AXME

Durable execution for AI agents and services - workflow orchestration with built-in human approval gates, agent crash recovery, and multi-agent orchestration. No cluster required.

Submit work once. AXME drives it to completion through crashes, retries, timeouts, and human-in-the-loop steps. A managed Temporal alternative with durable execution without cluster setup, an agent kill switch, and an agent monitoring dashboard out of the box.

Quick Start · Docs · Examples · Spec

intent = client.send_intent({
    "intent_type": "intent.deployment.approval.v1",
    "to_agent": "agent://myorg/prod/deploy-checker",
    "payload": {"service": "api-gateway", "version": "3.2.1"},
})
result = client.wait_for(intent["id"])  # waits hours if needed. survives crashes.

Why AXME?

• Human approval gates that survive for hours - AI agents block forever waiting for human approval - no reminders, no escalation, session dies. AXME provides built-in human-in-the-loop task types that wait as long as needed.
• AI agent crash recovery - agent crashes mid-task - state is gone, start over from scratch. AXME preserves intent state across restarts so work resumes, not restarts.
• Webhook retry is everyone's problem - backoff, jitter, DLQ, HMAC, idempotency. AXME handles delivery guarantees so you don't.
• Multi-agent orchestration across machines - most frameworks only work in one process. AXME provides an agent coordination protocol that works across services, languages, and clouds.
• Temporal alternative without the overhead - Temporal is overkill for 80% of use cases - cluster, determinism constraints, no built-in HITL. AXME gives you durable execution without cluster setup.
• Agent monitoring dashboard and kill switch - agents in production with zero visibility - no health checks, no cost tracking, no way to stop a misbehaving agent. AXME includes a real-time agent monitoring dashboard and an agent kill switch out of the box.

Before and After

Without AXME - polling, webhooks, Redis, and glue code:

resp = requests.post("https://api.vendor.com/generate", json=payload)
job_id = resp.json()["job_id"]

for _ in range(120):                          # poll loop
    status = requests.get(f".../jobs/{job_id}").json()
    if status["state"] in ("completed", "failed"):
        break
    time.sleep(5)

@app.post("/webhooks/vendor")                 # webhook endpoint
def handle_webhook(req):
    redis.set(f"job:{req.json['job_id']}", req.json["result"])

result = redis.get(f"job:{job_id}")           # fetch from cache

With AXME - submit once, get result later:

from axme import AxmeClient, AxmeClientConfig

client = AxmeClient(AxmeClientConfig(api_key="axme_sa_..."))
intent = client.send_intent("agent://myorg/prod/generator", payload)
result = client.wait_for(intent["id"])        # retries, timeouts, delivery - handled

No polling. No webhooks. No Redis. No glue code.

Quick Start

# Install CLI
curl -fsSL https://raw.githubusercontent.com/AxmeAI/axme-cli/main/install.sh | sh

# Authenticate
axme login

# Run your first example
axme examples run human/cli

Full walkthrough: cloud.axme.ai/alpha/cli

Agent Monitoring Dashboard - See and Control Your Agents

Every agent you deploy gets a real-time agent monitoring dashboard, policy enforcement, and an agent kill switch.

Open the live dashboard at mesh.axme.ai, or from your terminal:

axme mesh dashboard

Dashboard - all agents on one screen with health, intents, and cost tracking (day/week/month).

Policies - restrict which intent types an agent can send or receive, set cost and rate limits, auto-block on breach.

Kill switch - isolate a misbehaving agent in one click. All intents blocked instantly. Reversible.

from axme import AxmeClient, AxmeClientConfig

client = AxmeClient(AxmeClientConfig(api_key="axme_sa_..."))
client.mesh.start_heartbeat()  # agent appears in dashboard with live health
client.mesh.report_metric(success=True, latency_ms=230, cost_usd=0.02)

AXME vs Temporal vs Inngest

What You Can Build

Backend teams - workflow orchestration for approval flows, long-running API coordination, replace polling and webhooks.

AI agent builders - human-in-the-loop approvals that wait hours not minutes, multi-agent orchestration with handoffs across services, AI agent crash recovery across restarts, framework-agnostic (LangGraph, CrewAI, AutoGen, raw Python).

Platform teams - one agent coordination protocol instead of webhook-polling-queue stack, a Temporal alternative that ships in minutes.

CREATED -> SUBMITTED -> DELIVERED -> ACKNOWLEDGED -> IN_PROGRESS -> WAITING -> COMPLETED
                                                                            \-> FAILED
                                                                            \-> CANCELLED
                                                                            \-> TIMED_OUT

from axme import AxmeClient, AxmeClientConfig

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

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

{
  "scenario_id": "human.cli.v1",
  "agents": [
    { "role": "checker", "address": "deploy-readiness-checker", "delivery_mode": "stream", "create_if_missing": true }
  ],
  "humans": [
    { "role": "operator", "display_name": "Operations Team" }
  ],
  "workflow": {
    "steps": [
      { "step_id": "readiness_check", "assigned_to": "checker" },
      { "step_id": "ops_approval", "assigned_to": "operator", "requires_approval": true }
    ]
  },
  "intent": {
    "type": "intent.deployment.approval.v1",
    "payload": { "service": "api-gateway", "version": "3.2.1" }
  }
}

axme scenarios apply scenario.json --watch

POST https://mcp.cloud.axme.ai/mcp
Authorization: Bearer <account_session_token>

{"jsonrpc": "2.0", "id": 1, "method": "tools/call",
 "params": {"name": "axme.intents_send", "arguments": {
   "to_agent": "agent://myorg/production/my-agent",
   "intent_type": "task.process.v1",
   "payload": {"data": "..."}
 }}}

Topics

durable-execution workflow-orchestration temporal-alternative human-in-the-loop human-approval-gates ai-agent-crash-recovery agent-kill-switch agent-monitoring-dashboard multi-agent-orchestration agent-coordination-protocol durable-execution-without-cluster ai-agents hitl intent-protocol agentic-workflows

contact@axme.ai · @axme_ai · Security · License