Intelligence engine

Run governed natural-language analytics using facts-grounded responses.

Last updated:

The Intelligence Engine follows one strict split:

  • Facts = truth (deterministic operational data)
  • Model = interpretation (summaries, prioritisation, next-step guidance)

For the full current end-user capability list in Assistant, see Eden capabilities.

User-facing access

All Intelligence access is session-authenticated in ChannelWeave.

Naming split

The product terminology is intentionally split:

  • Assistant names the workspace/surface in the app.
  • Eden names the model/persona that responds to the user.
  • chat remains the internal transport and persistence term in routes, API contracts, and storage.

This keeps user-facing copy clear without introducing unnecessary contract churn in the intelligence stack.

Request lifecycle (high level)

  1. Validate request shape and bounds.
  2. Resolve tenant context from the signed-in workspace.
  3. Gather deterministic facts for the question.
  4. Return either:
    • a direct fact answer (for explicit factual prompts), or
    • a model-composed answer grounded in those facts.
  5. Persist audit metadata for traceability.

Response behaviour

  • If required facts are unavailable, Eden returns an explicit uncertainty note.
  • If model generation fails, Eden returns a safe deterministic fallback where possible.
  • Assistant links to in-app pages are rendered as safe clickable links.
  • User-facing wording is normalised to British English.
  • Custom insight rule questions now use saved rule state, finding history, and template-specific evidence through safe facts routes.

Data safety

  • Tenant scoping is enforced for all facts used in responses.
  • The assistant does not execute arbitrary user-authored SQL.
  • Sensitive operational setup details are documented in SuperAdmin docs only.

Local startup order

On the local development machine, Eden runs as its own launchd job:

  • com.graham.channelweave-eden starts the intelligence engine from intelligence-engine.
  • com.graham.channelweave starts the main app through a small wrapper that waits for http://127.0.0.1:8787/health to report healthy.

This keeps Assistant startup deterministic: ChannelWeave does not serve the app until Eden is reachable and healthy.

Super admins can monitor both launchd services from Platform → Runtime logs. The screen reads only the allowlisted ChannelWeave/Eden stdout and stderr log files.