AI

External AI agents

Invite Claude Code, Codex or OpenRouter into the chat, one agent at a time, with explicit permissions and an encrypted API key.

Icon for the External AI agents module in PANACHES
App access
External AI agents
Target module
AI · Agents
Reference
Claude Code, Codex, OpenRouter: cloud power invited on demand.
Screenshot of the PANACHES External Agents settings with Claude Code, Codex and OpenRouter.

Module overview

Cloud power, only when you invite it

Beyond the local Amber AI engine, PANACHES can talk to recognized external agents: Claude Code (Anthropic) and Codex CLI (OpenAI) for tool-using code assistance through their command-line interfaces, and OpenRouter to access many cloud models with a single API key. Each integration is strictly opt-in: nothing is enabled by default, the agent's presence on the machine is detected automatically, and the permission level (read-only, workspace write, full access) is set agent by agent. A custom OpenAI-compatible endpoint (Ollama, LM Studio, enterprise provider) can also be added from settings, with the API key encrypted at rest.

What the module brings

  • Invite a tool-using CLI agent (Claude Code, Codex) for code assistance, with an explicit permission level.
  • Keep every agent read-only by default in the general chat: editing a file requires an explicit turn and a dedicated permission.
  • Access many cloud models through OpenRouter with a single API key.
  • Add your own OpenAI-compatible endpoint (Ollama, LM Studio, enterprise provider), with the key encrypted at rest.
  • Switch from one agent to another from the chat's provider selector, without changing windows.

In practice

What the module brings.

Claude Code (Anthropic), Codex CLI (OpenAI) and OpenRouter are strictly opt-in: nothing is enabled by default, every agent is automatically detected on the machine, and stays read-only in the general chat until write permission is explicitly granted.

For : Developers and teams who want to combine local AI with cloud power, without losing control

Capabilities

What you can do.

  • Opt-in per agent
  • Read-only to full access
  • Custom OpenAI endpoint

Module overview

Real situations

Tool-using code review

Enable Claude Code from settings, then invite it into the chat for read-only code assistance.

Comparing engines

Switch between the local Amber AI, Codex and OpenRouter from the provider selector, depending on the task at hand.

Internal infrastructure

Add the endpoint of an internal company Ollama or LM Studio server, with its own encrypted API key.

Granting temporary access

Switch Claude Code to workspace-write for the duration of a refactor task, then return it to read-only.

Workflow

A concrete use case.

  1. Open AI settings, External Agents tab
  2. Enable Claude Code, Codex, OpenRouter or add a custom endpoint
  3. Choose the agent's permission level (read-only by default)
  4. Select the active agent from the chat's provider selector
  5. Grant explicit write permission if the agent needs to edit a file

PANACHES connections

Amber AI Code editor Word Excel Settings

Same family

Related modules.

Icon for the Amber AI module in PANACHES

Amber AI

PANACHES' built-in AI engine to summarize, write, code and document, with conversation memory and a RAG knowledge base over your own sources.

Explore