Platform¶
The antcrew platform is the managed cloud service your team uses to observe, store, and gate AI agent workflows. It does not execute agent code — it receives runs from the engine, stores their full history, and surfaces review queues to your team.
Access it at antcrew.org.
What it manages¶
Runs — every agent execution started via the engine is a run. The platform stores its full history: status transitions, TraceLog events, generated tickets, and any HITL reviews it triggered.
Tickets — structured action items extracted from run output. Each workspace has a configurable prefix (PROJ, ENG, BUG…). Tickets get sequential display IDs — PROJ-00001, PROJ-00002 — and link back to the run that created them.
HITL reviews — when an agent calls hitl_checkpoint(), the engine sends a review request to the platform. The platform queues it, notifies the right reviewer, and holds the resume signal until they approve or reject.
Workspaces — isolated tenants. Each workspace has its own API keys, members, ticket prefix, provider keys, and run history. Nothing leaks between workspaces.
Evals — automated quality checks that run against past runs on a schedule or on demand.
Webhooks — outbound notifications to your own endpoints when runs complete, tickets are created, or reviews are resolved.
Real-time dashboard¶
The dashboard subscribes to live events and updates without polling. You see:
- Current run status and elapsed time
- TraceLog events as they stream in — each LLM call, tool invocation, intermediate result
- Tickets as they are created
- HITL review cards with the full context the agent passed in
Data model¶
erDiagram
Workspace ||--o{ Run : "contains"
Workspace ||--o{ ApiKey : "has"
Run ||--o{ Event : "logs"
Run ||--o{ Ticket : "generates"
Run ||--o{ Review : "gates on"
Review }o--|| Ticket : "may reference"Key concepts¶
| Concept | Description |
|---|---|
| Workspace | Isolated tenant — own keys, members, tickets, runs |
| Run | One agent pipeline execution — has status, events, tickets |
| Event | A single TraceLog entry — LLM call, tool result, state change |
| Ticket | Structured action item from a run, with a display ID |
| HITL Review | Human approval gate that pauses a run until resolved |
| Eval | Automated quality check over run output |
| Webhook | Outbound HTTP call to your system on any platform event |