- Category
- AI Agents
- Rank
- No. 1034Tools index
- Pricing
- Open Source
- Type
- AGENT
- Builder
- badlogic
- GitHub
- 805 stars
- Latest release
- v1.0.0
- Date
About
AI assistant that lives in your browser. You guide, it executes — automate repetitive web tasks, extract data from any site, and transform it into whatever you need.
What it does
Sitegeist adds a conversational side panel to Chrome or Edge. A person steers an agent through browsing, page interaction, research, form entry, and document creation. It keeps chat sessions locally, supports attachments, and lets users save site-specific interaction knowledge as reusable skills. Models can come from several hosted providers or a local Ollama setup.
Why it's ranked here
The strongest idea is guided browser work with reusable site knowledge, not unattended automation. Users can correct element selection, preserve learned procedures, resume sessions, and inspect model costs. Broad provider support also reduces lock-in. The repository shows serious handling of browser constraints, but setup dependencies, proxy caveats, and a high browser-version requirement narrow its appeal.
What's good
Provider choice is unusually broad: API keys, supported subscription logins, and local Ollama models all fit the same interface. Sessions, attachments, settings, and credentials are stored locally. Saved skills associate instructions and JavaScript libraries with matching domains, turning repeated discovery into reusable procedures. Session locking prevents two browser windows from editing the same conversation, while token and cost totals remain attached to saved sessions.
Tradeoffs
It requires Chrome 141 or an equivalent Edge release, plus user-script and file-URL permissions for development loading. Some subscription logins need a configurable CORS proxy, and the supplied tutorial says requests may pass through the default proxy when certain credentials or providers require it. Local storage does not mean prompts stay local: chat text and attachments go to the selected model provider. Contributor setup also requires two sibling repositories.
How to use it well
Use it when browser research or data collection benefits from human checkpoints, repeatable site procedures, and outputs such as Markdown, PDF, Word, HTML, or spreadsheets. It best suits users comfortable choosing a model provider and supervising page actions. Save stable workflows as skills after testing them interactively. It does not replace an unattended scheduler, a general integration API, or a fully local workflow unless paired with Ollama.
Technical notes+
package.json defines a private ESM TypeScript browser-extension package built with esbuild and Tailwind CSS, with local file dependencies on sibling mini-lit and pi-mono repositories. src/sidepanel.ts constructs the agent, model selection, local session persistence, usage accounting, tools, and custom renderers. src/background.ts and src/utils/port.ts coordinate side-panel connections and per-window session locks through Chrome runtime ports and session storage, including reconnect-and-retry behavior. src/oauth/index.ts supports OAuth login and token refresh for Anthropic, OpenAI Codex, GitHub Copilot, and Google Gemini CLI credentials. src/tools/skill.ts stores domain-matched skills and validates injected JavaScript inside a sandboxed iframe while rejecting navigation logic. src/debug.ts provides manual prompt scenarios, an element picker, and a JavaScript REPL rather than evidence of an automated test suite.
Observed
- License
- AGPL-3.0
- Primary language
- TypeScript
- Package format
- Private npm package using ECMAScript modules
- Install surface
- Chrome and Edge browser extension distributed through Sitegeist download links or loaded unpacked from a local build
- Interface
- Browser side panel with conversational agent controls
- Platform requirement
- Chrome 141 or equivalent Edge release
- Model access
- API keys, supported subscription OAuth logins, and local Ollama models
- Build dependencies
- Development requires sibling mini-lit and pi-mono repositories
Read from README.md, package.json, src/debug.ts, src/icons.ts, src/global.d.ts, src/sidepanel.ts, src/tutorials.ts, src/background.ts, src/utils/port.ts, src/oauth/index.ts, src/tools/index.ts, src/oauth/types.ts, src/tools/skill.ts, src/utils/logger.ts, src/types/three.d.ts.
What it can do
Automate repetitive web tasks
User guidance and web page interactions → Completed automated web workflows
Extract data from websites
Any website URL or web page → Structured data from the site
Transform extracted data into different formats
Raw web data and desired output format → Data in specified format (CSV, JSON, etc.)
Execute browser-based commands
User instructions and target web elements → Performed browser actions
Navigate and interact with web forms
Form fields and user-provided data → Completed form submissions
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.