Aider

Get going in 30 seconds

pip install aider-chat
export OPENAI_API_KEY=ni_live_YOUR_KEY_HERE
export OPENAI_API_BASE=https://northerninference.ca/v1
aider --model openai/vertex_ai/gemini-2.5-pro-ca

Replace vertex_ai/gemini-2.5-pro-ca with any route ID from northerninference.ca/portal (Models page). The openai/ prefix is required by Aider's provider convention; NI receives the part after it.

Aider is a CLI pair-programmer. It uses LiteLLM internally, so any OpenAI-compatible endpoint works.

Setup

pip install aider-chat
export OPENAI_API_KEY=ni_live_YOUR_KEY_HERE
export OPENAI_API_BASE=https://northerninference.ca/v1

Run

# Use a Canadian route via NI
aider --model openai/vertex_ai/gemini-2.5-pro-ca

# Use an exact route key from /models (Anthropic Claude via Bedrock, Canadian routing)
aider --model openai/bedrock/global.anthropic.claude-sonnet-4-6-global

# With a custom prompt-caching setup
aider --model openai/vertex_ai/gemini-2.5-pro-ca --cache-prompts

Note on the openai/ prefix

Aider's convention is <provider>/<model>. Because we're telling aider to treat NI as OpenAI-compatible (provider=openai), you need to prefix the model with openai/, which ends up making aider send the exact route key after that prefix to NI.

Verify

aider --message "ping"

Then check portal → Usage for the request.