Run your whole bolthub account from your agent
When you connect bolthub to Claude Desktop today, your agent can discover any listed API, preview its price, pay for a call over Lightning, buy prepaid credit across a provider, mint a scoped credential for a sub-agent, and keep verifiable receipts. That is the *consumer* side of a marketplace, and it works.
But one pair of tools already does something different. deploy_node and node_status do not consume a paid API. They act on your account: they provision infrastructure on your behalf, authenticated as you. That is the first step across a line worth naming out loud, because we are about to cross it deliberately and completely.
The direction: your agent stops being only a *buyer* on bolthub and becomes an agent-native control plane for your whole account. Everything you do today by clicking through the dashboard, you will be able to do by asking your agent, with one-click authorization and without bolthub ever holding your keys or your money.
Here is what that looks like.
List an API without leaving your agent
"Here is my OpenAPI spec, put it behind a Lightning paywall at 5 sats a call."
That sentence becomes the whole onboarding flow. The agent imports your spec, proposes a pricing model per endpoint, sets up your hosted gateway, and hands you back a live listing to approve. The pieces that make this real (spec import, origin configuration, per-endpoint pricing, directory publishing) already run in production behind the dashboard. Wrapping them as agent tools turns a ten-minute form into a conversation.
You still connect your own wallet, and you still approve the listing before it goes live. The agent does the assembly, not the deciding.
Spin up a Lightning node, still non-custodial
This one already exists in preview, and it is the clearest proof that the model holds. Your agent can provision a Lightning node for you and report on its progress, and at no point does bolthub generate or hold the node's keys. The node is yours. We orchestrate the deploy; your infrastructure holds the keys.
The natural extensions follow the same rule. Check a node's status, order inbound liquidity so you can actually receive payments, watch it come online, all from the agent. Money and keys stay on your side of the line the entire time. That is the whole point: convenience that never asks you to hand over control.
Let your agent audit your own API
This is the one we are most excited about, because it turns bolthub's own quality bar into a tool you can run.
We already measure a lot about every listed endpoint: uptime, latency, response samples, schema shape, pricing. Expose that to your agent and ask the obvious question. "Audit my listing. What would make an agent choose it and pay again?" The agent reads your metadata, checks it against what actually makes an endpoint agent-friendly (clean typed responses, a stable versioned schema, machine-actionable errors, tight payloads, a pricing model that matches the workload), and hands you back a prioritized punch list.
It is the seller's guide we wrote, except your agent runs it against your real endpoints instead of you reading it and hoping.
A few more on the same track
The pattern generalizes, so a handful of high-value tools come almost for free once the authenticated surface is there:
- •Usage and revenue, on demand. "How much did my weather API earn this week, and which endpoint is busiest?" Your reporting is already computed; this just lets the agent ask.
- •Tune pricing from evidence. Pair the usage numbers with the audit and the agent can propose a price change ("your 402 rate on this endpoint is high, try 3 sats"), which you approve.
- •Watch and react. Standing checks the agent can arm for you (an endpoint's uptime dips, an invoice settles, a node needs liquidity) so the agent tells you rather than you refreshing a dashboard.
The one promise underneath all of it
Every capability above moves metadata and infrastructure. None of them moves money, and none of them asks you for a key. That is not an accident we are being careful to preserve; it is the design. A single Lightning payment settles to one wallet, and it is always yours. Your node holds its own keys. Your wallet connection stays a thing you do. The agent orchestrates around your keys; it never becomes their custodian.
One-click authorization will make the whole thing feel seamless: approve once in a browser, and your agent holds a scoped, revocable token, never your password and never your wallet. Seamless and non-custodial are not in tension here. We built the account layer that way on purpose.
When
Soon, and deliberately without a hard date in this post. Most of what is described here is a thin wrapper over an API that is already in production, which is why we are comfortable showing you the direction now rather than after the fact. We would rather tell you where bolthub is pointed and let you tell us which of these you want first.
If you are building agents, or listing APIs for them, this is the part where it stops being "a payment rail you call" and starts being "the account your agent runs for you." That is the next level, and it is close.