BytesBrains Cruise for VS Code

Every model your Cruise key can reach — in Copilot Chat and agent mode —
with budgets, per-project keys, and one cost ledger that stay on the gateway.

Install

Where	Link
Visual Studio Marketplace	BytesBrains.bytesbrains-cruise
Open VSX	bytesbrains/bytesbrains-cruise
Product	bytesbrains.com/cruise

This extension is a client of Cruise, not a second place spend can happen. Your keys, budgets and ledger stay on the gateway.

Try it before anyone issues you a key

Install the extension.
Command palette → Cruise: Manage API key → Try the demo.
Paste a cru_demo_ key.

That points cruise.endpoint at https://cruise-demo.bytesbrains.net/v1 — production’s model ids exactly, every price zero, and no provider credential in the deployment. Answers are fabricated. It costs nothing to take.

For real traffic, set the endpoint back to https://cruise.bytesbrains.net/v1 and enter a cru_live_ key the same way.

What you get


Vendor	Registers as `cruise`, so models appear in Copilot Chat’s picker and in agent mode — not behind a chat prefix
Models	Fetched from `GET /v1/models` whenever the list is asked for. Never frozen into the extension
Key	Lives in the editor’s `SecretStorage`, entered only through Cruise: Manage API key
Endpoint	`cruise.endpoint` — a user setting a workspace cannot override, because the key is sent wherever it names
Output bound	`cruise.maxOutputTokens`, clamped to each model’s measured ceiling

Why the list is fetched, not shipped

Cruise publishes only what it can actually serve for the key you present, with prices. A measurement older than 30 days is treated as unverified rather than last-known-good, and the model stops being routable. A list baked into an extension would offer models the gateway refuses — and you would blame the extension. Correctly.

Why the key never touches `settings.json`

Settings sync would replicate a live key onto every machine you sign into. That is a leak with no event to notice it by.

Lanes are models here too

A bb/… row in the picker is a lane: Cruise picks a member per request by price and by what the request needs. Hover shows the members and the dearest rate any of them can charge — because that is what a budget has to survive.

When it refuses

Cruise answers both spending refusals with 429 and OpenAI’s insufficient_quota on purpose — that is the shape every OpenAI client already understands. They are not the same event; this extension says which one you hit:

Refusal	Meaning
Out of budget	A period cap. It names when it resets; waiting is the fix
Out of credit	A lifetime cap. Waiting does nothing; a credit grant lifts it

Anything else — rate limit, stale measurement, a model your key may not call — is passed through with Cruise’s own sentence (project and figures), plus whatever the editor can add about what to do next. Detail goes to the BytesBrains Cruise output channel.

What it does not do

No telemetry
No update check against a second host
No traffic anywhere but the configured base URL
No runtime dependencies — the whole extension is one bundled file

It holds a cru_ key and never a provider credential. If this machine is compromised, the blast radius is one key you can revoke, capped by the budget and rate limit set when it was issued. Ask for yours with --rate-limit and --account-rate-limit set.

Develop

npm ci
npm run typecheck
npm test
npm run build

npm run package writes a .vsix you can load with Extensions: Install from VSIX….

Publishing to the Marketplace and Open VSX is a person running a command — not something a merge does. See CONTRIBUTING.md and SECURITY.md.

Licence

See LICENSE.txt. Proprietary: use with Cruise; no modified redistributions.

BytesBrains Cruise

BytesBrains