Configuration Reference

View Source

All OpenResponses configuration lives under the :open_responses application key.

Provider routing

Controls which adapter is used for each model name pattern.

config :open_responses, :routing, %{
  ~r/^gpt-/ => OpenResponses.Adapters.OpenAI,
  ~r/^claude-/ => OpenResponses.Adapters.Anthropic,
  ~r/^gemini-/ => OpenResponses.Adapters.Gemini,
  ~r/^llama|^mistral|^phi|^qwen/ => OpenResponses.Adapters.Ollama,
  "default" => OpenResponses.Adapters.Mock
}

Keys may be a Regex (matched with Regex.match?/2) or the string "default". Patterns are evaluated in insertion order; the first match wins. The "default" key matches any model that no pattern matched.

Provider credentials

config :open_responses, :provider_config, %{
  openai: [
    api_key: System.fetch_env!("OPENAI_API_KEY"),
    base_url: "https://api.openai.com/v1"         # optional override
  ],
  anthropic: [
    api_key: System.fetch_env!("ANTHROPIC_API_KEY"),
    base_url: "https://api.anthropic.com/v1"      # optional override
  ],
  gemini: [
    api_key: System.fetch_env!("GEMINI_API_KEY"),
    base_url: "https://generativelanguage.googleapis.com/v1beta"  # optional
  ]
}

Put this in config/runtime.exs so secrets are read at runtime, not compile time.

Hosted tools

Maps tool names to modules that implement OpenResponses.Tool.

config :open_responses, :hosted_tools, %{
  "get_time" => MyApp.Tools.TimeZone,
  "search_docs" => MyApp.Tools.DocSearch
}

Default: %{} (no hosted tools — all function calls are external).

Middleware

List of modules implementing OpenResponses.Middleware, applied in order.

config :open_responses, :middlewares, [
  MyApp.Middleware.AuditLog,
  MyApp.Middleware.TokenBudget
]

Default: [] (no middleware).

MCP servers

Pre-configured MCP server connections available to all requests.

config :open_responses, :mcp_servers, [
  %{
    name: "docs",
    url: "https://docs-mcp.example.com",
    headers: [{"authorization", "Bearer #{System.get_env("MCP_TOKEN")}"}]
  }
]

Default: [] (no pre-configured MCP servers).

Ash domains

Required for Ash framework integration. The installer sets this automatically.

config :open_responses, :ash_domains, [OpenResponses.Responses]

Ash async

Controls Ash's internal async behaviour. Set to false in development to simplify debugging.

config :ash, :disable_async?, false

Cachex

The response cache is a standard Cachex instance named :response_cache. Configure it in your supervision tree:

# Limit to 10,000 entries, default TTL 24h
{Cachex, name: :response_cache, limit: 10_000}

See the Cachex documentation for full options.

PromEx

config :open_responses, OpenResponses.PromEx,
  manual_metrics_start: :no_async,
  drop_metrics_groups: [],
  grafana: :disabled,
  metrics_server: :disabled

See the PromEx documentation for full options including Grafana dashboard upload.

Request timeouts

The default timeout for non-streaming requests waiting for loop completion is 30 seconds. For long-running agentic loops, use streaming mode ("stream": true).

To change the non-streaming timeout, override the controller in your application and adjust the receive ... after block.

Full example

A complete config/runtime.exs for production:

import Config

config :open_responses, :provider_config, %{
  openai: [api_key: System.fetch_env!("OPENAI_API_KEY")],
  anthropic: [api_key: System.fetch_env!("ANTHROPIC_API_KEY")],
  gemini: [api_key: System.fetch_env!("GEMINI_API_KEY")]
}

config :open_responses, :routing, %{
  ~r/^gpt-/ => OpenResponses.Adapters.OpenAI,
  ~r/^claude-/ => OpenResponses.Adapters.Anthropic,
  ~r/^gemini-/ => OpenResponses.Adapters.Gemini,
  "default" => OpenResponses.Adapters.OpenAI
}

config :open_responses, :hosted_tools, %{
  "search_docs" => MyApp.Tools.DocSearch
}

config :open_responses, :middlewares, [
  MyApp.Middleware.AuditLog,
  MyApp.Middleware.RateLimit
]