agent platform with managed models, retries and config validation built in.

Deploy Your Agent — Start FreeBook a Demo

Free plan available · Pro $49/mo · BYOK · 7-day free trial