axme-sdk-python

Python SDK for AXME - send intents, listen for deliveries, resume workflows. Durable execution without polling or webhooks.

Quick Start · Docs · Examples

Install

pip install axme

Requires Python 3.11+.

Quick Start

from axme import AxmeClient, AxmeClientConfig

client = AxmeClient(AxmeClientConfig(api_key="axme_sa_..."))

# Send an intent - survives crashes, retries, timeouts
intent = client.create_intent({
    "intent_type": "order.fulfillment.v1",
    "to_agent": "agent://myorg/production/fulfillment-service",
    "payload": {"order_id": "ord_123"},
}, idempotency_key="fulfill-ord-123-001")

# Wait for resolution - blocks until done, even across process restarts
result = client.wait_for(intent["intent_id"])
print(result["status"])

Connect an Agent

for delivery in client.listen("agent://myorg/production/my-agent"):
    intent = client.get_intent(delivery["intent_id"])
    result = process(intent["payload"])
    client.resume_intent(delivery["intent_id"], result)

Human Approvals

intent = client.create_intent({
    "intent_type": "intent.budget.approval.v1",
    "to_agent": "agent://myorg/prod/agent_core",
    "payload": {"amount": 32000},
    "human_task": {
        "task_type": "approval",
        "notify_email": "approver@example.com",
        "allowed_outcomes": ["approved", "rejected"],
    },
})
result = client.wait_for(intent["intent_id"])  # waits until human acts

8 task types: approval, confirmation, review, assignment, form, clarification, manual_action, override. Full reference: axme-docs.

Observe Lifecycle Events

for event in client.observe(intent["intent_id"]):
    print(event["event_type"], event["status"])
    if event["status"] in {"RESOLVED", "CANCELLED", "EXPIRED"}:
        break

Agent Mesh - Monitor and Govern

# Start heartbeat - agent appears in dashboard with live health
client.mesh.start_heartbeat()  # background thread, every 30s

# Report metrics after each task
client.mesh.report_metric(success=True, latency_ms=230, cost_usd=0.02)

# List all agents with health status
agents = client.mesh.list_agents()

# Kill a misbehaving agent - blocks all intents instantly
client.mesh.kill(address_id="addr_...")

# Resume it
client.mesh.resume(address_id="addr_...")

Set action policies (allowlist/denylist intent types) and cost policies (intents/day, $/day limits) per agent via dashboard or API. Agent Mesh overview.

Open the live dashboard at mesh.axme.ai or run axme mesh dashboard from the CLI.

Examples

export AXME_API_KEY="axme_sa_..."
python examples/basic_submit.py

More: axme-examples

Development

python -m pip install -e ".[dev]"
pytest

Related

hello@axme.ai · Security · License