A reference manager that actually lets your AI read the papers
Agentero is a local-first research workbench that unifies your papers, notes, and AI context so your own agent can actually follow the literature.

What it does
Agentero is a desktop research workbench built in Tauri and React that wraps your PDFs, annotations, and Markdown notes into a local Vault. It reads like a hybrid of Zotero and Obsidian, but every feature—from importing papers to translating passages—is designed with an AI agent in the loop. You bring your own local agent via the Agent Client Protocol, and the app keeps your research context grounded in one place instead of scattering it across browser tabs and chat windows.
The interesting bit
The unusual part is the ACP integration: Agentero does not ship with a model or lock you into a vendor. It treats your agent as a native participant that can install Skills, answer questions about highlighted text, and help manage citations. That BYOA philosophy, combined with built-in MCP support to plug into external clients like ChatGPT, makes it a research environment built around agents rather than a reference manager with a chat sidebar.
Key highlights
- Local-first Vault stores papers and notes; syncs via S3-compatible storage or SSH tunnel to remote libraries
- Imports from Zotero (preserving tags and attachments), Cool Papers, ModelScope, RSS feeds, arXiv recommendations, and browser plugins
- Exports BibTeX / BibLaTeX to slot into existing LaTeX writing workflows
- Parses PDF figures, tables, formulas, and algorithms with contextual understanding for agent consumption
- Cross-platform desktop app (macOS, Windows, Linux) with theming and bidirectional Markdown linking
Caveats
- Linux desktop support is currently limited to Ubuntu 22.04 and newer because of webkit2gtk 4.1 dependencies
- The project is Chinese-first: the README, screenshots, and default docs are in Chinese, with English as a secondary option
- It expects you to bring your own local agent via ACP; there is no bundled AI model
Verdict
Worth a look if you want your agent to actually participate in literature reviews rather than just summarizing pasted text. Skip it if you need a zero-config reference manager or run an older Linux distro.
Frequently asked
- What is poco-ai/Agentero?
- Agentero is a local-first research workbench that unifies your papers, notes, and AI context so your own agent can actually follow the literature.
- Is Agentero open source?
- Yes — poco-ai/Agentero is open source, released under the MIT license.
- What language is Agentero written in?
- poco-ai/Agentero is primarily written in TypeScript.
- How popular is Agentero?
- poco-ai/Agentero has 515 stars on GitHub.
- Where can I find Agentero?
- poco-ai/Agentero is on GitHub at https://github.com/poco-ai/Agentero.