Cellaflow
Cellaflow is a runtime that makes retries safe for AI agent tool calls by focusing on side effects rather than just agent state: it tracks which agent is allowed to perform an operation, what has already been done, and how to recover safely when an agent crashes or disappears mid-execution. It sits underneath existing agent frameworks (LangGraph, CrewAI, AutoGen, or custom agent code) without requiring the agent itself to be migrated.
What it addresses
Recovering an agent's state after a crash is the easy half, but recovering after the agent has already performed an irreversible operation is hard. Without idempotent downstream APIs, retries can repeat side effects like sending emails, creating tickets, triggering webhooks, provisioning resources, or mutating another system.
Shared over time
Each entry reflects what was shared on that date. Details may have changed since then.
Cellaflow
Cellaflow is a runtime that makes retries safe for AI agent tool calls by focusing on side effects rather than just agent state: it tracks which agent is allowed to perform an operation, what has already been done, and how to recover safely when an agent crashes or disappears mid-execution. It sits underneath existing agent frameworks (LangGraph, CrewAI, AutoGen, or custom agent code) without requiring the agent itself to be migrated.
Problem: Recovering an agent's state after a crash is the easy half, but recovering after the agent has already performed an irreversible operation is hard. Without idempotent downstream APIs, retries can repeat side effects like sending emails, creating tickets, triggering webhooks, provisioning resources, or mutating another system.
Audience: Talks to people running AI agents in production who have hit duplicate tool calls, duplicate side effects, or workflows that quietly never finished.
Posted by @druhinbala · Original comment ↗
“Cellaflow — Safe retries for AI agent tool calls.”