NousResearch/hermes-agent · error · ValueError

Model {agent.model} has a context window of {_ctx:,} tokens,

Error message

Model {agent.model} has a context window of {_ctx:,} tokens, which is below the minimum {MINIMUM_CONTEXT_LENGTH:,} required by Hermes Agent.  Choose a model with at least {MINIMUM_CONTEXT_LENGTH // 1000}K context.  If your server reports a window smaller than the model's true window, set model.context_length in config.yaml to the real value (this must be at least {MINIMUM_CONTEXT_LENGTH // 1000}K).

What it means

Hermes requires at least MINIMUM_CONTEXT_LENGTH (64K) tokens of context for reliable tool-calling workflows; this ValueError fires at init when context_compressor.context_length is positive but below that floor. One documented escape hatch: provider 'lmstudio' with an explicit positive integer model.context_length in config.yaml is allowed below the floor.

Source

Thrown at agent/agent_init.py:2653

    agent.codex_app_server_auto_compaction = codex_app_server_auto_compaction
    agent.codex_responses_native_compaction = codex_responses_native_compaction
    agent.codex_responses_compact_threshold = codex_responses_compact_threshold
    agent.max_compression_attempts = compression_max_attempts
    agent.compression_idle_compact_after_seconds = (
        compression_idle_compact_after_seconds
    )

    # Reject models whose context window is below the minimum required
    # for reliable tool-calling workflows (64K tokens).
    _ctx = getattr(agent.context_compressor, "context_length", 0)
    _allow_lmstudio_explicit_below_floor = (
        str(getattr(agent, "provider", "") or "").strip().lower() == "lmstudio"
        and isinstance(agent._config_context_length, int)
        and not isinstance(agent._config_context_length, bool)
        and agent._config_context_length > 0
    )
    if _ctx and _ctx < MINIMUM_CONTEXT_LENGTH and not _allow_lmstudio_explicit_below_floor:
        raise ValueError(
            f"Model {agent.model} has a context window of {_ctx:,} tokens, "
            f"which is below the minimum {MINIMUM_CONTEXT_LENGTH:,} required "
            f"by Hermes Agent.  Choose a model with at least "
            f"{MINIMUM_CONTEXT_LENGTH // 1000}K context.  If your server "
            f"reports a window smaller than the model's true window, set "
            f"model.context_length in config.yaml to the real value "
            f"(this must be at least {MINIMUM_CONTEXT_LENGTH // 1000}K)."
        )

    # Nous Hermes 3/4 are chat models, not tool-call-tuned. The interactive
    # CLI already warns via cli.py show_banner() (richer output + /model hint),
    # so skip platform=="cli" here to avoid emitting the warning twice per
    # startup. (Gateway/TUI/cron construct with quiet_mode=True and are already
    # gated off by the `not agent.quiet_mode` check above; this guard's active
    # job is the CLI dedup, and it leaves the door open for any non-quiet
    # non-CLI surface to still surface the warning.)
    if not agent.quiet_mode and (agent.platform or "cli") != "cli":
        try:

View on GitHub (pinned to c896c09c42)

Solutions

1. Switch to a model with at least 64K context via `hermes model`
2. If the server under-reports the true window, set model.context_length in config.yaml to the real value — the below-floor override is honored only for provider lmstudio with an explicit positive integer
3. Fix the local server's reported context so the resolved value is accurate
4. Remove/repair any model.context_length that coerces to 0 or False

Example fix

# config.yaml — before (provider: lmstudio, server reports 8K)
model:
  name: my-local-model

# after (explicit override of the server's under-report)
model:
  name: my-local-model
  context_length: 131072

from agent.constants import MINIMUM_CONTEXT_LENGTH  # 65536

def context_ok(resolved_context: int, provider: str, config_ctx) -> bool:
    if not resolved_context or resolved_context >= MINIMUM_CONTEXT_LENGTH:
        return True
    return (
        provider.strip().lower() == "lmstudio"
        and isinstance(config_ctx, int)
        and not isinstance(config_ctx, bool)
        and config_ctx > 0
    )

assert context_ok(resolved_ctx, provider, config_ctx), "model context below 64K floor"

try:
    agent = AIAgent(...)
except ValueError as e:
    if "context window" in str(e) and "below the minimum" in str(e):
        suggest_model_with_64k()  # or set model.context_length for lmstudio
    else:
        raise

When it happens

Trigger: The provider/model catalog (or server-reported value) resolves a context length < 65536 for the chosen model, and either the provider is not lmstudio, or it is lmstudio but model.context_length is missing, zero, or a bool — so _allow_lmstudio_explicit_below_floor is False.

Common situations: Pointing hermes at a small local model (8K/32K Ollama or LM Studio quant); a server under-reporting the model's true window; a stale catalog entry with a low context length; model.context_length accidentally set to 0.

Related errors

• Invalid profile name: ${value}
• SSH host is required.
• Malformed custom endpoint URL: {candidate!r}. Run `hermes se
• Auxiliary compression model {aux_model} has a context window
• iron-proxy rejected the new config (validation failed; the r

AI-assisted analysis of NousResearch/hermes-agent@c896c09c42 (2026-08-14). Data as JSON: /api/errors/9b0d6cc61c16180a. Report an issue: GitHub.