EventLog & Replay

Every run executed by antcrew-engine produces a structured EventLog — an append-only record of every capability dispatch, result, HITL checkpoint, and error.

What gets logged

graph TD
    A[EngineLoop.run called] --> B[EventLog: EngineStarted]
    B --> C[Capability selected]
    C --> D[EventLog: CapabilityDispatched]
    D --> E[Capability runs]
    E --> F[EventLog: CapabilityCompleted]
    F --> G{Goal satisfied?}
    G -->|no| C
    G -->|yes| H[EventLog: EngineFinished]
    E -->|error| I[EventLog: EngineError]

Event types

Event class	When emitted
EngineStarted	Loop begins
StateObserved	After each validator pass
EngineDecision	Capability selected by the decision policy
CapabilityDispatched	Before capability.run() is called
CapabilityCompleted	After capability.run() succeeds
ConditionSatisfied	A goal condition transitions to satisfied
CapabilityProgress	Mid-run progress update (token stream, partial output)
EngineFinished	All conditions satisfied — loop exits normally
EngineError	Loop exits with STUCK / TIMEOUT / NO_PROGRESS / CANCELLED

Collecting events in memory

from antcrew_engine.engine import EventLog

log = EventLog()
engine = EngineLoop(registry, validators, log)
engine.run(store, goal)

for event in log.events:
    print(event.kind, event.timestamp)

Streaming events to the platform

EventBusBridge forwards each event to antcrew-platform in real time so runs appear live in the dashboard:

from antcrew_engine.engine import EventBusBridge

bridge = EventBusBridge(
    platform_url="https://antcrew.org",
    api_key="acw_live_...",
    run_id="run_01HXYZ...",
)
engine = EngineLoop(registry, validators, bridge)

All events are visible in the run's Trace tab — one card per capability, with duration, cost, and produced artifact keys.

Trace tab in the dashboard

When a run is started via POST /engine/run, the platform assigns a run_id and the engine streams events to it automatically. The run detail page exposes a Trace tab that renders the EventLog as a human-readable timeline:

• One card per capability, in execution order
• Status indicator (running / done) with duration and cost
• Produced artifact keys shown as chips
• Live token stream while the capability is still running

For programmatic access to the same data, use GET /runs/{run_id}/events.

Per-agent model config

The model shown in the Trace tab reflects the resolved model after applying run.model_overrides → workspace.agent_models → platform default. See Model configuration to configure which model each capability uses.