poco-ai/Agentero

Agentero – A One‑Stop Research Workspace for Human‑Agent Collaboration

What it is

• Agentero is a desktop application (macOS, Windows, Linux) that lets researchers read PDFs, take Markdown notes, and interact with local AI agents—all while keeping the data in plain files on the user’s machine.
• It is built around two open protocols:
  • ACP (Agent Client Protocol) – lets any locally‑run LLM (Claude, OpenAI, Codex, etc.) act as a “research assistant”.
  • MCP (Model Context Protocol) – exposes the local paper library as a context source for external AI clients (e.g., ChatGPT web, Claude Desktop).

Why it exists

• Traditional tools (Zotero, Obsidian, web‑based chat windows) keep papers, notes, and LLM interactions in separate silos, forcing users to copy‑paste.
• PDFs are hard for models to parse because of multi‑column layouts, tables, and LaTeX formulas.
• Agentero bridges that gap: it extracts structured Markdown/LaTeX from PDFs and makes the content instantly available to an attached agent.

Core features

1. Structured reading & note‑taking
   • PDF viewer with outline navigation, smooth text selection, highlighting, and full‑text search.
   • Automatic extraction of equations, pseudo‑code, and tables into clean Markdown/LaTeX.
   • WYSIWYG Markdown editor with bidirectional links ([[link]]), slash commands, and live LaTeX rendering.
   • Built‑in translation modes (word‑level, page‑by‑page, full‑document) that can use custom API keys.
2. Open agent integration
   • Connect any local LLM via the ACP protocol; no vendor lock‑in.
   • Run an MCP server so external AI clients can query the local library for context.
   • Headless CLI (agentero) for scripting bulk imports, metadata extraction, and note generation.
3. Academic radar & paper discovery
   • Integrated browsing of Cool Papers and ModelScope community feeds; one‑click import.
   • RSS subscription for labs, blogs, and pre‑print servers.
   • Daily arXiv recommendations based on the user’s existing collection.
   • Automatic parsing of reference lists, bulk fetching of cited PDFs, and citation‑graph tracking.
4. Zotero compatibility & transparent local storage
   • One‑click migration of an entire Zotero library (tags, notes, attachments) into Agentero’s folder structure.
   • All data stored as plain PDFs and Markdown files; no hidden databases.
   • Optional sync to any S3‑compatible bucket or via SSH tunnel for remote access.

Typical workflow

1. Import papers (directly from Zotero, Cool Papers, RSS, or manual upload).
2. Read a PDF, select a passage, and ask the attached agent to explain, summarize, or generate a note.
3. The agent works on the structured Markdown/LaTeX context, reducing hallucinations.
4. Save the generated note alongside the paper; link it with [[double‑link]] for easy navigation.
5. Use the CLI or ACP to automate bulk tasks (e.g., nightly arXiv fetch, mass metadata update).

Getting started

• Desktop client: download the installer from the website (agentero.app) or GitHub Releases. macOS users can also install via Homebrew (brew tap poco-ai/agentero && brew install --cask agentero). Linux requires Ubuntu 22.04+ with webkit2gtk 4.1.
• CLI: brew tap poco-ai/agentero && brew install agentero; then run agentero --help for commands.

License

• Open‑source under the MIT License.

Who might benefit

• Researchers who already use Zotero/Obsidian and want a tighter loop with LLM assistants.
• Teams that need a reproducible, file‑based knowledge base that can be synced across devices.
• Developers building custom AI agents who need a standard protocol (ACP/MCP) to access a local paper corpus.

---

All details above are taken directly from the repository’s README.