Aider with Claude, GPT and Gemini

Connect in three fields

Aider talks to any OpenAI-compatible endpoint through two env vars and the litellm openai/ prefix.

Setup

export OPENAI_API_BASE=https://api.avots.ai/openai/v1
export OPENAI_API_KEY=av_mcp_YOUR_KEY
aider --model openai/anthropic/claude-sonnet-4.6

The openai/ prefix is mandatory, it tells litellm (the Aider backend) to use the OpenAI-compatible /chat/completions path, while the rest, anthropic/claude-sonnet-4.6, is the model id passed through. Aider may print a harmless model not known warning; set limits in .aider.model.settings.yml if defaults are wrong.

Example request

curl https://api.avots.ai/openai/v1/chat/completions \
  -H "Authorization: Bearer av_mcp_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-4.6",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

The same key also unlocks MCP tools

Your av_mcp_ key is not only for chat. avots also runs as an MCP server, so Aider can call avots creative tools, image, video, music, talking avatars, on the same key and the same balance.

MCP setup guide →

Models and prices

Set the model field to any id below, or a lazy name like claude / gpt / gemini that maps to a sensible default.

* Price in ⚡ per 1,000,000 tokens, input / output (1000⚡ ≈ $1). Free models need a positive balance. Full list via GET /openai/v1/models.

Questions

Other clients