Execution
Runtime
Agent orchestration
Declarative team definition. Parallel + sequential, fan-out + gather, branch on outcome.
Long-running execution
Runs survive restarts. Agents resume mid-flight. Marathons measured in hours, not seconds.
Scheduling & triggers
Cron, webhooks, queue events. Continuous loops or one-shot — both first-class.
Human-in-the-loop
Approval gates. Inline edits. Rollback policies. Reviewer-chops on risky actions.
State
Data & Memory
Agent memory
Persistent, per-tenant memory via pgvector + AstraMemory. Recall, consolidate, share.
Durable run history
Every run, every span, every input — replayable, immutable, exportable to your warehouse.
Tool registry
Filesystem, git, shell, HTTP, MCP, custom — uniform auth + audit envelope.
Multi-tenant isolation
Row-level Postgres security. Separate budgets, retention, secrets per tenant.
Insight
Telemetry
Distributed traces
OTel-compatible hierarchical spans for every agent run.
Real-time metrics
Latency, throughput, success rate, cost — sliced any way you need.
Reasoning replay
See what the agent thought, why it picked a tool, why it failed.
Eval validation
Prompt versioning and behaviour validation catch regressions before a crew ships.
Reach
Integrations
REST + MCP API
Everything is an endpoint. SDKs for TypeScript, Python, .NET. Auto-generated OpenAPI.
Webhook fan-out
Notify Slack, Linear, Jira, your own services on any run event.
Chat surfaces
Embed crews in Slack, Teams, Discord, or your in-app chat without writing glue.
SSO + SAML
Okta, Azure AD, Google Workspace, custom OIDC. SCIM-provisioning ready.
Architecture
Composable from the wire up.
Stable, boring primitives at every layer. Replace any single block with your own if you'd rather.
See the runtime in action.
Join the production open beta for the runner, memory, tools, observability, and approval controls.