Research Engines

Goal

Understand why Thalix Radar runs research locally on your machine, which engines are supported, and how your credentials never leave your machine.

Use Cases

—Choosing which research engine to use for repository analysis
—Understanding the security model: why credentials stay local
—Setting up engine authentication before running radar add or radar discover
—Troubleshooting engine detection and configuration

Prerequisites

•A Thalix Radar account (from Getting Started)
•At least one of: Codex CLI, Claude Code, omp, or LM Studio

Why Local

Research runs **locally, on your own machine**, through a coding agent you already have installed.

Radar never reads that engine's credentials. What reaches the hosted service is the finished guide, not your API keys.

Your Anthropic, OpenAI, Alibaba Cloud, or localhost API keys remain on your machine. The CLI talks directly to those providers; the Radar server only receives the structured guide after research completes.

Supported Engines

Codex CLI

Priority 1

First choice. Auto-detected when authenticated. Uses your configured OpenAI or Anthropic keys.

Claude Code

Priority 2

Second choice if Codex is unavailable. Uses your Claude API key configured in Claude Code.

Install Claude Code and authenticate with your Anthropic API key.

omp (oh-my-pi)

Priority 3

Third choice. Uses Qwen models via an Alibaba Cloud plan. Set RADAR_OMP_MODEL to choose a model (default: qwen3.8-max).

Install omp and authenticate with your Alibaba Cloud credentials. The CLI will use the model configured in your omp settings.

LM Studio

Priority 4

Fourth choice for fully local inference. Requires a running LM Studio server on localhost.

Start LM Studio server. Set LMSTUDIO_URL and LMSTUDIO_MODEL if non-default. The CLI auto-detects a running server on the default port.

Commands

radar status

Shows which engines are authenticated and available. Green checkmarks indicate ready engines.

radar add <repo> --engine codex

Override auto-detection: force research through a specific engine.

RADAR_ENGINE=claude radar add <repo>

Set engine via environment variable (alternative to --engine flag).

Engine Selection

The CLI tries engines in order: Codex → Claude Code → omp → LM Studio. Use --engine or RADAR_ENGINE to override. The engine is displayed at the start of every research run.

Security Model

•All research commands (add, discover, watch, content update, pulse) run through the selected engine.
•The engine communicates directly with its configured provider (OpenAI, Anthropic, Alibaba Cloud, localhost).
•Only the final structured guide is sent to the Radar API for cataloging.
•Your API keys, conversation history, and research prompts never leave your machine.