compressToolSchemaWithParams includes param names so the LLM knows what to pass. This prevents "missing required params" errors on first call.

Build a context string from past experience, ready to inject into an LLM system prompt. This is the key integration point: A

Explain WHY memories were recalled. Shows all scoring signals. For debugging and observability — developers can see exactly w

Forget a specific memory by ID.

Remove all memories in a namespace.

Ingest from ARK Runtime's governor registry JSON. Extracts model performance stats and tool success rates.

Ingest from ARK Runtime's ark-memory.json. Extracts tool usage stats and learned patterns.

Ingest from ARK Runtime's router learning JSON. Extracts learned routing rules.

scoreModelFor computes a composite score for how well a model handles a specific context.

Close flushes and closes the event file.

EmitTaskComplete records a full task execution.

EmitVerification records a code verification result.

FormatReport generates a human-readable report of all model capabilities.

FormatResult returns a human-readable verification result.

NewAgentWithGovernor creates an agent with the cognitive governor enabled.

NewEventEmitter creates an emitter that appends to the given file.

NewGovernorForRuntime creates a Governor that can be passed to runtime.NewAgentWithGovernor.

Compute word-level Jaccard similarity between two texts.

Remove oldest low-importance memories if over limit.

Initialize SQLite database and create tables if needed.

Route an event to the appropriate experience recorder.

Boost accessed memories, decay non-accessed.