autonai

Official Python SDK for AUTON — the operating system for AI-augmented companies. Full docs at docs.getauton.ai.

pip install autonai

Requires Python 3.9+.

Quickstart

import os
from autonai import AutonClient

client = AutonClient(
    base_url="https://api.getauton.ai",
    api_key=os.environ["AUTON_API_KEY"],  # Settings → API keys → auton_pat_…
)

# 1. Spawn a run
run = client.runs.spawn({
    "agent_id": "…",
    "company_id": "…",
    "area_id": "…",
    "actor_user_id": "…",
    "domain_config": {"template_id": "marketing.performance_marketing_manager"},
    "initial_messages": [
        {"role": "user", "content": "Pull last week's perf data and draft a Slack update"},
    ],
})

# 2. Stream deltas (generator; stops at the terminal delta)
for delta in client.runs.stream(run["run_id"]):
    print(delta["seq"], delta["kind"])

# 3. Triage approvals
inbox = client.approvals.list(status="pending")
client.approvals.decide(inbox[0]["id"], verdict="approve", by="…user uuid…", reason="Looks good")

More examples

# Knowledge: upload → search → cite
client.knowledge.upload(
    company_id=company_id,
    title="Brand book",
    source="upload",
    body="Voice: warm, direct, never jargon-heavy…",
)
hits = client.knowledge.search(company_id=company_id, query="brand voice")["hits"]

# Billing: this month's spend vs plan cap
usage = client.billing.usage("this_month")
print(usage["totals"]["cost_usd_micros"] / 1e6, "USD spent")

# Agent manuals: resolve with user preferences inlined
manual = client.agent_manuals.resolve(
    "marketing.performance_marketing_manager",
    company_id=company_id,
    user_preferences={"manual_tone": "concise", "preferred_locale": "pt-BR"},
)

# Heartbeats: a weekday 9am curator tick
hb = client.heartbeats.upsert(**heartbeat_fields)
client.heartbeats.fire_now(hb["id"])

Behaviour

• Auth — Authorization: Bearer <api_key> on every call.
• Retries — idempotent GETs retry on 5xx/network (max 3, exponential backoff). Non-idempotent calls never retry unless you pass idempotency_key=.
• Rate limits — Retry-After on 429 is honoured automatically; the final failure raises RateLimitError with .retry_after.
• Errors — typed: ValidationError, AuthenticationError, PermissionError_, NotFoundError, ConflictError, RateLimitError, ServerError — all subclassing AutonError with .status / .request_id.
• Self-hosting — pass orchestrator_url= / knowledge_url= / approvals_url= when those services are exposed on their own origins.

Versioning follows the gateway major: autonai==1.x ↔ gateway 1.x.