Welcome to abstract

abstract is a research workbench that runs on your own computer. You talk to an AI agent in a chat; it searches the literature, reads your papers, and writes with you: literature reviews, related-work sections, answers about your sources, rebuttals, and more.

What makes it different is that citation integrity is enforced in code, not just requested in a prompt:

• Bibliographic metadata is never written by a model. Titles, authors, venues, and DOIs are copied from Crossref and OpenAlex records.
• Every cited sentence is checked. A separate verification step must quote the source word for word, and code confirms the quote is really in the passage. Each sentence is marked supported, partial, or unsupported.
• Export is gated. Exporting re-checks every DOI and stops on unsupported claims unless you explicitly accept them.

Drafting is agentic, verification is deterministic, approval is human.

Start here

InstallOne command, nothing else needed.

Quick startFrom an empty folder to a verified draft.

Connect a modelYour ChatGPT plan, your API keys, or a local model.

TroubleshootingCommon messages and how to fix them.

How the docs are organized

• Start here: install, a ten-minute walkthrough, and choosing a model.
• Guides: one page per part of the workflow: workspaces, searching, drafting, exporting, steering the agent, skills, and memory.
• Reference: the command line, configuration files, privacy, and troubleshooting.

abstract is free and open source under the Apache 2.0 license. The code is on GitHub and the package is abstract-cli on npm.