fix(run_agent): honor model.context_length override for sub-64K models

[c0nSpIc0uS7uRk3r]

Summary

Commit c8aff74 ("prevent agent from stopping mid-task") added a MINIMUM_CONTEXT_LENGTH=64K guard in AIAgent.__init__ that rejects agent initialization for any model whose native context window is smaller than 64K. The raised ValueError explicitly documents an escape hatch:

"or set model.context_length in config.yaml to override"

But the guard fires unconditionally — it never checks whether the operator actually set model.context_length in config.yaml. The promised escape hatch is cosmetic.

Fix

Add and _config_context_length is None to the gate condition. _config_context_length is already a local variable in the same __init__ method, populated from the config.yaml value, so no additional plumbing is needed.

Behavior change

• Default (operator has not set model.context_length): unchanged — guard still fires, agent still refuses to start against sub-64K models.
• Override (operator explicitly sets model.context_length in config.yaml, accepting sub-64K risk): agent now starts, matching what the error message already promises.

Repro

1. In config.yaml:

   model:
     context_length: 32000

2. Run hermes chat -q "hello" -Q against a model with a 32K native context (e.g. Qwen 3.5 35B-A3B).
3. Before: ValueError: Model ... has a context window of 32,000 tokens, which is below the minimum 65,536...
4. After: Agent starts normally, runs to completion.

Related

Addresses Bug 1 from #11096 (combined issue covering three v0.9.0 rough edges).

[vivganes]

This is an important fix. Can we get this out in the next release? my hermes is bricked until then since i run a local model with less context window. The error message asks to set model.context_length but that does not work at all. This fixes it.

[alt-glitch]

Likely duplicate of #8962 — same fix: honor model.context_length config override below 64K minimum. Also related to #9142 (same fix, different PR). Addresses Bug 1 from #11096.

[teknium1]

Thanks for the PR. Closing — this is working as intended, not a bug.

The 64K context-window floor is deliberate: Hermes' tool schemas + system prompt consume a large fixed prefix, and sub-64K windows aren't reliable for agentic tool-calling. We don't want anyone running below 64K, so there is intentionally no escape hatch to override the floor with a smaller value.

The real defect here was the error message — it advertised "or set model.context_length in config.yaml to override", which falsely implied a sub-64K override existed and generated a cluster of well-meaning PRs (this one included) trying to wire it up. We've corrected the message in #53569 so it no longer promises an override that was never meant to exist.

Note: model.context_length is still honored for the legitimate case where a local server under-reports its true window — but the declared value must itself be ≥64K. It is not a way to run below the floor.

Appreciate the clean, minimal diff — the reasoning just points the other way on this one.