AGENTIC COMMAND OS

consequence clearing for autonomous AI systems — a premise changes, the fleet is planned, delegated, challenged, governed and corrected, and every step leaves evidence

JUDGE DEMO Sandboxed · Deterministic · Evidence-backed

No credential required. Replays a real, captured, zero-model mission trace — recon through governance through knowledge distillation — computed once by the exact same pipeline "Run autonomous mission" below uses, so a judge sees the full chain immediately with nothing to configure. No token is requested or sent by this button, and nothing it does can mutate real state.

OPERATOR MODE 🔒 credential required — run a LIVE mission under your own identity
change this and the PLAN changes — the specialists, tools and action kinds are computed from the objective, not fixed

Google model stack

every model string this system pins, joined to what a real call actually did — model IDs from lib/config.py, statuses from evidence/models/verification-*.json, never from a hardcoded list on this page. A model that has not been exercised says so.

Mission Media Lab — one mission, three modalities

all three read the same grounded brief, built from the checkpoints this mission actually persisted (GET /api/media/mission/{id}/brief returns the exact input, so it can be checked against the checkpoints themselves). Each card plays a committed render immediately and can also make a live call when a credential is present.

    Agent Fleet

    six identities with their own principals and bounded scope — served from the same constants the registry enforces. Verify it yourself: GET /api/architecture/proof

    Warrant Market — live pricing

    move a signal, watch every price change. Computed in warrant/economics.py, which cannot reach a model

    Consequence preview — what breaks if this executes

    the question this product is named after, asked of an AGENT ACTION. Every number below is a real traversal of the committed reverse index (spine/cascade.py) over the premises the recon agent actually parsed — zero model calls. Drive it yourself: GET /api/command-os/consequence-preview?subject=supplier_K&predicate=lead_time_days&value=20

    System Reality

    every feature below carries its own honest status — nothing here claims to be live that is not

    Mission Time Machine

    historical mission state, reconstructed from real checkpoints — not a literal reversal of time, a replay of what was actually persisted. Loads with this page; the most recent mission opens on its own.

    Mission arc — past to current trusted state

      Missions

        Checkpoints

          every stage above calls a real engine one layer down · every status label is served by /api/command-os/status, a second independently queryable source that must agree with this page or this page is wrong