PluelyPluelyDocs

Guides

How to Connect a Local AI CLI to Pluely

Route Pluely's questions through a CLI you already run — Claude Code, Gemini CLI, OpenAI Codex, Qwen Code, or Ollama — so answers come from a tool on your own machine. Once connected, your CLI sits at the top of the overlay's model picker.

What "Your CLI" means in Pluely

The overlay's model picker is split into groups — Your CLI, Web search / Pluely models, and Your providers. "Your CLI" is a local AI command-line tool installed on your machine that Pluely connects to directly. While a CLI is connected, it's the active choice: answers run locally through it, not through any hosted model.

Supported CLIs

  • Claude Code — Anthropic's CLI; Sonnet / Opus / Haiku on your Claude subscription.
  • Gemini CLI — Google's CLI with a generous free tier.
  • OpenAI Codex — GPT-5.1-class models on your ChatGPT plan.
  • Qwen Code — free Qwen3-Coder models.
  • Ollama — open models running fully offline on your hardware.

Step-by-step: connecting a CLI

  1. Confirm the CLI is installed and working from your terminal (sign in there first if it needs an account).
  2. Open the dashboard (Home icon in the overlay's top-left) → Dev space → Connect your CLI.
  3. Pluely detects installed CLIs automatically and shows each one's version — no paths to type. Just installed one? Tap the refresh button beside the section heading to re-scan.
  4. Click Connect on the CLI you want.
  5. A Model panel appears under the connected row: pick a well-known model, type a custom model id, or leave Default (CLI decides). Ollama is the exception — picking one of your locally pulled models is required.
  6. Open the overlay's model picker — your CLI is the first entry, shown with its model (e.g. Claude Code · opus).

Screenshots and thinking models

  • The Model panel tells you whether the current setup can see images: Claude Code, Gemini CLI, OpenAI Codex, and Qwen Code are vision-capable; Ollama depends on the pulled model (llava, gemma3, qwen-vl and other multimodal families).
  • Reasoning/thinking models work on every CLI — the chain-of-thought is stripped automatically so you only see the final answer.

Why local-first matters

A connected CLI runs on your own machine, so requests never have to leave your device to reach Pluely's managed infrastructure. If you already pay for or maintain a CLI subscription, Pluely becomes a convenient overlay on a tool you already trust — no new managed model in the loop.

Switching back to managed models

Open the model picker and select any option under "Pluely models" or "Your providers" — that disables CLI mode for now, and reconnecting is one click in Dev space. To remove the connection entirely, use Disconnect on the CLI's row.

Troubleshooting a CLI that won't answer

  • Run the CLI in your terminal, outside of Pluely — if it fails there (expired login, missing subscription), fix it there first; Pluely follows.
  • Pluely surfaces the CLI's own error message when something goes wrong, so the fix is usually named right in the overlay.
  • Just installed or updated a CLI? Tap the refresh button beside "Connect your CLI" to re-detect it.
  • Picked a custom model id that errors? Switch back to Default (CLI decides) — it always works.
  • Gemini CLI's "trusted directory" requirement and Codex's git-repository check are handled by Pluely automatically — no flags needed.

Tips

  • Claude Code, Gemini CLI, OpenAI Codex and Qwen Code require their own accounts — Pluely doesn't manage that billing; it connects to what's already installed.
  • Ollama is a strong choice for fully offline, zero-cost local inference.
  • Keep your CLI updated independently; Pluely calls whatever version is on your machine.
  • If you switch between a CLI and managed models often, the model picker at the top of the overlay makes it one click.

Last updated 2026-07-10 · pluely.com