ln.bot

The official Python SDK for ln.bot — Bitcoin for AI Agents.

Give your AI agents, apps, and services access to Bitcoin over the Lightning Network. Create wallets, send and receive sats, and get real-time payment notifications.

from lnbot import LnBot

ln = LnBot(api_key="uk_...")
w = ln.wallet("wal_...")

invoice = w.invoices.create(amount=1000, memo="Coffee")

ln.bot also ships a TypeScript SDK, C# SDK, Go SDK, Rust SDK, CLI, and MCP server.

---

Install

pip install lnbot

---

Quick start

Register an account

from lnbot import LnBot

ln = LnBot()
account = ln.register()
print(account.primary_key)
print(account.recovery_passphrase)

Create a wallet

ln = LnBot(api_key=account.primary_key)
wallet = ln.wallets.create()
print(wallet.wallet_id)

Receive sats

w = ln.wallet(wallet.wallet_id)

invoice = w.invoices.create(amount=1000, memo="Payment for task #42")
print(invoice.bolt11)

Wait for payment (SSE)

for event in w.invoices.watch(invoice.number):
    if event.event == "settled":
        print("Paid!")
        break

An expired invoice is not always final: one paid just before it expired can turn settled a moment later. Before giving up, check again with w.invoices.get(invoice.number), or keep listening for invoice.settled on w.events.stream() or a webhook.

If the wait runs out first (timeout= seconds, 60 by default, 300 at most), the last event is timeout: event.data is the current state; read again or watch again.

Event stream and webhook payloads use snake_case on the wire (max_fee, created_at). The SDK converts event payloads to the same field names as its REST types; a webhook handler reads the snake_case JSON as is.

Send sats

payment = w.payments.create(target="alice@ln.bot", amount=500)

Leave max_fee unset. The server defaults it to 1% of the amount (minimum 20 sats), which covers the network fee (0.25% plus Lightning routing) for most payments. Only the fee estimate plus a small buffer is held while the payment is in flight. On success the wallet is debited amount + actual_fee + service_fee.

If the fee estimate is above the limit, create raises BadRequestError (for example Network fee estimate (X sats) exceeds maxFee (Y sats).). Pass a higher max_fee only if you accept that fee.

A payment that can't be sent fails quickly, and its funds are released at once: within about half a minute if the Lightning service refuses it, or after 10 minutes of automatic retries for a temporary problem (stage "retrying", until retry_until). If its invoice expires first, it fails then, with Invoice expired before it could be paid. A payment stays processing longer only when its funds are already on the Lightning network: it settles or fails as soon as the network does, usually within seconds, and is never refunded while it could still be paid. If watch() runs out of time first, it ends with a timeout event whose data is the current state: check again with w.payments.get(payment.number) or watch again. A failed payment carries the reason in failure_reason: show it as is, and when it ends in "Please try again.", sending the payment again later can work.

Payment stages

Every payment carries stage, which says where it is more precisely than status, and status_detail, a sentence to show your user as is:

from lnbot import PaymentStage

p = w.payments.get(payment.number)
print(p.status, p.stage, p.status_detail)

if p.stage == PaymentStage.RETRYING:
    print("Retried until", p.retry_until)

stage	With status	status_detail
queued	pending	Waiting to be sent.
retrying	pending	Not sent yet; retrying for a few minutes. If it still can't be sent, it fails and the funds are released.
sending	pending or processing	Being sent. Between ln.bot wallets: Being transferred to the recipient's wallet.
in_flight	processing	Sent; the Lightning network is completing it, usually within seconds.
confirming	processing	Checking with the Lightning network whether it went out. The funds stay on hold until then.
on_hold	processing	Interrupted by an earlier version of the service and on hold until it is reviewed. The funds are held, not lost.
settled	settled	None
failed	failed	The failure_reason

A retrying payment carries retry_until: if it still can't be sent by then, it fails within about a minute and the funds are released. A payment can go from processing back to pending when it is retried, so treat both as in progress.

stage is a plain string and more stages may be added: for one you don't know, go by status. When the API doesn't send them, stage, status_detail and retry_until are None.

watch() events carry the same fields, but the stream only ends with settled, failed or timeout (the payment's current state): it doesn't send stage changes as they happen. Webhooks and w.events.stream() don't carry stage.

Price a payment before sending

quote = w.payments.resolve(target="alice@example.com", amount=500)
print(quote.network_fee_estimate, quote.service_fee, quote.total)

if quote.fee_warning:
    print(quote.fee_warning)  # sending without max_fee would be rejected

For a BOLT11 invoice, leave out amount (or pass the invoice amount). quote.internal is True when the recipient is a wallet on ln.bot, which makes the transfer free.

Pricing a Lightning address or LNURL fetches an invoice from the recipient, so it is slower. Call it once the amount is settled, not on every keystroke.

resolve raises BadRequestError when the target can't be paid, for example Recipient not found. for an unknown ln.bot address or Cannot transfer to yourself. for your own. When the amount is outside the recipient's range, the JSON in the error's body also carries min and max.

Send everything

Resolve a Lightning address or LNURL without an amount to get max_sendable, the most this wallet can send there now after fees:

quote = w.payments.resolve(target="alice@example.com")

if quote.max_sendable:
    w.payments.create(target="alice@example.com", amount=quote.max_sendable)

Don't pass max_fee here: max_sendable assumes the default. It is 0 when the balance can't cover the recipient's minimum.

Check balance

info = w.get()
print(f"{info.available} sats available")

---

Wallet-scoped API

All wallet operations go through a Wallet handle obtained via ln.wallet(wallet_id):

w = ln.wallet("wal_abc123")

# Wallet info
info = w.get()
w.update(name="production")

# Sub-resources
w.key           # Wallet key management (wk_ keys)
w.invoices      # Create, list, get, watch invoices
w.payments      # Send, list, get, watch, resolve payments
w.addresses     # Create, list, delete, transfer Lightning addresses
w.transactions  # List transaction history
w.webhooks      # Create, list, delete webhook endpoints
w.events        # Real-time SSE event stream
w.l402          # L402 paywall authentication

Account-level operations stay on the client:

ln.register()                       # Register new account
ln.me()                             # Get authenticated identity
ln.wallets.create()                 # Create wallet
ln.wallets.list()                   # List wallets
ln.keys.rotate(0)                   # Rotate account key
ln.invoices.create_for_wallet(...)  # Public invoice by wallet ID
ln.invoices.create_for_address(...) # Public invoice by address
ln.status()                         # Service status now (no key needed)
ln.status_history("7d")             # Service status over time (no key needed)

---

Service status

Whether ln.bot is working, measured about once a minute. Both calls are public: they need no API key and no wallet.

from lnbot import LnBot

ln = LnBot()
s = ln.status()
print(s.status)  # "operational", "degraded", "outage" or "unknown"

if s.notice:
    print(s.notice.level, s.notice.message)  # a message from the operators

sending = s.components.sending
print(sending.in_progress, sending.by_stage, sending.median_seconds)

s.components has sending (payments out), receiving (payments in), node (the Lightning node) and webhooks (deliveries), each with its own status. unknown means nothing was measured for 5 minutes, and the numbers are as old as s.updated_at. A number that wasn't measured is None, for example sending.median_seconds when nothing was sent.

h = ln.status_history("7d")
print(h.summary.uptime.overall.operational)  # percent of the measured time

for b in h.buckets:
    if b.samples == 0:
        continue  # not measured: no data, neither up nor down
    uptime = (b.samples - b.overall.degraded - b.overall.outage) / b.samples
    print(b.at, f"{uptime:.1%}", b.sending.sent, b.sending.p95_seconds)

range	One bucket per	bucket_seconds
24h (default)	5 minutes	300
7d	hour	3600
30d	6 hours	21600
90d	day	86400

Buckets are aligned to UTC and the last one is still filling, so don't assume how many there are. h.summary covers the whole range, with settle_times as a histogram of settlement times, and h.notices lists the operators' notices during it. The range starts at h.from_ (from in the JSON). Any other range raises BadRequestError.

Statuses and notice levels are plain strings, like stage, and new values may be added: compare them with the constants in ServiceState and NoticeLevel, and handle a value you don't know.

Both calls share the API's limit for public routes, 10 requests a minute per IP; over it they raise TooManyRequestsError (status 429). The status changes about once a minute, so there is no point asking more often.

---

Async support

Every method has an async equivalent via AsyncLnBot:

from lnbot import AsyncLnBot

async with AsyncLnBot(api_key="uk_...") as ln:
    w = ln.wallet("wal_...")
    info = await w.get()
    invoice = await w.invoices.create(amount=1000)

    async for event in w.invoices.watch(invoice.number):
        if event.event == "settled":
            print("Paid!")
            break

---

Error handling

from lnbot import LnBot, BadRequestError, UnauthorizedError, NotFoundError, ConflictError, TooManyRequestsError, LnBotError

try:
    w.payments.create(target="invalid", amount=100)
except BadRequestError:
    ...  # 400
except UnauthorizedError:
    ...  # 401
except NotFoundError:
    ...  # 404
except ConflictError:
    ...  # 409
except TooManyRequestsError:
    ...  # 429
except LnBotError as e:
    print(e.status, e.body)

Configuration

from lnbot import LnBot

ln = LnBot(
    api_key="uk_...",                 # or set LNBOT_API_KEY env var
    base_url="https://api.ln.bot",    # optional — this is the default
    timeout=30.0,                     # optional — request timeout in seconds
)

The API key can also be provided via the LNBOT_API_KEY environment variable. If both are provided, the constructor argument takes precedence.

---

L402 paywalls

w = ln.wallet("wal_...")

# Create a challenge (server side)
challenge = w.l402.create_challenge(amount=100, description="API access", expiry_seconds=3600)

# Pay the challenge (client side). max_fee works as for payments: leave it unset for the default.
result = w.l402.pay(www_authenticate=challenge.www_authenticate)
print(result.fee)  # network fee charged

# Verify a token (server side, stateless)
v = w.l402.verify(authorization=result.authorization)
print(v.valid)

---

Features

• Zero extra dependencies — only httpx
• Wallet-scoped API — ln.wallet(id) returns a typed scope with all sub-resources
• Sync + async — LnBot and AsyncLnBot with identical APIs
• Typed exceptions — BadRequestError, NotFoundError, ConflictError, UnauthorizedError, ForbiddenError, TooManyRequestsError
• SSE support — watch() returns an iterator/async iterator for real-time events
• Dataclass responses — all responses are frozen dataclasses

Requirements

• Python 3.10+
• Get your API key at ln.bot

Links

• ln.bot — website
• Documentation
• GitHub
• PyPI

Other SDKs

• TypeScript SDK · npm
• C# SDK · NuGet
• Go SDK · pkg.go.dev
• Rust SDK · crates.io · docs.rs

License

MIT