Agents & Models

OpenResearch runs your existing coding agent with your own subscription or API key. If an agent isn't installed, the app's onboarding can install it for you.

Supported agents

• Claude Code: sign in with claude auth login.
• Codex: sign in with codex login.
• OpenCode: sign in with opencode auth login, use its free models, or connect a local model.
• Cursor: sign in with agent login or set CURSOR_API_KEY.
• Google Antigravity: run agy and sign in.

Other agents that speak the Agent Client Protocol (ACP) can be added under Settings > Harnesses.

Choosing an agent

Pick the agent and model from the model picker when you start a chat. A chat keeps its agent; start a new chat to switch.

Use OpenResearch from your own terminal agent

To drive OpenResearch from an agent running outside the app, install the OpenResearch skill into it. Without --agent, it targets every agent set up on this computer; --full also installs the detailed per-topic skills.

orx install-skills
orx install-skills --agent claude --full

Local models

OpenCode can use a model served on your computer. In the model picker choose OpenCode > Add Local Model (or Settings > Harnesses > OpenCode), pick the server, click Find models, choose a tool-capable model, and save. A 32K context window is a good start.

• LM Studio: enable Developer Mode, load the model with a 32768 context length, and turn on the local server (127.0.0.1:1234/v1).
• oMLX (Apple silicon): 127.0.0.1:8000/v1.
• Ollama: 127.0.0.1:11434/v1.
• Custom endpoint: any OpenAI-compatible server (such as vLLM) that exposes /v1/models and tool calls on a loopback address. Use an SSH port forward for a remote server.

Local inference keeps model calls on your machine, but literature search and remote compute still use the network.