🔧 How to Route AI API Calls Through a Local Model

Redirect cloud AI API traffic (OpenAI, Anthropic, Gemini) to a local Ollama instance or another backend, transparently to the application.

Prerequisites

HTTPS interception already set up for the target container (see HTTPS interception tutorial)
Ollama (or another OpenAI-compatible backend) running, e.g., on localhost:11434

1. Open the Routing Panel

Press R to open the Inference Routing panel.

2. Select a Preset Route

cella comes with built-in presets:

Source	Backend	Description
`api.openai.com`	`localhost:11434`	OpenAI → local Ollama
`api.anthropic.com`	`localhost:11434`	Anthropic → local Ollama
`api.business.githubcopilot.com`	`localhost:11434`	Copilot → local Ollama
`generativelanguage.googleapis.com`	`localhost:11434`	Gemini → local Ollama

3. Enable a Route

Navigate to the desired route and press Enter to toggle it on. The route takes effect immediately.

4. How It Works

When the MITM proxy intercepts a request for a routed domain, it rewrites the upstream destination to the configured backend. The application sees a normal HTTPS response — no code changes needed.

💡 All presets route to local Ollama by default. Ollama's OpenAI-compatible endpoint (/v1) handles requests from OpenAI, Copilot, and Gemini clients natively. The Anthropic route needs an API format adapter since Anthropic uses a different request schema.

← Block Syscalls Next: Export Security Policies →