Skip to main content
Orchestration lets you describe a goal in plain language and have Hexel plan and execute it across a group of agents. Instead of writing the control loop yourself (deciding which agent handles what, sequencing steps, retrying failures), you submit a task and the platform does the coordination. Use orchestration when work spans multiple agents or steps, such as triaging a batch of support tickets, running a multi-stage research pipeline, or processing documents with human approval gates. You submit a task to a fleet; the platform plans it, executes it across the fleet’s agents, gates it with approvals when needed, and streams progress you can follow and replay.

Capabilities

Tasks

Submit work and track it through its lifecycle.

Fleets

Group agents and configure how they run.

Routing & Modes

Control capacity, deployment, and routing.

Approvals

Require human or policy approval before execution.

Streaming & Replay

Follow execution live and reproduce it later.

How it works

A task is scoped to a fleet, workspace, and environment. The platform plans the work, optionally pauses for approval, executes across the fleet, and records everything for auditing and replay.

Access

Through the Python and TypeScript SDKs (client.orchestrator.*) or the REST API.

Rate limits

Pagination

List endpoints accept page (starting at 1) and page_size (1–100, default 20).

Errors

Orchestration endpoints return standard HTTP status codes with a JSON body describing the problem.
The SDKs retry 429 automatically with exponential backoff and surface the Retry-After value.

Security

Next steps

Start with Tasks to submit and track your first task.