Replay any agent run.
Track every dollar.
Full observability across the entire agent lifecycle — from the first LLM call to the final tool invocation. Time-travel debugging, real-time cost tracking, immutable audit trails, and configurable alerts.
Step through every decision your agent made.
Every agent execution is recorded as an ordered timeline of events. Click any step to inspect the full context: model used, tokens consumed, tool parameters, response payloads, and cost at each stage.
- Step-by-step execution replay
- Per-step token and cost breakdown
- Tool call parameters and responses
- Shareable run permalinks
- Compare runs side-by-side
from curate_me import Observer
# Stream execution events in real time
observer = Observer(api_key="cm_sk_xxx")
async for event in observer.stream(runner_id="r_abc"):
print(event.type, event.data)
# Output:
# runner.task_received {'agent': 'pr-reviewer'}
# llm.request {'model': 'claude-4', 'tokens': 2847}
# tool.invoked {'tool': 'github_api'}
# runner.completed {'cost': 0.0123}Stream execution events to your own systems.
The Observer SDK connects via WebSocket and delivers every lifecycle event in real time. Pipe events to your own logging, monitoring, or alerting infrastructure. Available in Python and TypeScript.
- WebSocket real-time streaming
- Python and TypeScript SDKs
- Filter by event type or agent
- Structured event payloads
- Automatic reconnection
Know exactly where every cent goes.
Real-time cost tracking broken down by model, agent, org, and time period. Set daily budgets and get alerts before you overspend. Costs are accumulated in Redis for sub-second updates and persisted to MongoDB for full auditability.
- Per-model cost breakdown
- Daily, weekly, and monthly views
- Per-org budget caps
- Cost-per-agent analytics
- Export to CSV or API
Immutable logs. Secrets auto-redacted.
Every API key validation, governance check, PII detection, cost recording, and runner lifecycle event is logged to an append-only audit trail. Secrets and PII are automatically redacted before storage. Queryable, exportable, and retention-configurable per org.
- Append-only, tamper-resistant log
- Automatic PII and secret redaction
- Full-text search across events
- Configurable retention periods
- Export for compliance reporting
Get notified before problems become incidents.
Define rules based on cost thresholds, error rates, latency percentiles, or security events. Route alerts to Slack, PagerDuty, email, or webhooks. Alerts evaluate continuously against real-time metrics.
- Cost threshold alerts
- Error rate anomaly detection
- Latency percentile breaches
- PII leak notifications
- Slack, PagerDuty, email, webhooks
Stop guessing.
Start observing.
Time-travel debugging, real-time cost tracking, and immutable audit trails — all included with every plan.