Ω-API Shadow
See what your agents are actually doing.
Shadow shows whether your agents are moving the process — before you ask Operator to act.
Your fleet is not one conversation. Shadow grades how each agent moves the process.
Trace tools show what ran. Semantic evals score what was said. Shadow measures whether the process moved.
Not another LLM judging an LLM.
Ω-API is not another LLM judging an LLM. It adds a third vertex: a geometric judge of process movement.
Shadow does not score what the words mean. It measures what each turn does to the process.
What you get
Process fingerprint. every conversation as a measured shape · e.g. C3·S2.
C / S / M trajectory. converging, holding, or moving away · turn by turn.
Agent grade A–I. each agent ranked by how its process moves.
Regime mix. the reads the desk actually shows · moves toward completion, keeps circling, circles — sometimes progresses, gets stuck or drifts away, too early to read.
CSV / JSON export. fields like csm_shape_rle, decision, d_last, estimated_cost_usd · ready to join to your own agent or chatbot log.
Shadow shows which agents move, hold, or drift. Operator can act on the same signal — by asking the agent to realign or stop before the response is released.
Zero configuration
One call per turn. No workflow redesign. The shape is computed for you, Ω-side.
Where it is used
Support bots
Catch conversations that keep circling instead of resolving, and grade which bots close and which stall.
RAG assistants
See retrieval drift as the process moves away from the objective, turn after turn.
Sales qualification
Know whether the conversation is progressing toward a decision or just holding.
Internal copilots
Surface prompting friction and context rot as repeated holding or drift.
Long-running agents
Spot runaway loops before tokens keep burning.
Agent handoffs
Detect where the process cannot close and should move to a human.
Bot↔bot workflows
Find coordination waste: one side moves, the other only holds.
Swarm monitoring
Grade agents across the fleet and see which ones drift.
Your first 100 judgments are free.
If the signal matters, you will know before the first invoice.
The geometry behind the judge
Shadow is based on Sub-Limit Dynamics, a recently formalized geometric framework for reading movement under finite constraints.
You do not need to accept the framework to use the signal: every judgment produces a replayable process shape, and the Gate Zero experiment can be inspected directly.