Skip to main content

​
Overview

The Draft SDK lets external AI agents create goals, generate tickets, monitor execution, and review results — all programmatically. Agents interact with Draft the same way a human would, but through code.

​
Install

pip install draft-sdk
Requires Python 3.10+ and a running Draft instance.

​
Quick Start

from draft_sdk import DraftClient

client = DraftClient("http://localhost:8000")

# One command: create goal → generate tickets → execute → wait
result = client.run_goal("Add dark mode support", auto_approve=True, wait=True)

print(f"Status: {result.status}")
print(f"Completed: {len(result.done_tickets)}/{len(result.tickets)}")
That’s it. The SDK handles the entire lifecycle:
  1. Creates a goal from your description
  2. Generates actionable tickets using AI
  3. Approves tickets automatically
  4. Executes each ticket in an isolated git worktree
  5. Runs verification (tests, lints)
  6. Auto-approves passing revisions
  7. Returns the final result

​
Agentic-First Design

Draft is built for agents. The SDK exposes the same API surface that the web UI uses, so agents are first-class citizens:

Resource-Based API

client.goals, client.tickets, client.jobs, client.revisions — mirrors the REST API structure.

Polling Built In

client.jobs.wait() and client.run_goal(wait=True) handle polling automatically with configurable timeouts.

Progress Callbacks

Pass on_progress to get real-time events as the pipeline progresses.

Full Control

Use the high-level run_goal() or go granular with individual resource methods.

​
Authentication

When Draft runs with auth enabled (AUTH_ENABLED=true), pass a JWT token: 
client = DraftClient("http://localhost:8000", token="your-jwt-token")
In development mode (default), no token is needed.

​
Connection

# Check connectivity
client.health()   # {"status": "ok"}
client.version()  # {"app": "Draft", "version": "0.1.0"}

​
Next Steps

API Reference

Full reference for all resource classes and methods.

Examples

Real-world integration patterns: CI/CD, LangChain, custom agents.