A native, cross-platform launcher that wires any Ollama coding agent to any cloud model — in one click.

---

What is this?

Ollama ships ollama launch <agent> --model <model> to point coding agents (Codex, Claude Code, Cursor, …) at models you run locally or in the Ollama Cloud. Doing it by hand means remembering agent tokens, the exact cloud model names, and re-typing the command every time.

Llaunchpad is a tiny, good-looking GUI around that command:

• 🧩 Pick an agent from a dropdown — populated live from ollama launch --help.
• ☁️ Pick a cloud model from a dropdown — the full Ollama Cloud catalog, fetched in the background.
• 🚀 One click builds and runs ollama launch <agent> --model <model> for you.
• 🔁 Already running? It closes and relaunches the app cleanly.
• 💾 Remembers your last selection between runs.

Features

Live agent list	Parsed from ollama launch --help — always in sync with your Ollama version.
Full cloud catalog	Every model from ollama.com/v1/models, refreshed in the background every 5s.
Custom Ollama host	Point at any local or remote Ollama server; Test button probes the URL and pulls its local models.
Local model badge	Local models from your server show up first in the dropdown, tagged with a teal local badge.
Correct model names	Cloud ids are auto-normalized to launchable refs (glm-4.6 → glm-4.6:cloud, gpt-oss:120b → gpt-oss:120b-cloud).
GUI & CLI agents	GUI apps (Codex, VS Code) are opened/relaunched; CLI agents spawn in Terminal.
Terminal selector	Choose which installed terminal emulator CLI agents launch in from the Settings panel.
Install awareness	Agents whose app or CLI is missing on the machine get a not installed badge and Launch is disabled — no more silent "success" toasts when the integration isn't really there.
Codex App fix	Strips the legacy profile = line modern Codex rejects, so launches just work.
One-click restore	Restore button reverts an agent to its original profile when Ollama has a backup; disabled otherwise.
Smart "running" badge	Detects the real GUI process by bundle path — no false positives from background helpers.
Persisted state	Your last agent + model are restored on the next launch.
Native & light	Pure Rust + Slint, single ~14 MB binary, no Electron.
Cross-platform	Runs on macOS, Linux, and Windows.

Install

Prerequisites

• macOS 11+, Linux, or Windows
• Ollama installed and signed in (ollama signin) for cloud models.

Homebrew (macOS, recommended)

brew install --cask draugvar/llaunchpad/llaunchpad

Homebrew strips the quarantine flag, so the app opens normally — no right-click dance.

Download (all platforms)

From the Releases page:

• macOS — llaunchpad-macos-universal.tar.gz → unzip → move Llaunchpad.app to /Applications. (Direct download isn't notarized; first launch needs right-click → Open. Homebrew avoids this.)
• Linux — llaunchpad-linux-x86_64.tar.gz → extract → run ./llaunchpad.
• Windows — llaunchpad-windows-x86_64.zip → extract → run llaunchpad.exe.

Ollama must be installed and on your PATH (or in a standard location) on every platform.

Build from source

git clone https://github.com/draugvar/llaunchpad.git
cd llaunchpad
cargo build --release
./bundle.sh            # produces Llaunchpad.app
open Llaunchpad.app

Usage

1. Open Llaunchpad.
2. (Optional) Paste your Ollama host URL (default http://localhost:11434) and hit Test to pull that server's local models into the dropdown.
3. Choose an agent and a model from the inline dropdowns.
4. Hit Launch. The status bar confirms what was started. Click the × to dismiss it.

That's it — Llaunchpad runs ollama launch <agent> --model <model> -y under the hood and brings the agent up configured against your chosen model.

How it works

┌─────────────┐   ollama launch --help   ┌──────────────┐
│   agents    │ ◀──────────────────────  │              │
├─────────────┤                          │  Llaunchpad  │
│   models    │ ◀── ollama.com/v1/models │   (Rust +    │
└─────────────┘                          │    Slint)    │
        │  ollama launch <agent>         │              │
        └──── --model <model> -y ──────▶ └──────────────┘

• Agents come from the Supported integrations block of ollama launch --help.
• Models come from the Ollama Cloud catalog and are normalized to runnable refs.
• Launching spawns ollama launch; for GUI agents the running app is quit first, then reopened.

Supported agents

claude · codex-app · codex · vscode · cursor · opencode · copilot · droid · kimi · cline · hermes · openclaw · pi · pool

(whatever your installed ollama reports — the list is dynamic.)

⚠️ Not every cloud model supports agentic tool-calling. For coding agents prefer qwen3-coder, deepseek, glm-*, kimi-k2*, gpt-oss, minimax-m2*. Small/preview/vision models may return Invalid tool type.

Development

cargo run                 # debug run
cargo test                # unit tests (agent parsing, model naming)
cargo build --release && ./bundle.sh

Project layout:

src/
  main.rs            UI wiring, background refresh, state persistence
  config.rs          last-used selection + ollama_host (prefs.json)
  ollama/
    agents.rs        parse `ollama launch --help`
    models.rs        cloud catalog + local /api/tags + server probe
    launch.rs        spawn / quit / relaunch, Codex config fix
ui/app.slint         Slint UI + theme
assets/              icon, banner, screenshot
bundle.sh            assemble Llaunchpad.app

Contributing

Issues and PRs welcome. Good first contributions: more agent ⇄ process mappings, Linux/Windows support, a model capability filter.

License

MIT © draugvar — see LICENSE.

_{Not affiliated with Ollama. Built with ❤️ and a lot of cargo build.}