ai-gateway-config

Radicalbit AI Gateway - Config Generator

Generate or update the config.yaml for the Radicalbit AI Gateway based on: $ARGUMENTS

Your Task

1. Understand the request - determine whether the user wants to:

   • Generate a new config.yaml from scratch
   • Add or modify a specific section (model, route, guardrail, cache, etc.)
   • Review and improve an existing config

2. If a config.yaml already exists in the current directory, read it first before making changes.

3. Gather missing information - if the user's request is missing required fields, ask:

   • Which AI provider and model? (e.g., openai/gpt-4o, anthropic/claude-3-5-sonnet)
   • What should the route name be?
   • Are there any guardrails, caching, rate/token limits, or fallbacks needed?

4. Use the reference files:

   • examples/example-config.yaml — structural skeleton with all sections
   • references/schema.md — all valid fields, types, and constraints
   • examples/full-config.yaml — a real-world complete example

5. Write the output to config.yaml in the current working directory.

6. Summarize what was generated: list routes, models, and features configured. Flag any fields that still need real values (e.g., API keys).

Rules

• API keys must always use the !secret ENV_VAR_NAME syntax — never hardcode secrets.
• prompt and prompt_ref are mutually exclusive on a model — use one or the other.
• Guardrails are defined globally and referenced by name inside routes.
• String matching guardrails use values: [...] (a list), not a single pattern string.
• Caching requires type: exact or type: semantic — the field is mandatory.
• Embedding models are required when semantic caching is used on a route.
• Fallback can be defined at the top level or inside a route — both are valid.
• Every model_id referenced in fallback (both target and fallbacks) must also be listed in the route's chat_models (or embedding_models for embedding fallbacks). Missing this causes a runtime error.
• Route names should be kebab-case and descriptive (e.g., customer-service, internal-qa).
• For self-hosted or OpenAI-compatible models (Ollama, vLLM, OpenRouter), always add base_url to credentials and use openai/ as the model prefix.
• Advanced routing (keyword, token_length, time, budget) is defined at top level under routing and referenced by name in routes.
• budget_limiting at route level is required when using the budget routing rule.
• The mock provider (mock/gateway, mock/embeddings) can be used for testing without real API calls.