Data & AI · wired today
Drive OpenAI with governed agents
Models and datasets are infrastructure now, and agents need eyes on them. The OpenAI connector is one of 222 executable connectors and gives agents 2 tools, with the credential held in the vault and every call receipted.
What agents can run against OpenAI
| Tool | What it does |
|---|---|
list_models | List the GPT, embedding, and other models available to this API key, with their ownership and creation timestamp |
chat_completion | Send a chat completion to a GPT model. This spends the account's token usage, so it is HITL-gated and never fired by the shape-proof harness (write:true means blind proof always SKIPs it; a real credential would also need --include-writes) (write) |
What the OpenAI connector needs
You provide API key from your OpenAI account; Quox sends it as a bearer token. The credential is stored encrypted in the vault and resolved just-in-time, so agents use it without ever seeing it.
Where to get it: OpenAI's own API documentation ↗ covers creating and scoping the credential.
How proven is this?
This connector was blind shape-proven against the real OpenAI API: every endpoint it declares answered measurably differently from a deliberately bogus control request, without any account or credential involved. It is Beta: proven against the live API surface, not yet run with a real credential.
- Credentialed The vault stores, tests and just-in-time resolves its credential; agents never see the secret.
- Approvable Its write tools execute only after a human approves them in the inbox.
- Agent-owned No dedicated agent owns it yet; the orchestrator can drive it on request.
- Schedulable Not yet exposed in the workflow builder; that integration is on the connector roadmap.
- Evidenced Per-call evidence envelopes are on the roadmap; vault credential use is already receipted.
OpenAI integration FAQ
What can Quox agents do with OpenAI?
Agents can run 2 tools against OpenAI: list models, chat completion. Every call is receipted in the evidence trail.
What does the OpenAI integration need?
You provide API key from your OpenAI account; Quox sends it as a bearer token. The credential is stored encrypted in the vault and resolved just-in-time, so agents use it without ever seeing it.
Do agents see my OpenAI credentials?
No. Credentials live encrypted in the Quox vault and are resolved server-side at call time. Agents invoke tools; they never receive the underlying secret.
How proven is the OpenAI connector?
This connector was blind shape-proven against the real OpenAI API: every endpoint it declares answered measurably differently from a deliberately bogus control request, without any account or credential involved. It is Beta: proven against the live API surface, not yet run with a real credential.