Proposal: BytesBrains Cruise as a curated provider #24

What this is

We run Cruise, an OpenAI-compatible gateway: one key across many hosts (Workers AI, DeepSeek, Mistral, Anthropic, OpenAI…), per-project budget caps that stop a runaway loop, and a ledger of what each request cost. We'd like to add it to PROVIDERS as a curated entry, and are asking here first, as the README invites, before sending a PR. Disclosure: we operate Cruise.

Why it may suit OpenHack users: a scan is a long agent loop. A hard daily cap means a stuck hunter stops at a figure the user chose, not at the end of their card, and the ledger answers "what did that scan cost" when the CLI has no pricing for a BYOK model (it reports $0.0000).

Tested before asking

OpenHack 0.2.4, openhack --scan, no code changes, pointed at Cruise through the existing openai provider overrides (OPENAI_BASE_URL / OPENAI_MODEL):

• A full scan of a small Flask app with three planted bugs, on workers-ai/@cf/zai-org/glm-5.3 (the same family as your current default, Default new sessions to GLM 5.3 Flash #11): recon, hunter swarm and validator swarm all completed, and command injection and SQL injection were confirmed. 54 tool calls, parallel calls included. About $1.02 for the scan, 16m 48s.
• prompt_cache_key is accepted, so supports_prompt_cache=True.
• Refusals arrive as OpenAI-shaped errors that OpenHack already displays: a model lacking a capability the request needs gets 400 unsupported_capability, and an exhausted cap gets 429 budget_exhausted, after which OpenHack offered --resume.

The same runs turned up #21, #22 and #23, which affect any OpenAI-compatible provider, not only this one.

The change

One entry in openhack/providers.py, plus whatever your tests do for the other curated providers:

"cruise": ProviderSpec(
    "cruise", "BytesBrains Cruise", "https://cruise.bytesbrains.net/v1",
    "CRUISE_API_KEY", "workers-ai/@cf/zai-org/glm-5.3", True,
    hint="One key across open models, with budget caps",
),

No hosted-service code and no new CLI surface (per AGENTS.md). CRUISE_BASE_URL / CRUISE_MODEL overrides come for free from resolve(). The model list stays dynamic: GET /v1/models returns only what the presented key can be served, with capabilities and prices.

We're also looking at a Models.dev listing, which would put Cruise in the "Other…" picker either way.

Would you accept a PR along these lines? Happy to change the shape, the default model or the wording.

AI-assisted: drafted with Claude Code; the test runs are described above and logs are available on request.