BitAgent

Payment rails for AI agents — Bitcoin Lightning, built in.

BitAgent is an open-source framework for building AI agents that autonomously send and receive Bitcoin payments over the Lightning Network. Agents discover each other via Nostr, verify identity with DIDs, and coordinate multi-step workflows — all settled in real sats.

---

Why BitAgent?

Most AI agent frameworks treat money as an afterthought. BitAgent treats it as a first-class primitive.

Agent A                          Agent B (PolyglotAgent)
  │                                    │
  │── POST /polyglot/translate ───────>│
  │<── { payment_request: "lnbc..." } ─│  (no pay, no work)
  │                                    │
  │── pay Lightning invoice ──────────>│ LNbits
  │── POST /translate + payment_hash ─>│
  │<── { translated_text: "..." } ─────│

• Agents charge sats for services — translation, transcription, data lookups, anything
• Agents pay other agents — chain services together, costs flow automatically
• No API keys or accounts — Lightning invoices are the auth layer
• Discoverable via Nostr — agents announce themselves; anyone can find and call them

---

Quick Start

Requirements: Python 3.11+, a LNbits wallet (free at legend.lnbits.com)

git clone https://github.com/intrinsicinvestment91/bitagent.git
cd bitagent
pip install -r requirements.txt
cp env.template .env
# edit .env: set LNBITS_URL and LNBITS_API_KEY
python main.py

The server starts on http://localhost:8000. Open /docs for the interactive API.

Translate text (100 sats):

# Step 1 — request service, get invoice
curl -X POST http://localhost:8000/polyglot/translate \
  -H "Content-Type: application/json" \
  -d '{"text": "Hello world", "target_language": "es"}'
# → {"payment_request": "lnbc...", "payment_hash": "abc123..."}

# Step 2 — pay the invoice with any Lightning wallet, then submit with payment_hash:
curl -X POST http://localhost:8000/polyglot/translate \
  -H "Content-Type: application/json" \
  -d '{"text": "Hello world", "target_language": "es", "payment_hash": "abc123..."}'
# → {"translated_text": "Hola mundo"}

---

What's Included

Agent	Endpoint	Service	Price
PolyglotAgent	/polyglot/translate	Text translation (100+ languages)	100 sats
PolyglotAgent	/polyglot/transcribe	Audio → text (Whisper)	250 sats
CoordinatorAgent	/coordinator/translate_audio	Audio → transcribe → translate	350 sats
PriceOracleAgent	/oracle/price/{coin}	Live crypto price (BTC, ETH, LTC…)	2 sats
WebFetchAgent	/fetch/url	Fetch & clean any public web page	25 sats
SearchAgent	/search/query	Web search (Brave → SearXNG → DDG)	10 sats
StreamfinderAgent	/a2a (JSON-RPC)	Streaming availability search	100 sats
IdentityAgent	/a2a (JSON-RPC) + /.well-known/nostr.json	Paid NIP-05 registration + trust lookup	1000 sats (register), 10-25 sats (queries)

All agents implement the A2A JSON-RPC protocol — they can call each other without human involvement.

---

Use with Claude (MCP)

BitAgent ships an MCP server so Claude can call your agents as tools directly — no HTTP, no payment gate in the loop.

# Clone and install
git clone https://github.com/intrinsicinvestment91/bitagent.git
cd bitagent
pip install -r requirements.txt

# Add to Claude Code
claude mcp add bitagent -- python /path/to/bitagent/mcp_server.py

Or drop a .mcp.json in your project root:

{
  "mcpServers": {
    "bitagent": {
      "type": "stdio",
      "command": "python",
      "args": ["/path/to/bitagent/mcp_server.py"]
    }
  }
}

Claude then has access to search, fetch_url, translate, get_price, and convert_sats as native tools. Each result includes a cost_sats field so you can track spending.

---

Architecture

src/
├── agents/
│   ├── polyglot_agent/       # Translation & transcription (FastAPI router)
│   ├── coordinator_agent/    # Multi-agent orchestration
│   ├── price_oracle_agent/   # Live crypto prices (CoinGecko → Binance fallback)
│   ├── web_fetch_agent/      # Web page fetcher with SSRF protection
│   ├── search_agent/         # Web search (Brave → SearXNG → DuckDuckGo fallback)
│   └── streamfinder/         # A2A reference implementation
├── core/
│   ├── agent.py              # Base class: DID identity + wallet + security
│   ├── agent_server.py       # FastAPI wrapper with auto-generated endpoints
│   └── payment.py            # Payment flow helpers
├── security/                 # JWT auth, AES-256 encryption, input validation
├── identity/                 # DID document management (did:key, did:nostr, did:bitcoin)
├── network/                  # Nostr discovery, DHT peer-to-peer
└── wallets/                  # LNbits client, Fedimint ecash wallet

mcp_server.py                 # MCP stdio server — exposes agents as Claude tools

Adding a new agent takes ~50 lines. Mount a FastAPI router, decorate your endpoint with @require_payment(min_sats=100), and you have a paid service.

---

IdentityAgent

IdentityAgent adds:

• paid identity.register_nip05 flow
• machine-readable identity.get_identity and identity.get_trust_signal
• dynamic NIP-05 resolution at /.well-known/nostr.json?name=<handle>

Example trust lookup:

curl -X POST http://localhost:8000/a2a \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 11,
    "method": "identity.get_trust_signal",
    "params": { "pubkey": "npub1examplepubkey" }
  }'

See IdentityAgent docs for full registration/payment flow, NIP-05 examples, and security warnings.

---

Use Cases

Paid translation API — instead of managing API keys and subscriptions, clients pay per request in sats. No account required.

Autonomous research pipeline — a coordinator agent calls a transcription agent, then a translation agent, then a summarization agent. Each hop is paid automatically from the coordinator's wallet.

Agent marketplace — agents announce services on Nostr. Other agents discover them by service type and price, pick the best option, and call them directly.

Metered AI services — wrap any ML model in an agent. Users pay per inference. No billing infrastructure needed.

---

Deployment

Cloud (Railway): Connect this repo on railway.app. Set LNBITS_URL and LNBITS_API_KEY as environment variables. Done.

Self-hosted (Start9): See docs/START9_DEPLOYMENT_GUIDE.md.

Docker:

docker-compose -f docker-compose.dev.yml up

---

Contributing

BitAgent is early. There's a lot of room to build.

See CONTRIBUTING.md for how to add new agents, payment backends, and discovery protocols.

Good first issues are labeled good first issue on GitHub.

---

License

MIT — use it, fork it, build on it.