AXME
Run tasks that take minutes, hours, or days — without polling or connection timeouts
Background jobs that run longer than 30 seconds hit HTTP timeouts. Workers that restart lose state. Polling for completion wastes resources and adds latency.
AXME Cloud's intent lifecycle handles background tasks that outlast HTTP connections, worker restarts, and human attention spans.
Your agent was on step 47 of 50 when the process died. Without durable state, you restart from scratch — or worse, you guess which steps completed. HTTP timeouts and job queues were not built for multi-day human delays.
Background work longer than a request timeout
Report generation, video processing, bulk enrichment, and human approvals over days cannot live inside an HTTP handler or a stateless lambda without glue. Job queues help but still need custom status APIs and webhook completion handlers per job type.
Durable intents model any long-running work as submit → wait → complete — surviving process restarts at step 47 and notifying clients when done.
Example: 45-minute analytics export
An agent kicks off a warehouse export that runs 45 minutes, then waits for analyst sign-off before publishing. HTTP times out at 30 seconds; a deploy mid-run used to force a full restart.
Submit intent, WAITING_FOR_TOOL on export job ID, then WAITING_FOR_HUMAN for sign-off — analyst's dashboard updates via SSE; Cloud resumes after crash without re-running completed steps.
Crash at step 47 → resume
Stateless worker
# process OOM at step 47 # restart from step 1 or manual reconcile
AXME intent
intent = axme.submit(pipeline) # Cloud resumes from last checkpoint
SOLUTION
How teams solve this with AXME.
10-minute reports
Outlive HTTP.
Multi-day pipelines
Survive restarts.
Notify on complete
Push or inbox delivery.
Design patterns
Use wait_for_tool for external job IDs; wait_for_time for scheduled windows; wait_for_human for approval before publishing results. Choose push or inbox delivery for completion callbacks to downstream systems.
Set intent-level timeouts and dead-letter handling for jobs that never complete.
Common questions
- Better than Celery or SQS?
- Queues are great for tasks; intents add HITL, multi-agent legs, and unified audit — often both coexist during migration.
- Multi-day workflows?
- Yes — waiting states persist for days; use explicit timeouts and escalation for SLA management.
- How do clients get notified?
- Delivery modes: push webhook, SSE stream, or poll — same intent completion event.
Related reading
Deeper dives from the AXME blog.
Your AI Agent Crashed at Step 47. Now What?
AI agents do multi-step work. They crash. The state is gone. Here's how to make agent operations durable without writing checkpoint code.
Read post →Your AI Agent Crashed at Step 47. Why Isn't Crash Recovery the Default?
Your agent ran 47 steps of a 50-step pipeline. Then it crashed. The state is gone. Every framework says you should have configured checkpointing. Why isn't durability built in?
Read post →
Related capabilities
Related links
Ship your first durable agent — in under 10 minutes.
Free tier. No credit card. Self-host or hosted — your choice.