Skip to main content

OpenAI

Direct connection to OpenAI's API. Broadest model selection of any single provider — if you want a multimodal flagship, OpenAI usually has it first.

What you need​

  • An OpenAI Platform account: platform.openai.com
  • A funded workspace (OpenAI requires payment-on-file before any API calls succeed; they no longer give free trial credit on new accounts)
  • An API key (starts with sk-… or sk-proj-…)

Steps​

  1. Create the key. Open platform.openai.com/api-keys → Create new secret key. Pick Restricted scope and grant only Model capabilities if you want a tighter blast radius — that's all Kenaz needs.
  2. Pin a project. OpenAI Platform organizes keys under projects so spend is attributable. Either create a project for Kenaz or pick an existing one.
  3. Copy the key. Shown once.
  4. Add it to Kenaz. Providers → Add provider → OpenAI. Paste the key, click Test, Save.

Models and what they're for​

OpenAI's canonical model list: platform.openai.com/docs/models.

  • gpt-5 — flagship. Strong at code, math, and tool use.
  • gpt-5-mini — cheaper drop-in for the same family. Good for hooks and batch.
  • o3 / o4 — reasoning-heavy models. Slower, much higher quality on multi-step problems. Use when you need it; don't pay for it routinely.
  • gpt-4.1 — older flagship, still capable, cheaper than gpt-5.

If you want vision or audio, check the model card — not every variant supports every modality. Kenaz reads capability hints and disables UI affordances that won't work on your selected model.

Pricing​

platform.openai.com/docs/pricing. Reasoning models bill on reasoning tokens (hidden CoT) in addition to input/output — the Usage view in Kenaz shows both, so a long o3 turn won't surprise you on the invoice.

Privacy posture​

  • API traffic is not used to train OpenAI's models by default. Policy: openai.com/policies/api-data-usage-policies.
  • Zero Data Retention (ZDR) is available for eligible customers — apply via your account manager.
  • Kenaz talks directly to api.openai.com over TLS.

Troubleshooting​

  • 401 Invalid API key — common when copying from the dashboard if a trailing space or newline came along. Re-paste from the original.
  • 429 You exceeded your current quota — your workspace's usage is over its hard limit, or you have no payment method. Check Settings → Billing on the platform.
  • model_not_found — your project doesn't have access to that model. Some models are tier-gated and only unlock once your workspace hits a usage threshold; see platform.openai.com/docs/guides/rate-limits.
  • Streaming cuts off mid-response. OpenAI occasionally drops connections on long generations; Kenaz auto-retries with exponential backoff but will surface the partial response with a "Generation interrupted" banner. Hit Retry on the message.