DocsGet started

Quick Start

This page takes you from "just downloaded" to "the model can answer, and it knows which workspace it is in". Model parameters, tools and skills have their own pages.

1. Install

Open https://apps.microsoft.com/detail/9MV82J9G95GS (Store ID 9MV82J9G95GS) and click Get; the protocol link ms-windows-store://pdp/?productid=9MV82J9G95GS opens the Store page directly.

The Store build is an MSIX package: the Store installs it and keeps it updated, it needs neither administrator rights nor a manual WebView2 install (Windows 10 2004+ / Windows 11 required). Two differences from the GitHub installer build:

  • Its app config is redirected to %LOCALAPPDATA%\Packages\XLBBB.RabbitChatStudio\LocalCache\Roaming\com.leonard.aichat, so it is not shared with the installer build; the .data folder inside your workspace is shared by both.
  • The in-app update check answers "up to date" for the Store build — version updates come from the Store (see Updates).

From the GitHub Release

Open Latest release and pick your platform:

Platform File
Windows Rabbit-Chat-Studio-Windows-x64-setup.exe (NSIS)
macOS Apple Silicon Rabbit-Chat-Studio-macOS-Apple-Silicon.dmg
macOS Intel Rabbit-Chat-Studio-macOS-Intel.dmg
Linux x86_64 Rabbit-Chat-Studio-Linux-x86_64.AppImage, .deb or .rpm

Package managers (not available yet)

There is no package-manager channel yet: the winget, Homebrew and Scoop manifests are not published, so winget install, brew install --cask and scoop install cannot find the package. Use the Microsoft Store or the GitHub Release installer above; the commands will be added here once the manifests exist.

First-launch notes

  • Microsoft Store build: nothing extra to do; it works on first launch. If you previously used the installer build you will see a "first run", because the two keep separate configs.

  • macOS: builds are ad-hoc signed (no Apple Developer certificate). Right-click the app → Open, or use System Settings → Privacy & Security → Open Anyway.

  • Linux AppImage: make it executable first.

    chmod +x Rabbit-Chat-Studio-Linux-x86_64.AppImage
    ./Rabbit-Chat-Studio-Linux-x86_64.AppImage
    
  • Windows: if WebView2 is missing, install the Microsoft Edge WebView2 Runtime. If you build from source and hit GNU linker errors, switch to the MSVC toolchain.

2. Configure a model

Open Settings → API & Model.

Required fields

Field Description Example
Base URL Root of the OpenAI-compatible API, usually ending in /v1 https://api.openai.com/v1
API key Key from the provider; local services may accept any value sk-...
Model Model id used by the current chat gpt-4o-mini

Common base URLs:

Service Base URL
OpenAI https://api.openai.com/v1
DeepSeek https://api.deepseek.com/v1
OpenRouter https://openrouter.ai/api/v1
Ollama (local) http://localhost:11434/v1
LM Studio (local) http://localhost:1234/v1
vLLM (local) http://localhost:8000/v1

Stop at /v1; do not append /chat/completions. The app adds the endpoint itself.

Fetch the model list

Click Fetch Models From API (the button shows a spinner while both requests are in flight):

  1. The app calls GET {base_url}/models and merges the returned ids into Model List.
  2. It also loads the capability catalogue — three public feeds (models.dev, OpenRouter and basellm/llm-metadata) merged into one flat file, <app_data>/model_metadata.json, refreshed weekly. The first build downloads roughly 7 MB and is cached for 7 days.
  3. Matched capabilities (tools, reasoning, vision, files, audio, context window) appear as badges and are stored in config.model_metadata. Names are normalised (lowercase, flattened) and matched exactly first, then by similarity.
  4. If the provider reports a context window in /models, it wins; otherwise the catalogue value is used.

A failed fetch does not block manual use: enter a model id under Add Model Manually and click Add.

Context window

The context window is used by the usage bar and the agent's context budget. Priority:

  1. config.model_context_lengths[model] (auto-filled from /models or entered manually);
  2. legacy model_settings.max_tokens;
  3. a conservative global default.

If the provider does not report it and the catalogue has no match, set the real value manually (for example 128000); otherwise long chats will compress earlier than necessary.

Save

Click Save Changes in the top-right corner. The new configuration is used by the next turn; no restart is needed.

3. First chat

  1. Return to the main window and type in the input box.
  2. Enter sends; Shift+Enter inserts a newline.
  3. While the reply streams, the send button becomes Stop; click it to cancel the turn.
  4. Reasoning-capable models show a collapsible Thought Process block above the answer.
  5. The turn footer reports what the turn cost (tokens and, when known, price) next to the latency.
  6. Type @ in the input to attach a workspace file (add #12-24 for a line range) and / to insert a skill reference, +agent to embed a sub-agent role, and @@name to address a file in .data/tmp. Images, PDFs and audio only attach cleanly when the selected model has the matching capability badge.
  7. Paste or drop a file into the composer and it is copied to .data/tmp and attached as an @ mention.
  8. A finished reply can be saved as Markdown or exported to a PDF file; drawings in the reply save as PNG (static) or HTML (dynamic).

More detail: Interface and Chat.

4. Choose a workspace

Open Settings → Workspace & Profiles and click Browse… next to Workspace Directory.

  • If unset, the default workspace is workspace/ inside the app data directory.
  • Once selected, the app creates a hidden .data/ folder there and stores the chat database, skills, memory and todos inside it.
  • Switching the workspace switches all of that state: history, skill enablement, memory and todos, and it lands on a fresh conversation.
  • If the saved path no longer exists (a moved or unmounted directory), the app falls back to the workspace in the app data directory instead of failing to start; pick the directory again from Settings.
  • If the workspace is a Git repository, add .data/ to .gitignore.

Full layout: Workspace and Data.

5. Enable capabilities as needed

None of these is required for the first chat:

Capability Entry point Tutorial
File access and command execution Tools Files and Commands
Reusable workflows (skills) Skills Skills
External MCP tools MCP MCP Servers
Sub-agents and orchestration Agents Sub-Agents
Memory / todos / timers Tools Memory, Todos and Timers

6. Where the data lives

Application-level files (config, MCP, sub-agents, profiles, logs) live in the OS app data directory:

Platform Path
Windows %APPDATA%\com.leonard.aichat\ (Roaming)
macOS ~/Library/Application Support/com.leonard.aichat/
Linux ~/.local/share/com.leonard.aichat/

Workspace-level files live in <workspace>/.data/:

<workspace>/.data/
  chat.db                 chats / request monitor / interaction log / agent missions
  memory/{user,repo}/…    memory files
  todos/<session>/…       todo lists
  skills/<name>/skill.md  skills
  skills.status           skill enablement for this workspace
  timers.json             this workspace's pending delay timers
  scripts/                run_cmd script library
  tmp/                    composer pastes + PDF mention cache
  deliver/<session>/…     sub-agent deliverables, one directory per batch

7. Next steps

  • To let the model read and write code: read Tools Overview first, then enable File Actions.
  • To run builds/tests: see run_cmd / run_shell in Files and Commands.
  • To make it follow a fixed workflow: write a Skill.
  • To connect external systems: configure MCP Servers.
  • To parallelize large tasks: configure Sub-Agents.

Quick troubleshooting

Symptom Check
401 Unauthorized API key correctness/expiry; organization headers
404 Not Found Base URL has too many or too few /v1 segments; did you include /chat/completions?
Empty model list Provider may not implement /models; add the model manually
Fetch Models spins for a while the first time The capability catalogue is downloading (~7 MB, cached for 7 days)
Attachment skipped The selected model lacks vision/files/audio capability
Slow or interrupted replies Provider rate limits, proxy, or context length; see Monitoring and Logs

Source: docs/wiki/en/quick-start.md ↗ · Docs version v3.3.3