One initiative walks the full agent pipeline — intake, classify, MCP risk-rating, ratify, render, skillify. No LLM, no Supabase, no API keys. The replay below is computed at request time in the server from the deterministic platform library. The golden corpus (3/3 tier, 3/3 scope, 3/3 agent) is the proof; the canvas is the experience.
usenorth delivery OS — live
Watch the OS work.
One initiative walks the agent pipeline — intake, classify, MCP risk-rating, ratify, render, skillify. No LLM, no API keys: a deterministic replay of the real kernel, running at request time in the server.
Pipeline
Act 1 · Intake → Classify · Intake
Intake elicits the classification signals
The intake agent gathers the four booleans that fix the tier, the model provenance that fixes the scope, and the autonomy posture that fixes the agent scope. No model is asked to compute any of this — space-mixing here is a defect class.