SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI writingopendraft

opendraft

OpenDraft is an open-source Python engine that uses 19 specialized AI agents to generate research drafts based on real literature. It verifies citation DOIs through three major databases: CrossRef, OpenAlex, and Semantic Scholar. By default, it only retains literature indexed by at least two databases at the same time, avoiding fabricated citations. It supports generating long-form documents such as research papers, literature reviews, and thesis chapters (10,000 to 20,000+ words), and can export to PDF, Word, or LaTeX formats, with support for more than 57 languages. The project uses the MIT license, is completely free, can be self-hosted, and allows users to use their own API keys. In addition, it also provides abstract and audio summary tools, data scraping functionality, draft revision, and a quick research overview mode.

Open website

About this tool

OpenDraft is an open-source Python engine for generating source-grounded research drafts. It consists of 19 specialized AI agents that simulate the workflow of a research team and is aimed at researchers who need long-form academic documents (10,000–20,000+ words). Users enter a topic and can obtain drafts of research papers, literature reviews, or thesis chapters with citation verification.

Main capabilities:

  • Generate research papers (5-15 pages), literature reviews (20-40 pages), thesis drafts (30-80 pages), and structured reports (10-100+ pages).
  • Citations in every output are checked for authenticity: by default, a citation is retained only when at least two of CrossRef, OpenAlex, and Semantic Scholar hold the DOI; single-source results are discarded and logged.
  • Each citation records its verification source in bibliography.json, and unconfirmed citations are not displayed as confirmed.
  • Citation claim verification: CitationClaimVerifier determines whether each source supports the claim for which it is cited, returning RELEVANT, IRRELEVANT, or UNCERTAIN. IRRELEVANT citations are removed only when the confidence is above the threshold (default 0.7).
  • Supports more than 57 languages, including English, Spanish, German, French, Chinese, Japanese, and others.
  • Export formats: PDF (LaTeX-quality typesetting), Microsoft Word (.docx), and LaTeX source code.
  • Provides a TL;DR summary tool: generates a 5-point summary for any paper (thesis, key findings, methods, significance, limitations).
  • Provides an abstract audio tool: uses ElevenLabs TTS to generate podcast-style audio summaries (requires ELEVENLABS_API_KEY).
  • Data scraping: pulls statistical data from the World Bank, Eurostat, and Our World in Data and saves it as CSV.
  • Draft revision: supports AI-assisted revision of existing drafts, preserving all citations, with automatic versioning (v2, v3…), and before-and-after quality scoring.
  • Research overview mode (expose mode): quickly generates a scoped overview of a research topic, including source overview, key research teams, structured outline, complete references, and next steps, at about 3 times the speed of a full draft.

How to use:

  • Local installation takes about Runs in 10 minutes, as a command-line tool and a Python library.
  • Can also be used as a Claude Skill in agents such as Claude Code or Codex: get it from getedge.cc/skills/autonomous-research, then enter "Write a paper on <topic>" in the agent.
  • Open-source maintainers can use Codex and OpenAI models to maintain OpenDraft itself, including automated PR review, regression test generation, issue triage, release workflow automation, and more.

Use cases:

  • Researchers preparing literature reviews, journal submissions, or structured drafts.
  • Open-source maintainers building tools on a reproducible research draft pipeline.
  • Graduate students writing master's or doctoral theses.
  • Scholars who want every cited DOI verified against public databases before it is written into the draft.
  • Developers extending the agent pipeline to support custom research workflows, citation verifiers, and export formats.

Limitations and caveats:

  • OpenDraft is not intended for one-click generation of final papers, cheating, fabricating citations, or bypassing peer review, nor does it replace human researchers.
  • Citation verification only indicates that a DOI is registered and indexed; it does not guarantee that the reference supports the sentence to which it is attached; OpenAlex and Semantic Scholar both ingest Crossref metadata, so the three are not fully independent.
  • Citation claim verification is a language model judgment, not proof; the judge reads only the title and abstract, not the full text. UNCERTAIN means not checked, not passed.
  • The engine does not query arXiv as a citation database; there is no arXiv API client.
  • Strict confirmation and claim removal shrink the reference list and may cause PipelineValidationError (no citations survive); relax the search or intentionally lower the settings.
  • The citation counts cited in the README and EVALUATION.md were measured before multi-source confirmation became the default and are historical data.
  • Human review is still necessary.
  • License: MIT, 100% free and open source, self-hostable and usable with your own API keys.
  • The project is hosted on GitHub:https://github.com/federicodeponte/opendraft

Related tools

Browse category