AI & agent behavior

What it can help with

Evaluate assistant workflows with explicit behavior, provider and review boundaries.

These are potential pilot applications, not claims of customer outcomes.

What the source implements

The local completion function selects a profile from keyword signals and renders an in-process persona template. The requested model identifier is metadata on this route, not proof that the named model ran. A hard constitutional veto substitutes a refusal in the returned local output.

Available today

Local deterministic routing and completion are implemented; model quality requires separate evaluation.

How to evaluate it

For a customer pilot, choose a concrete task and identify whether it uses local templates or an explicitly configured provider. Evaluate task correctness, escalation and data handling with representative inputs before making quality claims.

Supporting research and guides

Read the accompanying archive entries: Policy-Locked Tool Use in Multi-Agent Systems; Human Approval Gates That Operators Actually Use; Offline-First AI Workflows for Sensitive Data; Prompt Injection Drills for Agentic Systems; Multi-Persona Routing Under Uncertainty: Policy Weights and Request Classification. Each distinguishes implemented behavior from proposed applications and includes pinned implementation evidence.