Little Canary for Hermes Agent

Little Canary screens Hermes Agent user turns with structural checks and a local powerless canary model. A BLOCK verdict withdraws downstream tool access for that turn. The original prompt still reaches Hermes's model.

With Hermes Agent 0.21.3 or later, local Ollama, and qwen2.5:1.5b:

ollama pull qwen2.5:1.5b
hermes plugins install little-canary
hermes plugins enable little-canary
hermes plugins list

The catalog entry installs this directory at a reviewed commit. The directory bundles Little Canary's runtime and compatible requests dependency. It runs inside Hermes; it does not require the separate little-canary serve adapter.

For a direct install before the catalog entry reaches your client:

hermes plugins install hermes-labs-ai/little-canary/integrations/hermes-agent --no-enable
hermes plugins enable little-canary

Set LITTLE_CANARY_MODEL in the Hermes Agent process environment to an installed Ollama tag to select the sacrificial model, for example LITTLE_CANARY_MODEL=gemma3:1b. The default remains qwen2.5:1.5b. The model receives no tools; selecting a model does not grant it any host authority.

The pre_llm_call hook screens the current user message once and can add a bounded warning. It cannot prevent the prompt from reaching the model. On a genuine BLOCK, pre_tool_call refuses downstream tools for that turn. If screening is unavailable, the plugin leaves tools available and reports degraded coverage. on_session_end clears the session's stored decisions.

For setup, verification and behavior, read the Little Canary Hermes Agent guide.