Integrate the Martian Code Router With Aider

This document describes how to set up Aider to route all your LLM requests through Martian.

Prerequisites

Ensure you have Aider installed. To install it using pip, execute:

pip install aider-chat

Configuration

Step 1: Configure Base Settings

1. Locate or create a new YAML file named .aider.conf.yml.

2. Populate the following fields:

openai-api-base: https://api.withmartian.com/v1
openai-api-key: sk-abc1234.....
model: openai/gpt-4.1
model-settings-file: ~/.aider.model.settings.yml

• openai-api-base: Martian's base url – https://api.withmartian.com/v1. This routes all Aider requests through Martian.
• openai-api-key: Your Martian API key.
• model: Name of the default model.
• model-settings-file: Name of the model settings YAML file (created in subsequent steps).

Step 2: Configure Model Metadata

1. Locate or create a new JSON file named .aider.model.metadata.json.

2. Create a JSON object named openai/{provider-name}/{model-name} and populate the following fields:

{
  "openai/openai/gpt-3.5": {
    "max_tokens": 8192,
    "max_input_tokens": 200000,
    "max_output_tokens": 8192,
    "input_cost_per_token": 0.80e-6,
    "output_cost_per_token": 4.00e-6,
    "mode": "chat"
  }
}

• max_tokens: Soft fallback limit.
• max_input_tokens: Maximum number of input tokens. Set to the smallest amount across router models.
• max_output_tokens: Maximum number of output tokens. Set to the smallest amount across router models.
• input_cost_per_token: Estimated average input cost.
• output_cost_per_token: Estimated average output cost.
• mode: Set to chat.

Step 3: Configure Model Settings

1. Locate or create a new YAML file named .aider.model.settings.yml.

2. Populate the following fields:

- name: openai/{vendor-name}/{model-name}
  edit_format: diff
  use_repo_map: true
  reminder: sys
  examples_as_sys_msg: true
  caches_by_default: true
  extra_params:
    routing_constraint:
      quality_constraint:
        numeric_value: 0.1

Field descriptions:

• name: Set to openai/{vendor-name}/{model-name}.

• edit_format: Set to diff. This sets the edit format that the LLM is to use (the default depends on the model).
• use_repo_map: Set to true. This configures Aider to use the repo map to understand your code base.
• reminder: Set to sys. This configures Aider to take the system prompt that Aider normally uses and automatically re-inject it as the reminder.
• examples_as_sys_msg: Set to true. This configures Aider to put examples inside the system message or false to put examples into the chat history as user/assistant.
• caches_by_default: Set to true. This configures the LLM API to cache system messages and other repeated context.
• extra_params.routing_constraint: Defines routing constraints for model selection.
  • quality_constraint: Set to either:
    • numeric_value (e.g. 0.1) to specify how strict the quality requirement is.
    • model_name (e.g. openai/gpt-4o) to require a specific model for quality.
  • cost_constraint: Set to either:
    • numeric_value (e.g. 0.1) to specify how strict the cost requirement is.
    • model_name (e.g. openai/gpt-4o) to require a specific model for cost.

These follow Martian's standard OpenAI-compatible router semantics.

Common Commands

The following are common commands to run Aider when configured with Martian:

# Run Aider with default model (from .aider.conf.yml)
aider

# Run Aider with the router explicitly
aider --model openai/martian/code

# Prevent Aider from committing changes to Git
aider --no-auto-commit

# Use both flags
aider --model openai/martian/code --no-auto-commit

# Call a specific model (e.g. OpenAI GPT 4.1 via Martian)
aider --model openai/gpt-4.1