One native agent with three capability domains. It understands PALISEC's records, tools and workflows, and it works for your operators. It is not an authority: the deterministic engine is.
The PALISEC Agent is a controlled application principal with explicit capabilities, not a super-administrator. It retrieves only what a question needs from the authoritative records, through 30 typed tools, and every answer cites the records it used. Citations that no tool returned are flagged as unverified.
It runs in the console with any OpenAI-compatible model provider (the deployment uses Groq's free tier). Quick investigations run the same tools directly, without a model, so the agent's view of your records is available even when the model is not.
Understand and investigate activity across PALISEC.
Questions it answers
Turn intent into a canonical, validated, simulated policy draft.
Questions it answers
Make PALISEC easier to operate, integrate and troubleshoot.
Questions it answers
Every tool declares a strict input schema, a risk class (read, draft, action), the minimum autonomy level and the role it needs. The model is shown only the tools the signed-in member may use, and the server checks again on every call.
Read tools retrieve real records: security state, agents, policies and versions, operations and decisions, approvals, audit events, security findings and integration status. Draft tools build and simulate policies. Action tools only create requests.
PALISEC AGENT RUNTIME (isolated module in the API, never inside the policy engine) ├─ Orchestrator runs stored in Postgres; a few steps per request, within a time budget ├─ Provider adapter OpenAI-compatible chat completions (Groq by default; swap by configuration) ├─ Context builder only the records a question needs, minimized and redacted ├─ Tool registry 30 tools · schema · risk · autonomy level · role · timeout ├─ Permission resolver member role × organization autonomy × disabled tools, checked per call ├─ Confirmation action requests → a member confirms → existing control-plane function └─ Audit and tracing run → tool call → evidence → action → audit event
Read the full model in Security → agentic trust boundaries.