███╗   ███╗ ██╗   ██╗  ██████╗ ███████╗ ██╗
████╗ ████║ ╚██╗ ██╔╝ ██╔════╝ ██╔════╝ ██║
██╔████╔██║  ╚████╔╝  ██║      █████╗   ██║
██║╚██╔╝██║   ╚██╔╝   ██║      ██╔══╝   ██║
██║ ╚═╝ ██║    ██║    ╚██████╗ ███████╗ ███████╗
╚═╝     ╚═╝    ╚═╝     ╚═════╝ ╚══════╝ ╚══════╝
Star
comparison

Mycel vs LangGraph

A framework for building agents, versus a harness for running them as a business.

What LangGraph is good at

LangGraph is a genuinely good way to express an agent's control flow — cycles, branches, state machines, human-in-the-loop interrupts. If you're designing how an agent thinks, it gives you the vocabulary.

The actual difference

LangGraph is a library you build with. Mycel is a service you run on. The graph is not the hard part of an AI service business — tenancy, credential custody, an audit a client's accountant would accept, scheduling, and a customer portal are. Mycel takes an opinion on all of those and comparatively few opinions about how your agent reasons; it drives a coding agent in a sandbox rather than composing nodes.

Side by side

LangGraphMycel
ShapeLibraryRuntime + product
Multi-tenancyYou build itOrg → project → member, tested
AuditYou build itHash-chained, verifiable
Customer portalYou build itIncluded, separate credential plane
SchedulingYou build itCadences + a Postgres queue
Agent control flowRich and explicitDelegated to a coding agent

Choose LangGraph

  • The agent's reasoning structure is the interesting problem
  • You're embedding an agent in an existing product
  • You want full control of the graph

Choose Mycel

  • The interesting problem is running a business, not the control flow
  • You need multi-tenancy, audit and approvals without building them
  • Your customers need somewhere to log in

Other comparisons: Temporal, Zapier & n8n, CrewAI & AutoGen, Building it yourself

Read the kernel Docs