boletus 0.1.4

Your first AI company — Open-source multi-agent framework that runs your business autonomously

Boletus 🍄

Your first AI company — open-source framework that runs an autonomous AI team via Slack + Claude CLI.

Send a business plan to your CEO agent. It hires a team, delegates tasks, writes code, creates marketing — all autonomously.

Quickstart

Prerequisites

• Python 3.11+
• Claude CLI installed and authenticated
• A Slack workspace (create one free)
• (Optional) Connect services in claude.ai/settings → Integrations (Gmail, Notion, Figma, etc.) — your AI team will use these automatically

1. Install

pip install boletus

2. Create a project & connect Slack

mkdir my-ai-company && cd my-ai-company
boletus init

This walks you through connecting Slack:

1. Create a Slack app at api.slack.com/apps → "From a manifest" → paste slack-app-manifest.json
2. Paste your App Token (xapp-...) and Bot Token (xoxb-...)
3. Paste your Slack Member ID

3. Start your AI team

boletus setup

The wizard DMs you in Slack, asks about your business, generates the team config, creates channels, and starts everything.

Upload a business plan PDF or describe what you want to build — the AI team takes it from there.

4. Watch it work

• #ceo — strategy, delegation, progress reports
• #engineering — CTO builds the product, hires devs
• #growth — CMO runs marketing, hires content writers

Type help in any channel for available commands.

How it works

You: "Build a SaaS for restaurant analytics"
        │
        ▼
   ┌─────────┐     ┌────────────┐     ┌──────────┐
   │   CEO   │────>│ Task Board │────>│ Workers  │
   │ (plans) │     │            │<────│(execute) │
   └─────────┘     └────────────┘     └──────────┘
        │                                    │
        ├── hires @cfo, @sales_rep...       │
        └──────── Slack channels ───────────┘

1. CEO receives your business plan, breaks it into tasks, delegates
2. CTO is a tech lead — scaffolds the project, hires @backend_dev, @frontend_dev, @tester, @devops
3. CMO is a growth lead — hires @content_writer, @designer, @seo_specialist
4. Workers claim tasks, write real code/content, self-verify before reporting
5. Auto-testing — after code tasks, test tasks are auto-created
6. Manager review — CTO/CMO approve or reject worker output with feedback
7. Auto-hiring — agents create new roles on the fly when workload demands it
8. Self-correction — workers escalate blockers, fix tasks auto-created

Dynamic team

Start with 3 agents (CEO, CTO, CMO). They hire more as needed:

CEO writes: @sales_rep: Build a list of 50 target companies and start outreach
→ system auto-creates sales_rep agent, starts working

CTO writes: @tester: Write integration tests for the payment API
→ system auto-creates tester agent, assigns first task

No manual configuration needed — the team grows organically.

Slack commands

Command	What it does
tasks	What everyone is working on
my tasks	Tasks for this channel's agent
team	Who's on the team and their roles
files	All local files the team created
links	Notion pages, GitHub repos, designs — all URLs
report	Ask CEO for a progress report
hire <description>	Add a new team member
help	All available commands

CLI

boletus setup       # Slack wizard — creates team + channels + starts bot
boletus run         # Resume where you left off
boletus init        # Manual setup — creates crew.yaml + .env
boletus doctor      # Check prerequisites
boletus validate    # Verify crew.yaml
boletus agents      # List configured agents
boletus tasks       # Show task board

Integrations (27 services)

Three tiers — zero to full setup:

Tier	Setup	Examples
Claude.ai Connectors	Connect once at claude.ai	Gmail, Notion, Figma, Canva, Gamma, Calendar, PostHog, Cloudflare, Miro
CLI tools	Token in .env	GitHub (gh), AWS (aws), Stripe (stripe), Vercel
Local MCP	Auto-configured	PostgreSQL, custom servers

Your agents use whatever services you've connected in your Claude account — no API keys needed for Claude.ai connectors. CLI tools are configured during boletus setup.

Configuration

Everything lives in crew.yaml (auto-generated by the setup wizard):

name: "My AI Company"

slack:
  app_token: ${SLACK_APP_TOKEN}
  bot_token: ${SLACK_BOT_TOKEN}

owner:
  slack_id: ${OWNER_SLACK_ID}

git:
  author_name: "AI Team"
  author_email: "[e-mail]"
  github_org: "my-org"  # repos created under this org

integrations: [github, gmail, notion, figma, canva]

projects:
  my-app:
    name: "My App"
    codebase: "./my-app"

# Custom MCP servers
# mcp_servers:
#   my-api:
#     command: "npx"
#     args: ["-y", "@my-org/my-mcp-server"]

Safety

• Circuit breaker — agents auto-pause after consecutive failures
• Manager review — worker output verified before marking complete
• Auto-test loop — code tasks automatically get a follow-up test task
• Escalation — workers flag blockers, fix tasks auto-created
• Fuzzy dedup — prevents re-delegating the same task
• Stuck task backoff — exponential backoff prevents planning loops
• Structured memory — agents learn from rejections, share knowledge cross-team

License

MIT