A browser for AI. Chromium as an execution environment for agents: the browser supplies the tools, page state and permissions — the agent is the program. Ships a local model that runs offline with no API key, or use Claude, OpenAI, Ollama, or the CLI you're signed into. Also an MCP server, so your editor can drive your real browser.
See the codeA browser with an AI assistant built in. Read pages, compare tabs, and complete browser tasks using a bundled local model, a cloud provider, Codex, or Claude Code.
Website · Downloads · Security · Contributing
Get an available build for your platform from GitHub Releases.
Nabsun-<version>-x64-setup.exe to install, or
Nabsun-<version>-portable.exe to run without installing.The bundled Qwen3 1.7B model runs on your CPU without an account or API key. Windows downloads are about 1.4 GB because they include the model. Local inference works offline; accessing websites and cloud assistants requires internet access.
Windows builds are unsigned; Mac builds use ad-hoc signing without notarization.
Compare your download with the release's SHA256SUMS.txt, for example:
Get-FileHash .\Nabsun-<version>-x64-setup.exe -Algorithm SHA256
Open the assistant with Ctrl+Shift+A and ask about a page or give it a task. The built-in model is suitable for simple tasks; use a larger model for complex work.
To connect Codex or Claude Code:
Nabsun installs the official native CLI when needed and reuses existing sign-ins. No terminal commands, npm packages, or separate Node.js installation are required. Your provider account must have access to the assistant. Use Repair / update if the connection needs fixing.
Credentials remain in the provider CLI's store. Sign out also signs that CLI out outside Nabsun. Claude Code may open your system browser during sign-in.
For Anthropic, OpenAI, Ollama, or a different local model, choose a provider in Settings. Cloud API providers require an API key.
soul.md, shared only with local models by default.To let an external agent use Nabsun, copy the MCP configuration from Settings → Integrations → Connect an external agent. Browser tools use the same approval settings as the sidebar. Only install plugins and MCP servers you trust. See Security for permissions, credential handling, and limitations.
With Node.js and npm installed, run these commands from the repository root:
npm ci
npm run fetch:model
npm start
Run npm run setup if Electron did not download during installation.
fetch:model downloads and verifies the engine, model weights, and licenses.
On Linux, it downloads the weights but exits with manual engine setup instructions.
npm run dev # Watch source files and launch
npm run typecheck
npm run verify # Automated checks, including bundled local inference
After the setup above, build on the target operating system:
npm run dist:win # Windows x64 installer and portable executable
npm run dist:mac # DMG and ZIP for the current Mac architecture
Output goes to release/. Packaging verifies the bundled engine, model, and
licenses. macOS packaging requires a Mac; public signed distribution also requires
a Developer ID certificate and notarization.
Nabsun is a developer preview. There is no automatic update, profile sync, or private browsing mode. The small local model can struggle with long tasks, and website compatibility varies. Review actions before approving them.
Architecture · MIT license · Third-party notices
Created by Naveen Alavilli.
TypeScript
82.5%
JavaScript
9.7%
HTML
4.4%
CSS
3.3%
A browser for AI. Chromium as an execution environment for agents: the browser supplies the tools, page state and permissions — the agent is the program. Ships a local model that runs offline with no API key, or use Claude, OpenAI, Ollama, or the CLI you're signed into. Also an MCP server, so your editor can drive your real browser.
See the codeA browser with an AI assistant built in. Read pages, compare tabs, and complete browser tasks using a bundled local model, a cloud provider, Codex, or Claude Code.
Website · Downloads · Security · Contributing
Get an available build for your platform from GitHub Releases.
Nabsun-<version>-x64-setup.exe to install, or
Nabsun-<version>-portable.exe to run without installing.The bundled Qwen3 1.7B model runs on your CPU without an account or API key. Windows downloads are about 1.4 GB because they include the model. Local inference works offline; accessing websites and cloud assistants requires internet access.
Windows builds are unsigned; Mac builds use ad-hoc signing without notarization.
Compare your download with the release's SHA256SUMS.txt, for example:
Get-FileHash .\Nabsun-<version>-x64-setup.exe -Algorithm SHA256
Open the assistant with Ctrl+Shift+A and ask about a page or give it a task. The built-in model is suitable for simple tasks; use a larger model for complex work.
To connect Codex or Claude Code:
Nabsun installs the official native CLI when needed and reuses existing sign-ins. No terminal commands, npm packages, or separate Node.js installation are required. Your provider account must have access to the assistant. Use Repair / update if the connection needs fixing.
Credentials remain in the provider CLI's store. Sign out also signs that CLI out outside Nabsun. Claude Code may open your system browser during sign-in.
For Anthropic, OpenAI, Ollama, or a different local model, choose a provider in Settings. Cloud API providers require an API key.
soul.md, shared only with local models by default.To let an external agent use Nabsun, copy the MCP configuration from Settings → Integrations → Connect an external agent. Browser tools use the same approval settings as the sidebar. Only install plugins and MCP servers you trust. See Security for permissions, credential handling, and limitations.
With Node.js and npm installed, run these commands from the repository root:
npm ci
npm run fetch:model
npm start
Run npm run setup if Electron did not download during installation.
fetch:model downloads and verifies the engine, model weights, and licenses.
On Linux, it downloads the weights but exits with manual engine setup instructions.
npm run dev # Watch source files and launch
npm run typecheck
npm run verify # Automated checks, including bundled local inference
After the setup above, build on the target operating system:
npm run dist:win # Windows x64 installer and portable executable
npm run dist:mac # DMG and ZIP for the current Mac architecture
Output goes to release/. Packaging verifies the bundled engine, model, and
licenses. macOS packaging requires a Mac; public signed distribution also requires
a Developer ID certificate and notarization.
Nabsun is a developer preview. There is no automatic update, profile sync, or private browsing mode. The small local model can struggle with long tasks, and website compatibility varies. Review actions before approving them.
Architecture · MIT license · Third-party notices
Created by Naveen Alavilli.
TypeScript
82.5%
JavaScript
9.7%
HTML
4.4%
CSS
3.3%