Request limits

Global request size limits are enforced as the final step of the proxy transform pipeline. Configure them on the Policies page (or via /api/settings/request-limits).

Available limits

Two global limits can be set:

• maxMessages - maximum number of entries in an OpenAI-style messages array.
• maxEstimatedTokens - maximum estimated prompt tokens for the input payload.

Both are optional. null means no limit.

Enforcement order

Limits run after:

• API-key allow-list checks.
• Routing-rule evaluation and model rewrites.
• Model alias resolution.
• API-key system-prompt injection.

This means injected system prompts count toward the estimated prompt-token limit.

Rejections

When a request exceeds a configured limit, llama-dash returns 422 with request_too_large and does not forward the request upstream.

The prompt-token limit is based on a local JSON estimate, including OpenAI messages and Anthropic top-level system and tools. Treat it as a protective size guard, not exact tokenizer output.