API keys

API keys authenticate proxy clients. Keys are SHA-256 hashed at rest and shown once on creation. Once any key exists, the proxy requires Authorization: Bearer sk-....

Each key supports:

• Rate limits — per-key RPM and TPM (token-bucket, in-memory, resets on restart).
• Model allow-lists — restrict which models a key may call.
• MCP relay allow-lists — which relays a key may use.
• System prompt — injected into matching requests.

The key detail view shows usage stats, a per-model breakdown, and recent requests.

Creating keys

When creating a key, choose:

• A display name.
• Optional allowed models. Empty means all models for normal /v1/* proxy requests.
• Optional allowed MCP relays. Empty means no MCP relay access.
• Optional RPM and TPM limits.
• Expiry: 30 days, 60 days, 90 days, 1 year, or never.
• Optional system prompt.

The secret starts with sk- and is shown only once. llama-dash stores only a SHA-256 hash and a short prefix.

Enforcement

For /v1/* requests, llama-dash accepts Authorization: Bearer sk-....

If no user-created API keys exist, /v1/* proxy requests are allowed without a key. After the first key is created, unmatched proxy traffic requires a valid key unless a matched routing rule uses passthrough auth.

API key checks include:

• Hash match using a timing-safe comparison.
• Revocation status.
• Expiry.
• RPM token bucket.
• TPM token bucket for normal proxy requests.
• Model allow-list before routing rewrites and again after routing rewrites.

TPM accounting records token usage after upstream usage data is available, so it is an operational rate limiter, not a hard preflight token budget.

Key management

The key detail view supports:

• Rename.
• Rotate. Rotation immediately invalidates the old secret and shows the new secret once.
• Revoke. Revoked keys stop authenticating but remain visible for audit context.
• Update system prompt.
• Update model access.
• Update MCP relay access.
• Inspect 30-minute activity, recent requests, and per-model usage.