Aider on your own GPUs.

Aider on a served model through its OpenAI-compatible provider.

Where a request from Aider goes: the two settings Connect writes, the warden in the middle, then one copy of the model per GPU.

In words: Aider sends its requests to LM Warden with OPENAI_API_BASE set to the warden and OPENAI_API_KEY holding a warden key. LM Warden checks the key, speaks OpenAI Chat Completions, queues the request by key priority and hands it to the replica router, which keeps each session on the replica that already holds its cache, one model copy per GPU.

Protocol: OpenAI Chat Completions
Support: Local models only

Local models only, over Chat Completions. Aider edits through text formats, so the model needs no tool calling. The metadata file only stops Aider warning about an unknown model and sets its context size.

Requirements

Works best with at least 32,768 tokens of context.

Setup

In the console, Connect writes these files with your warden’s address, your key and the model you picked. Here they are with placeholders.

Local models

Point Aider’s OpenAI provider at this warden.

Shell

export OPENAI_API_BASE=https://your-warden/v1
export OPENAI_API_KEY=vw_YOUR_KEY
aider --model openai/your-served-model-name

.aider.model.metadata.json (.aider.model.metadata.json)

{
  "openai/your-served-model-name": {
    "max_input_tokens": 32768,
    "max_output_tokens": 8192,
    "max_tokens": 8192,
    "input_cost_per_token": 0,
    "output_cost_per_token": 0,
    "litellm_provider": "openai",
    "mode": "chat"
  }
}

Session id per launch (optional)

# Run this instead of `aider ...`: it writes a fresh session id per launch.
SID=$(python3 -c 'import uuid; print(uuid.uuid4())')
cat > .aider.model.settings.yml <<EOF
- name: openai/your-served-model-name
  extra_params:
    extra_headers:
      X-Session-Id: $SID
EOF
aider --model openai/your-served-model-name

Official documentation: https://aider.chat/docs/llms/openai-compat.html (read 2026-10-04).