Clutch
The operational layer where you manage, monitor, and govern everything your agents do.
It sits between your intent and your agents' execution — giving you a structured pipeline, a human approval gate, real-time visibility, and performance accountability across your entire fleet.
Mission Pipeline
Every piece of work moves through a structured 7-stage pipeline. Missions are created with a title, description, priority (low / medium / high / critical), and optional project assignment. Agents are assigned per mission. Nothing advances without meeting the criteria for that stage. The pipeline is not advisory — it is enforced. A mission in Review means the agent has finished and the human operator must sign off before it moves to Done.
Human Approval Gate
Before an agent executes a plan, it submits an approval request. Clutch surfaces these with full context. The operator reviews and approves or rejects. Approved requests unblock execution. Rejected requests stop it. Every decision is timestamped and attributed.
Budget & Spend Control
Clutch tracks LLM spend in real time against operator-set limits. Budget limits are editable from the UI. Spend is broken down per agent and per task.
Agent Fleet
Clutch maintains a live registry of every agent in the cluster. The fleet view shows what every agent is doing right now, what it did last, and who it delegated work to.
| Field | What it shows | Registry Scope |
|---|---|---|
| Tier | Top (Squad Lead) or Workhorse | Squad Lead / Workhorse |
| Role | Squad Lead / Engineer / Analyst / Executor / Monitor | 5 Role Tiers |
| Status | running / idle / busy / standby / offline / error | 6 Lifecycle States |
| Current task | Active task and next queued task | Live Queue Inspection |
| Last event | Seconds, minutes, or hours ago | Heartbeat Precision |
Performance Rankings
Every agent accumulates a composite scorecard built from six dimensions. Scores roll into a composite score and a credit score, which determines the agent's rating tier:
Tasks finished vs attempted
Output quality per token consumed
Flagged fabrications
Accuracy of agent self-estimates
Execution time vs baseline
Deliverable quality rating
Done-Guard
No mission reaches Done without a deliverable attached. This is enforced at the database level — not a UI suggestion. Agents must produce and attach output before a task can close.
Live Activity Feed
Every agent action emits a real-time event over WebSocket. The activity feed shows the full event stream — task state changes, delegation events, model calls, errors — with agent attribution and timestamps.
Outbox
Completed deliverables are collected in the Outbox and can be exported in two formats.
Security
All bot POST endpoints require HMAC-SHA256 signed webhooks
Access is gated via lvls post-quantum credential tokens
Deployed on Tailscale — not exposed to the public internet
Explore the full platform
Five integrated components. One unified system.