Nanobrowser changelog: what's new each month
Every stable Nanobrowser release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 9 months covered; the current month updates daily.
October 2026
2 releases: 0.2.0 → 0.3.0
October brought major architectural changes to Nanobrowser, rebuilding the browser control layer and replacing the planner-navigator system with a unified agent. The new foundation supports more models, more reliable interactions, and gives users direct control over agent actions.
- Groq, Cerebras, and Llama API providers removed; saved settings and API keys deleted.
- Minimal reasoning effort option removed; previous Minimal settings converted to Low.
- Building the extension now requires Node.js 24 LTS and pnpm 10.
Highlights
- Switched to Chrome DevTools Protocol for more stable and reliable browser automation.
- Single unified agent per tab replaces the old planner and navigator system, with multiple tasks able to run in parallel.
- Ask mode (default) requires approval before the agent submits forms, sends messages, or uploads files; sensitive data like passwords and payments always require your confirmation.
- Support expanded to any model with tool calling: Anthropic, OpenAI, Gemini, OpenRouter, Grok, DeepSeek, Mistral, and any OpenAI-compatible endpoint.
- New attachments feature lets you send images, text files, or tab screenshots to the agent, which can upload them to pages and save results.
New
- Attach images, text files, or screenshots for the agent to work with.
- Voice input using OpenAI's or Gemini's voice model.
- Custom instructions to set persistent context for every chat.
- Saved prompts accessed with / shortcuts.
- Long chat summarization that preserves your full message history.
- Edit and rerun any message in a chat.
- Export chats as Markdown.
- Auto mode option to let the agent proceed without asking for approval.
Improved
- Element interactions now use real browser input with stable element IDs instead of injected scripts.
- Each action is sent once and confirms whether the page responded.
- Pick reasoning effort level and model per chat instead of globally.
- Default models updated for OpenAI, Anthropic, Gemini, Grok, DeepSeek, and Ollama.
- Ollama now connects through its OpenAI-compatible API while maintaining backward compatibility.
- Extension rebuilt as a single WXT package instead of multiple packages.
- Migrated from LangChain to Vercel AI SDK for LLM calls.
Fixed
- Ollama connection stability improved through OpenAI-compatible API.
November 2025
1 release: 0.1.13
November brought improvements to the chat interface, better language support, and several important fixes for stability and compatibility. The focus was on enhancing usability and resolving issues with complex web pages.
Highlights
- Added text file attachment support to the chat interface for easier document handling.
- Fixed a critical infinite loop affecting live nodes and sandboxed iframes on sites like Amazon.
- Replaced emoji icons with professional React icons throughout the settings interface.
- Added Portuguese (Brazil) language support.
- Improved handling of Gemini models when accessed through OpenAI-compatible API endpoints.
New
- Portuguese (Brazil) language support.
- Text file attachment support in chat interface.
- Support for Gemini models via OpenAI-compatible API endpoints.
Improved
- Replaced emoji icons with professional React icons in settings.
- Enhanced coding assistant guidance and discoverability.
- Upgraded clickable element detection to iterative implementation for better stability.
Fixed
- Fixed infinite loop affecting live nodes and sandboxed iframes.
- Fixed handling of Gemini markdown-wrapped JSON in structured output.
- Resolved recursion issues in element detection algorithm.
September 2025
1 release: 0.1.12
September brought model updates and UI improvements to Nanobrowser, along with fixes to analytics and options persistence.
Highlights
- Added gpt-5-chat-latest as a new non-reasoning model option.
- Fixed model parameter persistence so settings now save correctly in the options UI.
- Improved lightweight guardrails and system prompts for better performance.
- Added optional anonymous analytics to help inform future enhancements.
New
- gpt-5-chat-latest non-reasoning model
- Optional anonymous analytics
Improved
- Lightweight guardrails and system prompts
- Removed unnecessary identify calls to reduce useless identity events
Fixed
- Model parameter persistence in options UI (issue #207)
- Inconsistent anonymous ID bug
- Minor bugs including sponsor badge display
August 2025
1 release: 0.1.9
August brought expanded language support and improved API integrations, along with better handling of cross-origin frame access and a restructuring of agent responsibilities.
- Validator agent removed; its responsibilities now handled by planner
Highlights
- Added Llama API integration for expanded model support
- Implemented internationalization (i18n) to support multiple languages
- Improved handling of inaccessible frames blocked by CORS policies
- Transferred validator agent responsibilities to planner for streamlined workflow
New
- Llama API integration support
- Internationalization (i18n) framework
- Traditional Chinese (zhTW) locale
- CORS-blocked frame access handling
Improved
- Traditional Chinese README documentation updated and improved
- Synced with browser-use v0.5.11 for better compatibility
Fixed
- Handling of inaccessible frames from parent due to CORS restrictions
July 2025
1 release: 0.1.8
July brought internal improvements to Nanobrowser's architecture and performance. The release focused on resolving provider compatibility issues and optimizing core browser functionality.
- Historical task replay is now disabled by default; enable it explicitly if needed
Highlights
- Fixed model name collisions between different providers by storing them as provider-model pairs
- Optimized scrolling performance for browser elements
- Upgraded Puppeteer to the latest version
- Disabled historical task replay by default for safer operation
Improved
- Reorganized internal structure by moving DOM functionality into browser as a sub-package
- Enhanced scrolling element performance
Fixed
- Resolved model name collisions that could occur when using multiple providers
June 2025
1 release: 0.1.7
June brought expanded LLM provider support and improved onboarding for new users. The browser gained new capabilities for sidebar integration and audio input.
Highlights
- Added Groq and Cerebras as supported LLM providers for more model options
- New guided setup flow welcomes first-time users and eliminates API key confusion
- Opera sidebar support now available alongside Chrome compatibility
- Microphone button added for voice input capabilities
New
- Groq LLM provider integration
- Cerebras LLM provider integration
- Opera sidebar support
- Microphone button for audio input
- Guided setup flow for new users
Improved
- Onboarding experience with clearer API key guidance
- English text in placeholder messages
Fixed
- Removed invalid sidebaraction from Chrome manifest that caused compatibility issues
May 2025
2 releases: 0.1.5 → 0.1.6
May brought expanded provider support and new browser control actions to Nanobrowser. The month focused on adding cloud AI integrations and improving DOM processing compatibility.
Highlights
- Added Azure OpenAI and OpenRouter as AI provider options.
- New closetab and wait actions for browser control.
- Improved JSON extraction and zip bundle handling.
- Added guardrail and highlight capabilities.
New
- Azure OpenAI provider support
- OpenRouter provider support
- Closetab action
- Wait action
- Guardrail feature
- Highlight feature
Improved
- Synchronised DOM processing with browser-use v0.1.41
- Updated default models for all providers
- Fixed JSON extractor
- Enhanced zip bundle stream handling
- Added AbortSignal support
Fixed
- JSON extractor now handles data correctly
- String data in zip bundle streams is now processed properly
April 2025
1 release: 0.1.4
April brought bug fixes, security improvements, and enhanced interaction with web elements. The month focused on refining existing functionality and addressing stability issues.
Highlights
- Fixed Google search bug
- Added dropdown actions for select elements
- Security dependency updates to remove potential vulnerabilities
- Improved chat input handling with composition event checks
New
- Dropdown actions for select elements
Improved
- UI/UX enhancements to improve user experience
- Chat input handling with isComposing checks for better keyboard support
- Type improvements for activeTab option tracking
Fixed
- Google search bug
March 2025
4 releases: 0.1.0 → 0.1.3
March brought Nanobrowser from initial release through rapid expansion of model support and user-facing features. The month saw additions of major AI providers, local model execution, and significant UI improvements.
Highlights
- Added support for Ollama to run local LLMs like DeepSeek-r1, Qwen2.5, and Mistral-small.
- Expanded model compatibility to include Gemini 2.0, Claude 3.7, Grok, and DeepSeek official API.
- Completely redesigned Models Settings UI for easier configuration of LLM providers.
- Added dark mode with automatic system preference detection.
- Enabled custom OpenAI-compatible providers like OpenRouter.
New
- Support for Gemini 2.0 models.
- Support for Claude 3.7 and Grok.
- Built-in Ollama integration for local LLMs.
- Support for custom OpenAI-compatible providers.
- Support for DeepSeek official API.
- Dark mode with automatic OS theme adaptation.
- Per-agent customization of temperature and top-p parameters.
- Chrome Web Store link added.
Improved
- Completely redesigned Models Settings UI for better usability.
- Enhanced model selection to handle single available model correctly.
- Improved type safety for planner and validator LLM variables.
- Added recommendations for choosing optimal agent models in documentation.
- Added instructions for building from source in documentation.
- Added contributing guidelines and issue templates.
Fixed
- Fixed errors when using GPT-4o-mini.
- Fixed Gemini issues in production environment.
- Fixed model selection bug when only one model is available.
Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. Nanobrowser: pricing, features and alternatives · All changelogs