machine-sdk

Python SDK for the Machine AI platform — projects, agent sessions, and Machine Apps, driven from Python or the bundled machine-cli.

pip install machine-sdk

Authentication

Create a personal access token in the Machine app (Settings → API keys), then:

export MACHINE_API_KEY=kortix_pat_...
# self-hosted deployments only:
export MACHINE_API_URL=https://your-machine-api.example.com

Python

from machine import MachineClient

client = MachineClient()  # reads MACHINE_API_KEY / MACHINE_API_URL

for project in client.list_projects():
    print(project["project_id"], project.get("name"))

# Start an agent working on a prompt
session = client.create_session(project_id, prompt="Audit the repo's README files")
print(session["session_id"], session["status"])

# Poll it
session = client.get_session(project_id, session["session_id"])

# Stop it
client.stop_session(project_id, session["session_id"])

CLI

machine-cli projects
machine-cli sessions <project-id>
machine-cli new <project-id> "Fix the failing tests" --model anthropic/claude-sonnet-5
machine-cli session <project-id> <session-id>
machine-cli stop <project-id> <session-id>
machine-cli apps <project-id>

Output is JSON, so it pipes cleanly into jq.

Surface

Versioning

0.2.x is a ground-up rewrite against the current Machine API. The 0.1.x releases targeted a retired API surface and additionally shipped without any Python code due to a packaging defect — do not use them.