Run an AI company from your terminal.
Autonomous agent orchestration engine built in Rust.
Quickstart · Docs · Examples · Discord
$ hive status
🏢 Flowtivity AI — All Systems Operational
┌─────────────────┬────────┬──────────┬───────────┬──────────┐
│ Department │ Status │ Budget │ Spent │ Mode │
├─────────────────┼────────┼──────────┼───────────┼──────────┤
│ 📝 Content │ 🟢 │ $15/day │ $11.20 │ auto │
│ 🔍 Leads │ 🟢 │ $10/day │ $7.45 │ auto │
│ 📧 Sales │ 🟡 │ $15/day │ $9.80 │ approval │
│ 📊 Analytics │ 🟢 │ $5/day │ $3.10 │ auto │
│ 🏗️ Tenders │ 🟡 │ $10/day │ $5.60 │ approval │
│ 🔬 QA │ 🟢 │ $8/day │ $3.90 │ auto │
│ 🛡️ Governance │ 🟢 │ $2/day │ $1.25 │ auto │
└─────────────────┴────────┴──────────┴───────────┴──────────┘
Total: $42.30 / $65.00 (65%) · 7/7 agents active · 23 tasks today
Hive is an orchestration engine for AI agent companies. Define your company in a TOML file, hire agents from any provider, set budgets, and let them run. Monitor everything from your terminal.
If OpenClaw is an employee, and Paperclip is a company — Hive is the operating system.
Your terminal
↕
hive (this project)
↕
┌──────────┬──────────┬──────────┬──────────┐
│ OpenClaw │ Claude │ Codex │ Any LLM │
│ agents │ Code │ agents │ agent │
└──────────┴──────────┴──────────┴──────────┘
One binary. Zero dependencies. Runs anywhere.
| Without Hive | With Hive |
|---|---|
| 20 terminal tabs, no idea which agent is doing what | hive status — one command, full picture |
| Agents burn through API credits unchecked | Per-department budgets with automatic circuit breakers |
| Company knowledge lives in your head | hive.toml — your entire company is version-controlled |
| "Did the content agent already publish today?" | hive logs content --today |
| Agent sends embarrassing duplicate email | Governance agent catches it, hive audit shows the block |
| Sharing your agent setup means writing docs | hive export > company.toml — anyone can hive import it |
| Restarting agents after a reboot takes 20 minutes | hive up — everything back in 2 seconds |
# From crates.io
cargo install hive-ai
# Or download pre-built binary (macOS, Linux, Windows)
curl -fsSL https://hive.flowtivity.ai/install.sh | sh
# Or via npm (bundles native binary)
npm install -g @AJflow/hivehive init my-company
cd my-companyThis creates a hive.toml:
[company]
name = "My AI Company"
mission = "Build something amazing"
currency = "USD"
[board]
name = "Your Name"
[[departments]]
id = "engineering"
name = "Engineering"
mission = "Build the product"
budget = { daily = 20.00 }
mode = "autonomous"
runtime = "claude-code"# Authenticate with your LLM provider
export ANTHROPIC_API_KEY=sk-ant-...
# Start all departments
hive up
# Or run a single task
hive run engineering "Build a REST API for user authentication"
# Watch it work
hive logs engineering --followYour entire company is defined in hive.toml. Git-trackable. Shareable. Reproducible.
[company]
name = "Flowtivity AI"
mission = "Work smarter with AI, not harder with more hours"
currency = "AUD"
[board]
name = "AJ Awan"
contact = "telegram:aj"
notify = ["daily-briefing", "budget-alert", "error"]
[ceo]
name = "Flowbee"
runtime = "openclaw"
interface = "telegram"
memory = true # persistent memory across sessions
[[departments]]
id = "content"
name = "Content & AEO"
mission = "Own search rankings and social presence"
budget = { daily = 15.00, alert_at = 0.8, pause_at = 1.0 }
mode = "autonomous"
runtime = "openclaw"
skills = ["aeo-blog", "linkedin-post", "social-media"]
schedule = "0 8 * * *" # wake at 8am daily
kpis = ["posts_published", "impressions", "clicks", "avg_position"]
[[departments]]
id = "leads"
name = "Lead Research"
mission = "Keep the pipeline full with qualified prospects"
budget = { daily = 10.00 }
mode = "autonomous"
runtime = "openclaw"
skills = ["lead-researcher", "contact-enrichment"]
depends_on = ["analytics"] # waits for analytics insights before researching
[[departments]]
id = "sales"
name = "Sales & Outreach"
mission = "Convert leads to meetings"
budget = { daily = 15.00 }
mode = "approval" # first-touch emails need human approval
runtime = "openclaw"
skills = ["cold-email", "prototype-builder"]
approver = "board"
depends_on = ["leads", "qa"]
[[departments]]
id = "analytics"
name = "Analytics & Intelligence"
mission = "Surface opportunities from data"
budget = { daily = 5.00 }
mode = "autonomous"
runtime = "claude-code"
skills = ["gsc-analytics", "ga4", "cloudflare-stats"]
schedule = "0 7 * * *" # first to wake, feeds other departments
[[departments]]
id = "tenders"
name = "Bid & Tender"
mission = "Find and win government tenders"
budget = { daily = 10.00 }
mode = "approval"
runtime = "openclaw"
skills = ["austender-scraper", "bid-writer"]
approver = "board"
[[departments]]
id = "qa"
name = "Product & QA"
mission = "Nothing ships broken"
budget = { daily = 8.00 }
mode = "autonomous"
runtime = "claude-code"
skills = ["prototype-qa", "email-qa", "blog-qa"]
# QA reviews output from content and sales before it ships
reviews = ["content", "sales", "tenders"]
[[departments]]
id = "governance"
name = "Governance & Budget"
mission = "Keep the machine honest and efficient"
budget = { daily = 2.00 }
mode = "autonomous"
runtime = "claude-code"
skills = ["budget-monitor", "dedup-checker", "audit-trail"]
watches = ["*"] # monitors all departmentsHive doesn't care what agent you use. If it can receive a task and return a result, it's hired.
| Runtime | How it works | Best for |
|---|---|---|
openclaw |
Connects via OpenClaw gateway API | Autonomous 24/7 agents with memory |
claude-code |
Spawns Claude Code process | Coding, analysis, one-shot tasks |
codex |
Spawns Codex CLI | Code generation, auto-approved changes |
cursor |
Connects to Cursor workspace | IDE-integrated development |
http |
POST task to any HTTP endpoint | Custom agents, cloud functions |
shell |
Runs a shell command | Scripts, cron-like automation |
mcp |
Connects via Model Context Protocol | Any MCP-compatible agent |
Every department operates in one of three modes:
| Mode | Behaviour | Use when |
|---|---|---|
autonomous |
Agent acts freely within budget | Low-risk work: analytics, research, content |
approval |
Agent drafts, human approves before execution | External comms: emails, bids, social posts |
paused |
Agent stopped, no tasks processed | Debugging, budget overrun, maintenance |
Switch modes at any time:
hive mode sales approval # require approval for all sales actions
hive mode sales autonomous # let it run
hive pause sales # stop immediately
hive resume sales # restartBudgets are not suggestions. They're circuit breakers.
$ hive budget
📊 Budget Overview — March 7, 2026
Daily budget: $65.00 / Monthly: ~$1,950
Today spent: $42.30 (65%)
┌─────────────────┬──────────┬───────────┬──────────────┐
│ Department │ Budget │ Spent │ Status │
├─────────────────┼──────────┼───────────┼──────────────┤
│ Content │ $15.00 │ $11.20 │ ██████████░░ │
│ Sales │ $15.00 │ $9.80 │ ████████░░░░ │
│ Leads │ $10.00 │ $7.45 │ █████████░░░ │
│ Tenders │ $10.00 │ $5.60 │ ███████░░░░░ │
│ QA │ $8.00 │ $3.90 │ ██████░░░░░░ │
│ Analytics │ $5.00 │ $3.10 │ ████████░░░░ │
│ Governance │ $2.00 │ $1.25 │ ████████░░░░ │
└─────────────────┴──────────┴───────────┴──────────────┘
Cost efficiency:
Per lead: $1.85 (↓12% vs last week)
Per post: $3.20
Per bid: $2.80When a department hits its budget limit:
- 80% — Warning logged, board notified
- 100% — Department auto-paused, board alerted
- Manual
hive resumeor wait for daily budget reset
Departments can depend on each other. Hive resolves the execution order automatically.
Analytics (7am) → Leads → Sales
↓
QA → [ships to external]
↑
Content
↑
Governance (watches all)
$ hive graph
analytics ──→ leads ──→ sales
│ │
content │
│ │
└──→ qa ←──┘
│
[output]
governance ──→ [watches: *]hive init [name] # Create new company in current directory
hive up # Start all departments
hive down # Gracefully stop all departments
hive status # Overview of all departments
hive org # Display org chart
hive graph # Show dependency graph
hive config # Validate and display hive.toml
hive export [file] # Export company config
hive import <file> # Import company confighive run <dept> "task" # Run a one-shot task in a department
hive logs <dept> [--follow] # Stream department activity
hive logs <dept> --today # Today's activity only
hive pause <dept> # Pause a department
hive resume <dept> # Resume a department
hive mode <dept> <mode> # Set mode: autonomous | approval | paused
hive restart <dept> # Restart department agenthive hire <dept> [--runtime] # Add/replace agent in a department
hive fire <dept> # Remove agent from a department
hive agents # List all active agents
hive health # Health check all agents (heartbeat)hive budget # Budget overview
hive budget set <dept> <amt> # Set department budget
hive budget history [--days] # Historical spend data
hive audit [dept] # Full audit trail
hive audit --errors # Only errors and blockshive inbox # Show pending approvals
hive approve <id> # Approve a pending action
hive reject <id> [--reason] # Reject with optional reason
hive approve --all # Approve all pending (use carefully)hive mcp # Start MCP server (for Claude, Cursor, etc.)
hive mcp --stdio # MCP over stdio (for tool integration)
hive serve [--port 8080] # Start web dashboard
hive api # Start REST API serverhive skills list # List installed skills
hive skills add <url> # Install skill from GitHub
hive skills remove <name> # Remove a skillRun hive mcp and any MCP-compatible client can orchestrate your company:
# In Claude Desktop settings, add:
{
"mcpServers": {
"hive": {
"command": "hive",
"args": ["mcp", "--stdio"]
}
}
}Now Claude can:
hive.status()— check all departmentshive.run("content", "Write a blog about AI trends")— assign taskshive.budget()— check spendhive.approve(id)— approve pending actionshive.logs("sales")— read activity
Works with OpenClaw, Cursor, Windsurf, or any MCP client.
hive serve --port 8080Opens a real-time dashboard at localhost:8080 with:
- Org chart visualization
- Department cards with budget bars, KPIs, and mode toggles
- Live activity feed
- Approval inbox
- Budget charts and cost analytics
- Mobile-responsive (manage from your phone)
The dashboard is optional. Everything it shows is available via the CLI.
[company]
name = "My SaaS"
[[departments]]
id = "content"
name = "Content"
budget = { daily = 10.00 }
mode = "autonomous"
runtime = "claude-code"
[[departments]]
id = "sales"
name = "Sales"
budget = { daily = 10.00 }
mode = "approval"
runtime = "openclaw"[company]
name = "AI Agency"
# Each client gets their own department
[[departments]]
id = "client-acme"
name = "ACME Corp"
budget = { daily = 25.00 }
mode = "approval"
runtime = "openclaw"
tags = ["client", "enterprise"]
[[departments]]
id = "client-startup"
name = "Startup Inc"
budget = { daily = 10.00 }
mode = "autonomous"
runtime = "claude-code"
tags = ["client", "startup"][company]
name = "My OSS Project"
[[departments]]
id = "triage"
name = "Issue Triage"
budget = { daily = 5.00 }
mode = "autonomous"
runtime = "claude-code"
skills = ["github-issues"]
[[departments]]
id = "pr-review"
name = "PR Review"
budget = { daily = 10.00 }
mode = "approval"
runtime = "codex"
skills = ["code-review"]┌─────────────────────────────────────────────────┐
│ hive CLI │
│ (Rust binary — parsing, config, orchestration) │
├─────────────────────────────────────────────────┤
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │ Scheduler│ │ Budget │ │ Audit │ │
│ │ │ │ Engine │ │ Logger │ │
│ └────┬─────┘ └────┬─────┘ └────┬─────┘ │
│ │ │ │ │
│ ┌────┴──────────────┴──────────────┴────┐ │
│ │ Department Manager │ │
│ │ (lifecycle, heartbeat, delegation) │ │
│ └────┬──────────┬──────────┬────────────┘ │
│ │ │ │ │
│ ┌────┴────┐ ┌───┴────┐ ┌──┴─────┐ │
│ │OpenClaw │ │ Claude │ │ HTTP │ ...adapters │
│ │Adapter │ │ Code │ │Adapter │ │
│ └─────────┘ └────────┘ └────────┘ │
│ │
├─────────────────────────────────────────────────┤
│ Storage: SQLite (embedded) or Postgres │
│ Config: hive.toml │
│ Logs: Append-only audit trail │
│ API: REST + WebSocket + MCP │
└─────────────────────────────────────────────────┘
- Single binary — No runtime dependencies. Download and run.
- SQLite default — Zero-config embedded database. Postgres optional for scale.
- Append-only audit — Every agent action is immutable. Compliance-ready.
- Adapter pattern — New agent runtimes are a single trait implementation.
- Config-driven — All behaviour defined in
hive.toml. No code required. - Offline-first — Works without internet for local agents. Cloud optional.
| Hive | Paperclip | CrewAI | AutoGen | |
|---|---|---|---|---|
| Language | Rust | Node.js | Python | Python |
| Interface | CLI + Dashboard | Dashboard | Python API | Python API |
| Config | TOML file | Dashboard UI | Python code | Python code |
| Budgets | Per-department, circuit breakers | Per-agent limits | None | None |
| MCP | Native | No | No | No |
| Agent-agnostic | Any runtime | OpenClaw, Claude, Codex | LangChain-based | OpenAI-based |
| Audit trail | Append-only log | Basic logging | None | None |
| Install | Single binary | Docker/Node.js | pip install | pip install |
| Company-as-code | hive.toml | UI config | Python scripts | Python scripts |
| Dependency graph | Built-in | No | Sequential only | Round-robin |
- Core CLI — init, up, down, status, logs
- hive.toml config system
- Budget engine with circuit breakers
- OpenClaw adapter
- Claude Code adapter
- Audit trail
- MCP server (
hive mcp) - Web dashboard (
hive serve) - Approval workflow (
hive inbox) - Dependency graph execution
- Codex adapter
- HTTP adapter (webhooks)
-
hive publish— share company templates - Plugin system for custom adapters
- Multi-company support (
hive.workspace.toml) - Hive Cloud — managed hosting (coming 2026)
We welcome contributions! See CONTRIBUTING.md for guidelines.
# Clone and build
git clone https://github.com/AJflow/hive.git
cd hive
cargo build
# Run tests
cargo test
# Run with example config
cargo run -- --config examples/flowtivity/hive.toml statusHive is built and tested using AI agents — specifically OpenClaw and Claude Code. We eat our own cooking.
The Flowtivity team runs Hive in production to manage 7 departments, 186+ leads across AU and US, and $65/day in autonomous AI operations.
Read the full story:
- Zero-Human Companies Are Here
- OpenClaw vs Paperclip: The Framework Comparison
- The Agentic Company Pattern
MIT — see LICENSE for details.
Built by Flowtivity 🐝 — Work smarter with AI, not harder with more hours.