Product
Four surfaces. One coordinated runtime.
ADL is not a framework. It is production infrastructure — an orchestrator, an SDK, a runtime, and an observability plane, engineered to run agent fleets at enterprise scale.
01
Orchestrator
Declarative workflow graphs with typed handoffs. Compose sequential, parallel, and conditional agent execution with rollback guarantees.
- DAG scheduler
- Typed handoff contracts
- Idempotent retries
- Rollback checkpoints
02
Agent SDK
Author agents in TypeScript or Python. Bring your own model — OpenAI, Anthropic, open-weight — behind a uniform tool-calling interface.
- TS + Python SDKs
- Model-agnostic runtime
- First-class MCP support
- Streaming tool use
03
Runtime
Deploy to a managed edge or your own VPC. Autoscales per-agent, budgets tokens per workflow, and hard-caps runaway cost.
- Managed or BYOC
- Per-agent autoscale
- Token budgets
- SOC 2 Type II
04
Observability
Every handoff, prompt, and tool call is a first-class trace. Replay any run, diff prompts, and attribute cost to the exact node.
- OpenTelemetry native
- Prompt-level diffing
- Cost attribution
- Run replay