Description
The unsettling thing about letting AI help with research is not knowing whether a conclusion was found or invented. How was this figure produced, on which data, in what environment? If you cannot see that, the result cannot go into a paper. Reproducibility and provenance are what separate research AI from a chatbot.
AIPOCH Open-Science is an AI research workbench that runs on your own machine — one-click install on macOS, Windows and Linux, model-agnostic, local-first so the data stays local. It ships 22 scientific skills and 24 data connectors: literature imports directly by DOI, PubMed or arXiv id, and it searches Europe PMC, OpenAlex and other databases at once, preferring open full text so you stop hunting for papers.
The flow is plain: create a project, describe the research goal in ordinary language, and the agent reads your files, searches the literature, runs Python and R notebooks and generates reports, taking the repetitive work off your hands. It reached #1 on the public BiomniBench-DA leaderboard, so it is not a toy.
What makes it serious is provenance. Every output carries its source, charts are saved as tamper-evident versions, and one click shows the code that produced it, the input files and the environment it ran in — how a result was reached is visible, and reproducing it is just running it again. When the evidence is missing it marks the answer "unavailable" rather than making something up, which matters especially for research. Apache 2.0.
Local-first, cross-platform: a desktop app for macOS, Windows and Linux, one-click install, with data and compute kept on your own machine rather than forced to the cloud.
Model-agnostic: connect different providers' models and switch as needed, without being locked to one vendor.
Literature import and federated search: import by DOI, PubMed or arXiv id and search Europe PMC, OpenAlex and more at once, preferring open full text.
22 scientific skills: literature review, data analysis, figure generation and report writing tuned for research, so you do not assemble each step yourself.
24 data connectors: hooks into common scientific databases and sources, and with Python and R notebooks chains fetching, computing and plotting into one flow.
Reproducible provenance: each output records its source, charts are saved as tamper-evident versions, and the producing code, input files and run environment are all viewable — reproduction is a re-run.
Marks missing evidence: when there is no reliable basis it says "unavailable" instead of inventing one, keeping conclusions trustworthy rather than merely plausible.
AIPOCH Open-Science is an AI research workbench that runs on your own machine — one-click install on macOS, Windows and Linux, model-agnostic, local-first so the data stays local. It ships 22 scientific skills and 24 data connectors: literature imports directly by DOI, PubMed or arXiv id, and it searches Europe PMC, OpenAlex and other databases at once, preferring open full text so you stop hunting for papers.
The flow is plain: create a project, describe the research goal in ordinary language, and the agent reads your files, searches the literature, runs Python and R notebooks and generates reports, taking the repetitive work off your hands. It reached #1 on the public BiomniBench-DA leaderboard, so it is not a toy.
What makes it serious is provenance. Every output carries its source, charts are saved as tamper-evident versions, and one click shows the code that produced it, the input files and the environment it ran in — how a result was reached is visible, and reproducing it is just running it again. When the evidence is missing it marks the answer "unavailable" rather than making something up, which matters especially for research. Apache 2.0.
Features
Local-first, cross-platform: a desktop app for macOS, Windows and Linux, one-click install, with data and compute kept on your own machine rather than forced to the cloud.
Model-agnostic: connect different providers' models and switch as needed, without being locked to one vendor.
Literature import and federated search: import by DOI, PubMed or arXiv id and search Europe PMC, OpenAlex and more at once, preferring open full text.
22 scientific skills: literature review, data analysis, figure generation and report writing tuned for research, so you do not assemble each step yourself.
24 data connectors: hooks into common scientific databases and sources, and with Python and R notebooks chains fetching, computing and plotting into one flow.
Reproducible provenance: each output records its source, charts are saved as tamper-evident versions, and the producing code, input files and run environment are all viewable — reproduction is a re-run.
Marks missing evidence: when there is no reliable basis it says "unavailable" instead of inventing one, keeping conclusions trustworthy rather than merely plausible.
