Repository navigation
Releases
CoAutoResearch 2.0
CoAutoResearch 2.0 brings autonomous investigation and human–AI collaboration into one research workflow.
An autonomous research partner that self-improves recursively and works with you.
What changes
-
Recursive research improvement: use recorded evidence, review feedback and reusable lessons to refine subsequent trials within the researcher’s brief and budget.
-
Two developer-operated example papers demonstrate the Digits and Ising research workflows, including their findings and limitations.
-
Discuss findings in a parallel chat, review a suggestion, and send it into the main research session.
-
Follow bounded trials with clearer activity, review, pause, resume and restart controls.
-
Trace recorded findings, evidence and limitations into the manuscript.
-
Generate a paper from recorded evidence using four pinned scientific skills, local figures, citations and PDF tools. Preview it with the browser’s native PDF reader, cancel or retry writing, and retain a previous ready draft.
-
Set up the dashboard and paper tools through one instruction to your existing coding agent.
-
Read the refreshed walkthrough and architecture guide.
Get started
Give your coding agent this instruction:
Set up and launch CoAutoResearch from https://github.com/YihongT/CoAutoResearch. Follow docs/agent-setup.md, reuse my existing coding-agent login, configure paper generation, verify the setup, and open the dashboard.
The attached co-auto-research-2.0.0.tgz is an installable package; its SHA-256 is supplied separately. This GitHub release does not publish or update the npm registry package.
Existing projects
Back up your research before upgrading. Follow the upgrade guide for preserved legacy history and the v2 migration workflow. Research projects, credentials and local outputs are not included in the distribution.
Validation and scope
The release archive was rebuilt from the tagged source, inspected, and installed in an isolated prefix; the installed CLI reports 2.0.0. Dependency audit reported no vulnerabilities.
Verified locally on macOS with Chrome and Codex, including research controls, collaboration and manuscript-to-PDF generation. Claude installation, authentication and model discovery were checked; Claude research inference and Linux/Windows workflows were not fully exercised. Generated research and papers require human scientific review. This release targets one trusted researcher rather than public multi-user hosting.
See the full changelog and documentation.
Known limitation
A previously observed composer clipping issue at 390px width has not received final product revalidation. Desktop use is recommended for this release. README responsiveness checks do not establish mobile dashboard correctness.