Quickstart

Install the app

Download the build for your platform from the GitHub releases page.

# AppImage
chmod +x Anymo_*.AppImage
./Anymo_*.AppImage

# Debian / RPM
sudo apt install ./Anymo_*.deb
sudo dnf install ./Anymo_*.rpm

open Anymo_*.dmg   # then drag Anymo.app to Applications

First launch

Anymo opens to its home screen: "What should we build in your workspace?" A first-run banner lists anything left to set up. Pick or create a workspace, any folder where the agent should work; all of Anymo's state for it stays in that folder, local-only.

You can try the full loop immediately with the built-in mock provider: no API key, no network. It runs a real recorded scenario so you can watch a run, expand its timeline, and see an artifact land before any model is configured.

Connect a model

Open Settings → Model provider. Anymo speaks the OpenAI-compatible chat API, so anything that serves it works: local model servers, routers, proxies, or hosted APIs.

1. Pick a preset or enter a base URL

   Presets cover Ollama (http://localhost:11434/v1), LM Studio (http://localhost:1234/v1), OpenRouter-style routers, and generic hosted endpoints. Local servers usually need no key.

2. Enter the model, and a key for hosted endpoints

   Saved keys are sealed at rest, encrypted with a master secret in your OS keychain. They never appear in logs, events, or the UI.

3. Save and check

   Each saved profile has a health check that confirms the endpoint is reachable. Profiles persist across restarts, and your fallback chain spans them automatically if an endpoint fails mid-run.

Full detail, including environment variables for headless use: Model providers.

Your first run

1. Describe the work

   Type a goal into the composer: "summarize my notes", "research X and write a short report". The permission pill controls how much freedom the run gets; the default auto mode lets safe actions flow and asks before risky ones.

2. Watch it work

   Progress shows as quiet "Ran tool" lines, collapsing into a "Worked for Xs" summary you can expand into the full timeline. For parallel work, toggle Hive to watch the team live.

3. Decide when asked

   If the run needs something risky, it parks and asks. Approve or deny; either way it is recorded. See Approvals.

4. Review what it learned

   Outputs are saved as artifacts in your workspace. If the run proposes memories or a reusable skill, the From this run panel lets you accept or reject each one.

Where to go next

• Swarm mode: put a team of agents, up to 300, on one goal.
• Goals: set a finish line and let it keep working until an independent evaluator agrees.
• Skills: turn runs that worked into slash commands.
• Dreams: review what the agent wants to remember.
• Automations: schedule recurring agent work.
• Plugins and connectors: give the agent more tools.
• Worktrees: let coding agents work without touching your checkout.

Working from source

Contributors, Windows users, and anyone who wants the headless CLI can run from a clone. Prerequisites: a recent Rust toolchain (1.80+), Bun, and the platform WebView runtime Tauri v2 needs (WebView2 on Windows, WKWebView built into macOS, WebKitGTK on Linux).

git clone https://github.com/ychampion/anymo
cd anymo
cargo run -p anymo-cli -- setup     # verify environment, scaffold a demo workspace, run the offline mock demo
cargo run -p anymo-cli -- doctor    # grouped OK/WARN/FAIL report with a "what to fix" list
bun install
bun run dev:desktop                 # launch the app in dev mode

setup finishes by telling you exactly what to do next, and doctor is offline-first: a missing provider is a warning, never a failure, because the mock demo always works. The same kernel is fully scriptable; see the CLI reference and the App-server API.

Verify a checkout end to end:

cargo test --workspace --exclude anymo-desktop
cargo run -p anymo-evals -- run --all
bun run launch:gates

If something breaks

Run the in-app diagnostics or anymo doctor; every check has a stable id and a one-line fix. The most common ones are expanded in Troubleshooting.