1. Installation and first launch
Download Kaoru for your operating system. Clicking a button starts the download immediately, with no intermediate pages. Then open the installer and follow the first-launch guide.
Download Kaoru
Pick your operating system. The download starts when you click (version 2.3.1).
Windows
Linux
AppImage: right-click → Properties → allow executing as a program, or run chmod +x in the terminal.
macOS
Install from source (developers)
git clone https://github.com/Dregxmoon/Kaoru-Agent.git
cd Kaoru-Agent
npm install
npm run rebuild
npx electron .
Run these commands in your development environment. Then consult the repository reference for initial configuration, voice and system-specific dependencies.
Full installation instructions ↗Provisional RAM, CPU and disk requirements (Spanish) ↗
2. Provider, model and cost
Choose your provider and model in Kaoru’s settings. Use an authorized key or a compatible endpoint. Review your provider’s data policy, quotas and costs; open-source software does not include API usage. A local model does not automatically disconnect TTS, browsing or integrations.
3. Workspace and rules
Select the project folder from the chat toolbar. Prepare its dependencies and rules: Kaoru can incorporate AGENTS.md, CLAUDE.md or .cursorrules. Avoid including files unrelated to the task. Project rules guide the work; execution permissions are resolved separately.
4. Terminal with a side chat
Create a terminal in your workspace and choose Chat to open a linked conversation without stopping the shell. Select output and use Ask Kaoru to attach it as a reviewable card before sending. Explain error prepares a draft. Single-line shell snippets from chat can be pasted without running or run after explicit confirmation. The shell uses your user permissions.
5. A useful first task
Try: “Analyze how this form is validated, identify the relevant files and propose tests; do not edit yet.” Then authorize a specific change with an observable result. Review the plan, activity and diff, and cancel generation when you need to stop the run.
6. Tests and completion
Configure your project’s commands. In smart mode, verification after recognized mutations can run quality scripts and a fallback JavaScript syntax check. Automatic focused selection recognizes tests/run-all.sh; other runners need configuration. Impact searches are approximate and do not replace integration tests.
7. Voice, browser and operating system
For a task across apps, describe the results you want to observe: ‘open the report in the editor, create a reminder and confirm it appears’. Kaoru prepares and reviews a mission, asks for initial authorization and advances result by result; when accessibility cannot verify one, it reports the pending step. Neural speech is included and needs internet; microphone input is not available in the current interface. Opening a URL uses your default browser; page inspection uses managed Chromium. Linux uses AT-SPI2 and Windows uses UI Automation; macOS coverage is partial.
8. Integrations and teamwork
Connect only authorized accounts. Choose the scopes you need in GitHub, and revoke access you no longer use. Review plugin origins and permissions. Start with a test repository before using client data, and agree on what context may leave the device.
9. Continuity and recovery
Memory and durable objectives require working local storage. Keep backups before important operations. A checkpoint does not automatically undo actions in external services. Read-only subagents can investigate in parallel; subagents that modify the workspace run serially.
10. Before a client handoff
Check functionality, diffs, tests and reported limitations. Document the provider, permissions and data handling for that installation. For commercial distribution, use an avatar you own or have appropriate rights to: March 7th assets are not covered by MIT. This website does not offer a contractual support commitment or SLA.