Skip to content
Apkar
ع

Documentation

Build your first agent.

From a single step to a durable graph with retries, budgets, and an audit trail.

Getting started

QuickstartA working agent in about ten minutes.Core conceptsRuns, steps, checkpoints, and why the distinction matters.

Building agents

Defining agentsThe builder API, and what compiles to the same graph.Retry policiesBackoff, caps, and per-tool overrides in one file.Approval gatesPausing a run for a human without losing its state.

Running in production

Budgets and cost guardsCeilings that count tokens, tool credits, and retries together.ObservabilityTraces, spans, and getting run latency next to service latency.Audit exportProducing evidence an auditor can read.

Reference

SDK referenceEvery method on the agent builder.Error codesWhat each failure means and what to do about it.