"model context length below the 64K minimum"
Recent Hermes builds enforce a hard 64K context-length minimum, so previously-working 32K models like MiniMax-M2.7 and Kimi-k2.6 are now rejected at startup and can take Telegram bots fully down. provider: auto can also pick a sub-64K model without checking. Set model.context_length in config.yaml to override, or switch to a 64K+ model, then restart.
Likely cause
Recent Hermes builds enforce a hard 64K context-length minimum. Previously-working 32K models (MiniMax-M2.7, Kimi-k2.6) are now rejected at startup, which can take Telegram bots fully down. A second cause: provider: auto can resolve to a sub-64K model without checking the minimum.
The fix
- 1 Set model.context_length in config.yaml to override the detected value, or switch to a 64K+ context model.
- 2 If you use provider: auto, pin your main model and auxiliary.vision.provider to a specific 64K+ model instead of auto.
- 3 Restart the agent.
context_length: 64000Hit a different error?
Paste any agent error and get the cause and fix in seconds.
Frequently asked questions
My model worked last week and now fails to start. What changed?
Hermes added a hard 64K context minimum. Models under that (32K MiniMax-M2.7, Kimi-k2.6) are now rejected at startup even though they worked before. Set model.context_length to override, or move to a 64K+ model.
Why did provider: auto pick a model that's too small?
The auto resolver doesn't check the 64K minimum, so it can select a sub-64K model that then gets rejected. Pin a specific 64K+ model for both your main and auxiliary.vision providers instead of relying on auto.
Will overriding context_length cause problems?
Only if you set it above what the model actually supports. Match it to the model's real window (64000 or higher). Setting it correctly lets Hermes compress before hitting the wall instead of refusing to start.
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.
$19/month per agent · BYOK · 7-day money-back guarantee