Last updated: 2026-06-29
Zero-retention routing ensures LLM providers do not retain prompts or completions for training or logging beyond the request lifecycle.
When zero-retention applies
| Path | Retention |
|---|---|
Chat (POST /v1/chat) | Standard provider retention policies; Enterprise can enforce zero-retention |
| Inline completion | Separate path with x-use-case: code-completion-only — distinct from chat |
| Enterprise confidential | Zero-retention headers on all inference for flagged orgs |
Inline completion routing
The VS Code extension sends inline completion requests with:
x-use-case: code-completion-only
This routes through a dedicated completion path with stricter retention controls and faster models (e.g. Claude Haiku).
BYOK (Bring Your Own Key)
Enterprise customers can supply their own LLM provider keys:
- Keys stored in
.env.backendon your self-hosted API - Inference routes through your provider accounts
- Your existing enterprise agreements with Anthropic/OpenAI apply
Supported providers
| Provider | Env variable |
|---|---|
| Anthropic | ANTHROPIC_API_KEY |
| OpenAI | OPENAI_API_KEY |
| Google Gemini | GEMINI_API_KEY or GOOGLE_API_KEY |
| DeepSeek | DEEPSEEK_API_KEY (Enterprise, requires COOP_LLM_ALLOW_UNAPPROVED=true) |
Attestation
Enterprise plans include a zero-retention attestation document describing:
- Which code paths use zero-retention headers
- Provider retention policies referenced
- Data flow from extension → API → LLM provider
Contact hello@coop-ai.dev for attestation copies.