What Is Open Paper?
Open Paper's own homepage and About page position it as a research-library workbench built around an AI assistant, rather than a general chatbot with file upload.
- Built by Khoj Inc, whose founders, Saba Imran and Debanjum, also built Khoj, a separate open-source "second brain" AI assistant; the About page says the idea came from wanting better annotation tools and AI help without hallucinations or data concerns during Saba's own AI-safety reading
- First version launched June 2025, per openpaper.ai/about
- Maturity: the repository has roughly 0+ GitHub stars as of this check; it has published no formal Release or Tag, so the version in client/package.json and server/pyproject.toml (0.1.0) is the only version number this review could find
- What it is not: Open Paper is not optimized for self-hosting. The GitHub README states the project "isn't optimized for self-hosting" and is built primarily as a hosted service; a Docker-based self-hosting path exists in the repository, but Khoj Inc runs openpaper.ai itself as the intended way to use it
- Not to be confused with the company's other product, Khoj — see Open Paper vs. Other Research-Paper Tools below for how they differ
How Open Paper Works: Projects, Data Tables, and Zotero
Per the GitHub README and openpaper.ai/pricing, Open Paper's features beyond plain PDF viewing fall into a few groups.
- Reading and annotation — a user uploads a PDF, highlights passages, attaches comments, and can send a selection straight to the AI assistant; notes can be viewed in markdown, per the README
- Cited AI assistant — the assistant sits beside the paper and answers questions with citations linking back to the source passage; uploading a PDF generates a starter brief and starter questions automatically, per the README
- Projects — a user groups related papers into a Project to ask questions across all of them at once ("topic synthesis") and to run data-table extraction: defining a custom schema (for example, sample size or method) that Open Paper fills in for every paper in the project, with each cell tied back to its source and the whole table exportable to CSV
- Discovery — a built-in search over open-access literature lets a user find and add new papers to their library directly, per the README
- Zotero import and sync — a user can connect a Zotero account to import an existing library and keep it synced, per the README
- Audio overviews — Open Paper can generate an audio version of a paper, counted against a weekly limit on openpaper.ai/pricing
Get Open Paper
Open Paper is a hosted web app — there is nothing to install. Sign in at openpaper.ai and start uploading papers.
Path | How to get it |
|---|---|
| Website (sign in) | openpaper.ai |
| Pricing and plan limits | openpaper.ai/pricing |
| Privacy policy | openpaper.ai/privacy |
| Self-host the source (AGPL-3.0) | Docker Compose, see GitHub |
There is no stated version number on the website or in any GitHub Release — Open Paper has published no Releases or Tags as of this check. The only version signal found is 0.1.0, taken from client/package.json and server/pyproject.toml in the repository, checked 2026-11-18. A separate "Terms" page (openpaper.ai/terms) returned a 404 at the time of this check.
Privacy and Data Handling
Per openpaper.ai/privacy, checked 2026-11-18, Open Paper is operated by Khoj Inc and processes data in the United States.
- The privacy policy names the operator as Khoj Inc., with a registered address at 1111B S Governors Ave #6322, Dover, DE 19904, and states the company "transfer[s] the data to the United States and process[es] it there"
- This review did not find a stated governing-law or jurisdiction clause; the expected "Terms" page (openpaper.ai/terms) returned a 404 error at the time of this check
- Every AI feature sends the relevant text to a cloud provider — Open Paper's server-side setup requires Google Gemini, OpenAI, and Anthropic API keys, per DEVELOPMENT.md — so uploaded paper content is visible to whichever of those providers answers a given request, under that provider's own policies
- Self-hosting the open-source code is possible, but the GitHub README says the project "isn't optimized for self-hosting"; a self-hoster would supply their own cloud API keys and storage, not Open Paper's
Open Paper License and Pricing
Open Paper has two live pricing tiers today, plus a third that is announced but not launched, per openpaper.ai/pricing, checked 2026-11-18.
Tier | Price | Key limits |
|---|---|---|
| Base | $0/month | 10 uploads, 200MB, 5K chat credits/wk, 5 audio/wk, 2 projects, 2 tables |
| Researcher | $12/month | 500 uploads, 3GB, 150K chat credits/wk, 100 audio/wk, 100 projects, 50 tables |
| Teams | TBD (coming soon) | Unlimited uploads/storage/chat/audio, 500 projects, 100 tables |
License: AGPL-3.0, per the GitHub repository, checked 2026-11-18. Annotations are listed as a feature on the pricing page with no stated per-tier limit. The pricing page marks Researcher "Recommended" and Teams "Coming Soon."
Who Should Use Open Paper?
Whether Open Paper fits depends mainly on whether a hosted, cloud-AI workspace for a paper library is what you need, rather than a local or offline tool.
Use Open Paper if
- You want one workspace to read, annotate, and ask cited questions about papers without setting up any infrastructure yourself
- You want to pull the same field (method, sample size, result) out of every paper in a project into one exportable table
- You already use Zotero and want your existing library imported and kept in sync
- You are comfortable with your paper content reaching a cloud AI provider (Google Gemini, OpenAI, or Anthropic) to get an answer
Wait, or look elsewhere, if
- You need local or offline AI inference — Open Paper has no Ollama or local-model path, hosted or self-hosted
- You need the free tier's limits (10 paper uploads, 2 projects) to cover an active research workflow — that likely means the $12/month Researcher tier, since Teams is not yet launched or priced
- You need a stated company jurisdiction or legal terms page — this review could not reach openpaper.ai/terms (404) or find a governing-law clause
- You want a self-hosting-first tool — Open Paper's own README says it "isn't optimized for self-hosting," see Open Paper vs. Other Research-Paper Tools below for alternatives that are
Open Paper vs. Other Research-Paper Tools
Open Paper is one of several AI tools for working with research papers. Here is how it compares to others in the same general document-chat segment — see the Local AI Directory for the full catalog.
- PapersGPT — an AGPL-3.0, open-source Zotero plugin that chats with PDFs inside an existing Zotero library, with an MCP server for Claude and Cursor; unlike Open Paper, PapersGPT can run on a built-in local model or Ollama rather than only cloud providers. See the PapersGPT review.
- Khoj — an AGPL-3.0, open-source "second brain" AI assistant made by the same company (Khoj Inc) behind Open Paper; Khoj is a broader personal-knowledge assistant that syncs with Obsidian and Notion and does support local models (Ollama, LM Studio, llama.cpp), while Open Paper is narrower and paper-specific and runs on cloud models only. See the Khoj review.
- SurfSense — an Apache-2.0, open-source desktop research agent styled as an air-gapped NotebookLM alternative; unlike Open Paper, SurfSense's index stays on the user's own machine and documents only leave it if a remote model is chosen. See the SurfSense review.
Frequently Asked Questions
Is Open Paper free?
The Base tier is free with limited paper uploads (10), storage (200 MB), and weekly chat/audio credits. A $12/month Researcher tier raises those limits; a Teams tier is announced on the pricing page but not yet launched or priced.
Can Open Paper run without an internet connection?
No. Open Paper is a hosted web app, and every AI feature depends on a cloud provider (Google Gemini, OpenAI, or Anthropic) per its own DEVELOPMENT.md — there is no Ollama or local-inference option, hosted or self-hosted.
Is Open Paper open source?
Yes. The code at github.com/khoj-ai/openpaper is licensed AGPL-3.0. The project states it is built primarily as a hosted service and "isn't optimized for self-hosting," though a Docker-based self-hosting path exists in the repository for anyone who supplies their own cloud API keys and storage.
Is Open Paper the same product as Khoj?
No, though both are made by the same company, Khoj Inc. Khoj is a broader personal-knowledge assistant that syncs with note apps like Obsidian and supports local models; Open Paper is a narrower, paper-specific workspace that runs on cloud models only.
Does Open Paper support Zotero?
Yes. A user can connect a Zotero account to import an existing library into Open Paper and keep it synced, per the GitHub README.
