CONCEPT

The mycel Method

Practices for orchestrating AI agent teams.

01

Isolation

Without isolation, two agents editing the same file will overwrite each other. Each agent gets its own git worktree, so 10 agents can work on the same codebase with zero merge conflicts.

$ mycel agent create eng-01 --role engineer
02

Communication

Agents need to coordinate without you as the bottleneck. Persistent channels with @mentions let agents hand off work, request reviews, and converge — all logged and searchable.

$ mycel channel send eng "starting tests"
03

Visibility

You can't trust what you can't see. The web dashboard at localhost:9374 shows every agent's state, output, costs, and channel messages in real time.

$ mycel status
04

Control

AI tokens add up fast. Per-agent budgets with hard stops prevent surprise bills. Role permissions scope what each agent can do, so you delegate without losing control.

$ mycel cost budget set eng-01 --limit $5.00
05

Persistence

Agents crash. Machines restart. All state lives in .mycel/ — worktrees, channel history, cost records, and memory survive restarts and are backed by git.

$ mycel agent start eng-01 --resume
06

Simplicity

No runtime dependencies. No cloud accounts. No YAML config files. One Go binary, two commands, and your agent team is running.

$ mycel init && mycel up