Nanobrowser Web Automation
A Chrome and Edge extension that runs multi-agent web automation with your own LLM API keys.
Per-dimension scores and reasoning
Evidence: README claims privacy-first, local execution, user control, but no implementation details. Security policy (SECURITY.md) and dependency overrides (package.json) exist, but no permission list or data flow description. Deductions: lack of code-level evidence for least privilege, user confirmation, data flow transparency; no rollback mechanism mentioned; publisher identity unverified.
Evidence: README and package.json provide build and run instructions, but no details on error handling or failure messages. Deductions: static review cannot verify self-consistency, dependency availability, or actual quality of failure messages; only documentation-based.
Evidence: README details target audience (users seeking OpenAI Operator alternative), use cases (news summarization, GitHub research, shopping research), and environment requirements (Chrome, Node.js, pnpm). Deductions: capability boundaries not clearly stated, trigger precision (e.g., task instruction specificity) not detailed.
Evidence: README provides installation, build, configuration instructions, includes examples and community links; LICENSE is Apache-2.0; package.json includes version. Deductions: missing changelog, known limitations section not explicit, maintenance responsibility unclear (publisher unverified).
Evidence: README claims free, open-source, privacy-friendly, offers multiple LLM support and local model options, cost-effective. Deductions: output usability (e.g., quality of generated results) lacks concrete evidence; marginal value based on claims not measured.
Evidence: README includes feature claims and community test links, but no traceable verification data. Deductions: claims lack independent verification, facts and inferences not clearly separated.
- Publisher identity is unverified; treat source with caution.
- Privacy and local execution claims in README lack code-level evidence; further review needed.
- No permission list or data flow description provided; potential for excessive permissions.
- Dependency security only partially mitigated via overrides; full dependency tree should be checked.
What does this agent do, and when should you use it?
Nanobrowser is an open-source AI web automation extension that runs in the browser and officially supports Chrome and Edge. Its side-panel chat interface accepts tasks and presents real-time execution status. Specialized agents, including Planner and Navigator, collaborate on workflows and can use different configured models. Users supply their own LLM API keys, while automation runs in the local browser; the project states that credentials are not shared with a cloud service. Follow-up questions and conversation-history management are included for ongoing browser-task interactions.
After installation, the user opens the Nanobrowser sidebar from the toolbar, adds LLM API keys in Settings, and selects models for Navigator and Planner. The user submits a natural-language task; Planner plans the work and Navigator handles web-navigation work as part of the multi-agent workflow. The side panel shows real-time status, supports contextual follow-up questions after a task, and exposes Conversation History for accessing and managing interactions. README examples include extracting the top 10 TechCrunch headlines from the last 24 hours, finding popular Python repositories on GitHub, and finding an Amazon Bluetooth speaker matching price, water-resistance, and battery-life constraints.
- A researcher who needs the ten leading TechCrunch headlines from the last 24 hours can submit that extraction request in the sidebar.
- A developer researching popular Python repositories with the most GitHub stars can use it for browser-based repository research.
- A shopper looking on Amazon for a water-resistant Bluetooth speaker under $50 with at least 10 hours of battery life can state those constraints in one task.
- An individual who repeatedly browses and gathers information across websites can assign models separately to Planner and Navigator to balance cost and capability.
- A user who wants browser-local execution with self-managed API keys can configure Ollama or another custom OpenAI-compatible provider for local models.
What are this agent's strengths and limitations?
- Planner and Navigator can use separate models, allowing model choice to be split between reasoning, navigation, and cost considerations.
- It lists support for OpenAI, Anthropic, Gemini, Ollama, Groq, Cerebras, Llama, and custom OpenAI-compatible providers.
- It is delivered as a browser extension with an interactive side panel, real-time status, follow-up questions, and conversation history rather than only a one-shot script.
- The README places execution in the local browser and uses user-provided API keys.
- Full official support is limited to Chrome and Edge; Firefox, Safari, and other Chromium variants such as Opera and Arc are listed as unsupported.
- Users must provide their own LLM API keys. The extension is free, but model usage is billed by the chosen provider.
- The documented cost-effective model setup can be less stable on complex tasks and may require more iterations.
- Building from source requires Node.js 22.12.0+ and pnpm 9.15.1+; the Chrome Web Store version may lag behind the latest release because of review timing.
How do you install or deploy this agent?
Stable install: install Nanobrowser from the Chrome Web Store, click “Add to Chrome,” and confirm. For the latest manual install, download nanobrowser.zip from the official GitHub releases page, unzip it, open chrome://extensions/ in Chrome, enable Developer mode, click Load unpacked, and select the unzipped nanobrowser folder. To build from source, install Node.js v22.12.0 or higher and pnpm v9.15.1 or higher, then run:
git clone https://github.com/nanobrowser/nanobrowser.git
cd nanobrowser
pnpm install
pnpm buildThe built extension is in dist; load it using the manual-install steps. For first use, open Settings in the sidebar, add LLM API keys, and choose models for Navigator and Planner.
How do you use this agent?
Click the Nanobrowser toolbar icon to open the sidebar. In Settings, add an API key for the provider you choose and select models separately for Planner and Navigator. Enter a specific web task in chat, such as “Go to TechCrunch and extract top 10 headlines from the last 24 hours.” Watch the real-time status while it runs, then ask contextual follow-up questions or manage prior interactions in Conversation History. For local models, configure Ollama or another custom OpenAI-compatible provider; the README recommends explicit, clean prompts and breaking complex work into detailed steps.
How does this agent compare with similar options?
The project presents itself as a free alternative to OpenAI Operator. Its stated distinctions are use of your own API keys, per-agent model selection, and local-browser execution; the README does not provide feature benchmarks or a direct comparison table.