Linchpin vs CrewAI

CrewAI is a Python library for multi-agent orchestration. You define agents with roles, goals, and backstories, give them tools, then compose them into a crew that runs through tasks sequentially or hierarchically. CrewAI is imported into your own Python service. Open source (MIT). There is also a hosted CrewAI offering for managed runs.

Linchpin is an open-source managed-agent runtime. Three services and Postgres on a single VM. Agents, environments, and sessions are first-class resources behind an HTTP API. Every session runs in its own Docker container with controlled networking. Output streams over SSE from an append-only event log. Language-agnostic; CrewAI runs in Python, but Linchpin's API is callable from any language. Apache-2.0.

Frame: CrewAI is code you import. Linchpin is infrastructure you deploy. They solve different problems.

• Your agent needs to execute code in a sandbox. Linchpin's per-session Docker container with bash + file tools is the right primitive for agents that do, not just orchestrate text.
• You want a runtime callable from any language. HTTP + SSE; your Go service or your TypeScript app can drive Linchpin without writing Python.
• You want an append-only event log streamable to clients. Audit, replay, multi-client UIs all become trivial.
• You want a deployment shape — a VM running services — not a library. Linchpin is infrastructure; you run it once, all your apps call it.
• You want a single, small codebase to read. ~3k LOC.

• You want role-based multi-agent orchestration as a first-class abstraction. CrewAI is built around the crew metaphor — roles, goals, backstory, hierarchical / sequential processes. Linchpin treats every session as independent.
• Your team is Python-first and wants tight in-process control. CrewAI is just an import; the agent loop runs inside your process.
• You want a broad ecosystem of LLM providers and tools. CrewAI builds on top of langchain-style integrations.
• You want a managed runtime via the vendor. CrewAI Enterprise is an option if you don't want to operate it yourself.

A reasonable architecture: define crews in CrewAI inside a Python service, and call Linchpin's HTTP API when an individual agent needs a sandboxed environment (run a shell, edit files, do anything that should not run in your service process). Linchpin becomes the "sandboxed task" primitive your crew dispatches to.

Or skip CrewAI: Linchpin sessions are independent. Run several at once, post events between them at the application layer, and you have multi-agent without the abstraction overhead. The right choice depends on whether the crew metaphor maps onto your problem.

• Linchpin vs LangGraph — runtime vs Python state-machine library
• Linchpin vs AutoGen — runtime vs Microsoft multi-agent library
• Linchpin vs OpenHands — runtime vs coding agent
• Use case: self-hosted AI agents