Models & Engines
CLANKER is model-agnostic. Agents run on Anthropic (Claude) or OpenAI, and the two are interchangeable peers — same setup, same billing, same key flow. No provider is special-cased.
Choosing a model — the “engine”
You don’t pick a model per message. The model (the engine) is resolved from a cascade — most-specific wins:
agent pin > workspace default > account default > deployment default| Level | Where it’s set |
|---|---|
| Agent | The agent’s definition pins a model (model: in its frontmatter). |
| Workspace | A default engine for everyone in the workspace (Workspace Settings). |
| Account | Your personal default (Settings → Default Engine). |
| Deployment | The platform default + the allow-list of enabled models. |
The resolved model’s provider also selects the sandbox coding engine — an OpenAI model runs the OpenAI Agents SDK, a Claude model runs Claude Code. There is no separate switch to flip.
Bring your own key (BYOK)
Add your own Anthropic (sk-ant-…) or OpenAI (sk-…) key during onboarding or in Settings — one field, either provider, zero markup. The key is used for both chat and skill execution, and the platform detects the provider automatically.
One gateway
All LLM traffic — managed and BYOK, Anthropic and OpenAI — routes through a single AI gateway for observability, rate-limiting, and egress control. BYOK requests carry your key through the gateway (never a direct call to the provider).
Availability is driven by keys
There are no provider flags. A provider’s models are offered iff the deployment holds a key for it — Anthropic’s managed key (in the gateway) and/or an OPENAI_API_KEY. A provider the deployment has no key for is BYOK-only: bring your own key and it lights up for you. A provider with no deployment key and no BYOK key simply isn’t shown, so you can never land on an engine that can’t run.
Workspace owners can narrow the set further for their workspace; you only ever choose from what’s available to you.