One file to agentize
your Claude project.

Drop an Aidefile into any Claude Code project.
Budget, vault, hooks, triggers. Fire and forget.

v2.0.0-alpha

$ cargo install aide-sh

What is an Aidefile?

Aidefile
[persona]
name = "Senior Reviewer"
style = "direct, cares about edge cases"

[budget]
tokens = "100k"
max_retries = 3

[hooks]
on_spawn = ["inject-vault"]
on_complete = ["commit-memory"]

[trigger]
on = "issue"

[vault]
keys = ["GITHUB_TOKEN"]

That's it. Your Claude project is now an agent.

How it works

terminal
# Turn any Claude project into an agent
$ cd ~/projects/code-reviewer
$ aide init --persona "Senior Reviewer"
✓ Created Aidefile

# One-shot: fire and forget
$ aide run . "Review PR #42 and leave comments"
▸ Running task in ~/projects/code-reviewer
  agent: Senior Reviewer
  budget: 100000 tokens
✓ Task completed (23,847 tokens used)

# Or register and run by name
$ aide register . --name reviewer
✓ Registered 'reviewer'

$ aide run reviewer "Review all open PRs"
✓ Task completed (67,231 tokens used)

# Start daemon: agents wake up on triggers
$ aide up
▸ Daemon started (polling 1 agent)

What aide handles

Budget

Token limits per task. Auto-retry with remaining budget. Never burn more than you set.

Vault

age-encrypted secrets injected as env vars at spawn time. Never enters the LLM context window.

Triggers

GitHub Issues, cron, webhooks. Daemon polls and dispatches. Agents wake up when needed.

Hooks

on_spawn, on_complete. Inject vault, commit memory, notify, clean up. Lifecycle control.

Memory

Auto-compact when threshold is hit. Each agent has its own memory namespace. Claude Code native.

Teams

aide import / export. Share agent templates via git. Team as repo, member as directory.

Philosophy

Claude Code is already a great agent runtime.

aide doesn't replace it. aide is the lifecycle manager.

Who to wake up. How much to spend. What secrets to give. When to stop.

Aidefile is to Claude Code what Dockerfile is to Linux.