Agent credentials
Configure API keys and connected plans for coding agents.
Superconductor works on a BYOK (bring-your-own-keys) basis, meaning that requests to the coding agent models are made using your own keys or credentials.
Connecting credentials
Agent API keys can be defined on the Workspace, or on the Project.
For Claude Code and Codex agents, we additionally support individual users connecting their Claude and ChatGPT premium plans instead of using API keys. This can be done either in Workspace settings, Project settings, or Account settings.
How do I track usage? If an implementation uses an API key, the implementation card's Cost value estimates the API spend for that run. If it uses a connected Claude or ChatGPT plan, use the 5-hour and 7-day usage stats shown next to the agent name instead. The card's Cost value may still appear, but it is an API-price estimate based on the agent's reported token usage, not a measure of charges or credits consumed on your plan.
Deciding which credentials to use
If an implementation agent is launched automatically (e.g. by an external PR, Slack, Email triage, or the Meeting assistant), we use API keys only, following precedence: Project key first, then Workspace key.
If the agent is launched manually by a user, we follow this precedence order:
- That user's connected plan (if any)
- Project API key (if any)
- Workspace API key
What if my connected plan hits its usage limit? Superconductor does not automatically fall back to an API key. The implementation will stop, and you'll see a message in the chat indicating that your plan's limit has been reached and when it resets. For Claude, you can disconnect your plan in your account settings and use an API key instead until the limit resets.
Best available credentials
Some features don't ask you which agent to use — they automatically run with the best coding agent you have credentials configured for. Superconductor tries them in this order and uses the first one with a working credential:
- Claude Code (Opus 4.7)
- Codex (GPT-5.5, high reasoning)
- Amp
- OpenCode (GLM 5.1)
- Gemini (3.1 Pro)
Features that use this fallback:
- Assisted Environment Setup
- Guided review
- QA Checks [Beta]
- Email triage [Beta]
- Meeting assistant [Beta]
- The fallback default selection in the profile selector when you launch an implementation, if your Default profile for integrations is unset or unavailable for the project
Benchmarking
Agent API keys are also required to include agents in Benchmark runs. Only agents with configured keys will be available for selection when setting up a benchmark. You don't need to provide keys for the evaluation models — Superconductor handles that.