Bring your own keys (BYOK)
Point Recoder at the provider keys your agents already use. Recoder orchestrates and governs; it never proxies, intercepts, or resells inference tokens.
Recoder is bring-your-own-keys. You point it at the provider keys your agents already use. Recoder orchestrates and governs; it never proxies, intercepts, or resells inference tokens. You pay your providers directly for inference, and (on a paid Recoder plan) you pay Recoder for orchestration and control — never for tokens.
1. Your agents' own keys
When you run Claude Code, Codex, opencode, etc., the agent talks to its provider using its own auth — exactly as it does outside Recoder. Recoder launches each agent with your shell environment, so it inherits the same keys and config you normally use (ANTHROPIC_API_KEY for Claude Code, OPENAI_API_KEY for Codex, whatever opencode is configured with). Recoder does not sit between the agent and its provider.
2. Recoder-managed inference (optional)
Some Recoder features can call an LLM directly (assistive features, routing). These go through Recoder's BYOK LLM gateway, which you configure with your own provider keys. Missing keys degrade truthfully — the feature reports it's unavailable rather than silently failing or falling back to someone else's tokens. Key writes are governed, and secrets are stored encrypted, never in plaintext.
Provider keys vs platform keys
- •Provider keys (ANTHROPIC_API_KEY, OPENAI_API_KEY, …) — your AI provider credentials. These pay for inference. Recoder never resells them.
- •Recoder platform keys (sk_…) — identify you to a Recoder backend for control-plane operations. The desktop app mints one locally for its embedded backend; you don't manage it by hand for local use.