FAQ#
Short answers to the questions people ask most about Lockpin, with links to the pages that say more. It is for anyone deciding whether to use Lockpin or getting started with it.
Which models does it work with?#
Anthropic, OpenAI, Google Gemini, Meta, xAI, DeepSeek, Mistral, Groq, Together, Fireworks and OpenRouter, and through your own cloud account on Amazon Bedrock, Google Vertex AI and Azure OpenAI. Prices for over 100 models are built in, including cached input, cache writes and long-context rates. Add any other model, or an OpenAI-compatible server of your own, with one small file.
lockpin providers -models lists every model with its price. A model without a price is refused,
so the catalog matters. See Supported providers and
Prices and the price feed.
What happens when a task hits its budget?#
The call that would go over is held before it reaches the provider. If the meter, VS Code or the overlay is open, Lockpin asks: allow this call, raise the budget, or stop. With nobody watching, or no answer in 3 minutes, your agent gets HTTP 402 with the numbers, so it can save its work and stop cleanly.
A header can lower a task’s budget but not raise it; raises go through the meter, VS Code or the overlay. See Approvals at the cap.
Does it slow my agent down?#
Hardly. Far from the cap, Lockpin sizes each call itself and sends it straight on. Only near the cap does it ask the provider for an exact input count, where the provider offers one. The live meter shows how many milliseconds Lockpin adds to each call.
Does it change what the model sees?#
The spend cap never changes your prompts. So that the cap holds, it may give a call an output limit
or, near the cap, lower one (-fit-output and -degrade, both on by default;
X-Lockpin-Degrade: off turns degrading off for one request).
Trimming and project notes do change what the model sees, so both are off unless you turn them on. Trimming can also be switched per request. Configuration marks every key that changes what the model sees.
Do my provider keys go to Lockpin’s servers?#
No. Calls do not pass through our servers. On your own machine, the agent sends its API key to Lockpin, which passes it to the provider with the call. Lockpin does not write it to disk, the ledger or its logs. In a team, a team node can hold the provider keys so that agents never see them. See Security.
Does it work offline or if your servers are down?#
Yes. The decision to allow or refuse a call is made on your machine, from its own ledger file. It never waits on our servers, and it works with the network to us cut off. The hosted dashboard is optional.
What does Lockpin send to you?#
With the default settings, no spend data. There is no telemetry and no automatic update check.
Lockpin contacts our servers only for features you turn on: reporting to the dashboard, joining a
team, asking about held calls on your phone or in chat, and lockpin upgrade. See
What leaves your machine.
Do I have to route my agents through Lockpin to see my spend?#
No. lockpin import reads what Anthropic, OpenAI and Cursor billed from their admin APIs, with
admin keys from your environment. Capping a call needs the proxy; seeing last month’s bill does not.
lockpin import -days 7
Can it cap a Claude or ChatGPT subscription?#
No. A proxy cannot cap a subscription. lockpin plan reads the Claude Code and Codex session logs
on your machine and shows where a subscription’s usage went, by window, model and session. It needs
no Lockpin running and makes no network call.
Can it cap Cursor?#
Not from your machine. Cursor’s own servers make its calls, so they cannot reach a Lockpin on your
machine. lockpin connect cursor prints the steps that do apply, and lockpin import can read what
Cursor billed. See Cursor.
Is there a VS Code extension?#
Yes, free, on the VS Code Marketplace and on Open VSX for Cursor, Antigravity and Windsurf: search for Lockpin in the Extensions view. It reads the meter you already run and puts your tasks’ spend against their budgets in the status bar and side bar. The cap is still enforced by Lockpin, not by the editor. See VS Code extension.
Does Lockpin use AI itself?#
No. All of its logic is plain arithmetic and file snapshots, so it adds nothing to your bill, unless
you turn on cache keep-alive pings (-cache-keepalive), which are charged to the task.
Is a team budget across several machines exact?#
No. Each machine enforces its own cap, so a budget or total across several machines is soft: the machines together can pass it, and the dashboard labels it soft. A dashboard budget can only lower a machine’s own limit, never raise it. See Teams.
What if Lockpin stops?#
When its own parts fail, Lockpin fails closed by default: a call it cannot reserve and record is refused, never sent unseen. If the process is not running at all, your agent’s calls fail to connect. See Troubleshooting for the choices.
Lockpin used to be called Spendrail. Do old setups still work?#
Yes. Nothing was released under the old name, so this matters only to a machine that ran a pre-launch build. The old names are still accepted:
What |
New |
Old, still accepted |
|---|---|---|
Request headers |
|
The same names with |
Meter, health check and API |
|
|
Alert variables |
|
|
Default files |
|
|
Key prefix |
|
Existing |