config.example.toml

# Example operator config for the standalone CLI chat server.
# Start with: CHAT_CONFIG=config.toml mix run --no-halt

port = 8080

# The OpenAI-compatible chat-completions endpoint — Ollama, vLLM,
# LM Studio, llama.cpp, LiteLLM, or a compatible cloud API. Accepted with
# or without the trailing /chat/completions; a query string is passed
# through (e.g. Azure's ?api-version=...).
endpoint = "http://localhost:11434/v1"

default_model = "llama"

# Optional. Without it, no Authorization header is sent (local engines).
# Three forms — inline, from the environment, or from a command's stdout
# (e.g. a password manager):
#
# api_key = "sk-..."
#
# [api_key]
# env = "CHAT_API_KEY"
#
# [api_key]
# cmd = { run = "op", args = ["read", "op://vault/item/key"] }

# Optional. Replaces the default `Authorization = "Bearer {{api_key}}"`
# entirely — spell out every auth header you need. {{api_key}} expands
# wherever it appears.
# [request.headers]
# api-key = "{{api_key}}"

# Optional extra chat-completions body fields for engines that take them.
# stream/model/messages/tools are owned by the server and rejected here.
# Most engines (Ollama, vLLM, ...) only report token usage on streams
# when asked:
[request.body]
stream_options = { include_usage = true }

[[models]]
alias = "llama"
name = "Llama 3.2"
description = "Local llama3.2 via Ollama"
model = "llama3.2"

# Optional server-side web tools, the same providers the hosted service
# uses. Each configured key enables its tool for every chat turn:
# brave_api_key enables web_search (Brave Search API), firecrawl_api_key
# enables web_scrape (Firecrawl). Keys take the same three forms as
# api_key (inline string, { env = ... }, or { cmd = ... }).
# [web_tools]
# brave_api_key = { env = "BRAVE_API_KEY" }
# firecrawl_api_key = { env = "FIRECRAWL_API_KEY" }