"response remained truncated after N continuation attempts"
"Response remained truncated after N continuation attempts" is the retry-exhausted follow-up to a normal truncation. Hermes re-prompts the model to continue after hitting the output ceiling, but every attempt inherits the same broken model.max_tokens (bug #4404) or the same low provider default, while each partial answer refills the context window. Fix the underlying limit - set HERMES_MAX_TOKENS in ~/.hermes/.env, raise Ollama's num_ctx and num_predict, or run /compress - and the response completes on the first pass.
Likely cause
This is the retry-exhausted follow-up to a plain truncation, not a separate fault. When Hermes hits the output ceiling mid-stream it automatically re-prompts the model to continue where it left off. Every retry inherits the same broken model.max_tokens from config.yaml (bug #4404) or the same low provider default (Ollama defaults to 2048), and each partial answer is fed back into the context so the window fills faster - so all attempts hit the identical wall and Hermes gives up.
The fix
- 1 Fix the root output limit, not the retry message: set HERMES_MAX_TOKENS in ~/.hermes/.env - the config.yaml path is ignored.
- 2 On Ollama, raise num_ctx and num_predict in a Modelfile so continuations have room.
- 3 Free the context the retries have been filling: run /compress or start a fresh session, then re-run the prompt.
export HERMES_MAX_TOKENS=8192Hit a different error?
Paste any agent error and get the cause and fix in seconds.
Frequently asked questions
How is this different from "response truncated due to output length limit"?
Same root cause, later stage. The output-length message is the first truncation; this one means Hermes already retried the continuation several times and every attempt hit the same ceiling. Fixing the output limit clears both.
Mine says 4 attempts, the docs say 3. Does the number matter?
No. The number is just how many continuation retries that Hermes build budgets, and it varies between versions. Whether it says 3 or 4, the meaning is identical: every retry ran out of room.
The message says "Try again" - should I just re-run it?
Re-running sends the same request into the same ceiling, so it usually fails the same way. Raise HERMES_MAX_TOKENS (or num_predict on Ollama) and clear the context first, then retry.
Stop firefighting agent errors
Decoding errors one at a time is the manual version of what BetterClaw automates. Run your OpenClaw agents hosted with managed models, retries and config validation built in.
Free plan available · Pro $49/mo · BYOK · 7-day money-back guarantee