Checkpoint
Save portable execution state when work may outlive a process, model session, container, or machine.
A research agent saves its sources, progress and next step before its session ends.
Explore checkpoint.save, checkpoint.get, handoff.create, lock.acquire and recovery.register, with real API paths and concrete use cases.
Save portable execution state when work may outlive a process, model session, container, or machine.
A research agent saves its sources, progress and next step before its session ends.
Retrieve a live checkpoint owned by your wallet and continue from the state your application saved.
A new session loads the same task state and resumes without repeating completed work.
Delegate an encrypted snapshot to one named recipient wallet. Its lifetime cannot exceed the source checkpoint.
A planning agent passes a structured work package to a separate implementation agent.
Coordinate access to a global logical key so multiple agents do not simultaneously act on the same resource.
Two workers compete for a task key before processing the same queue item. Your application respects the lock result.
Requires setup with an approved callback destination. Register heartbeats so your system can receive a signed notification when an agent stops responding.
An orchestrator receives a recovery notification after an agent misses its expected heartbeat window.
Combine these primitives in your own agent framework. MantisMesh does not run your model, launch a replacement agent, or automatically discover a successor.
GET /v1/handoffs/{handoffId}DELETE /v1/locks/{key}POST /v1/recoveries/{recoveryId}/heartbeatAll protected operations require a bearer session and idempotency key. Locks have no fencing token or renewal protocol. Recovery callbacks require operator-approved origins and are at-most-once attempts.
The receiving agent loads the snapshot into its own runtime. MantisMesh carries state; your application carries execution.
Start with the wallet-authenticated REST API.