AgentRecover agent checkpoint auto-recovery

Automatically saves and restores multi-tenant agent state so you don't hand-write checkpointers and idempotency

Problem

LangGraph and similar frameworks let you save and load agent state, but devs must still write checkpointers, recovery logic, and handle idempotency themselves. As agent business logic explodes and serves multi-tenants, this becomes painful—and no product auto-adapts to it.

Solution

SDK that wraps agent nodes via decorators, automatically recording checkpoints, idempotency keys, and recovery policies; on failure it rolls back to the last stable node and retries. Multi-tenant isolation and an observability panel included.

Users

AI developers and platform teams building multi-tenant agent applications

MVP

Python/TS SDK to capture node entry/exit state automatically; built-in idempotency key generation and dedup; policy-based rollback to last checkpoint on failure; simple web panel to view run traces and recovery history

Revenue

Open-core + hosted cloud subscription (per tenant/run); enterprise tier with audit logs and SLA

Why now

Agents are moving from demo to production, making multi-tenancy and long-running recovery routine—yet everyone still rolls their own; a solo dev can focus on the SDK layer without competing head-on with platforms