# Open Science by AIPOCH: a local-first AI research workbench you run on your own desktop > Open Science is an Apache-2.0 Electron workbench that drives Claude Code, OpenCode, Codex or CodeBuddy agents over your own files, with Python and R execution and artifact provenance. The release page is the only supported install path, and the README says nothing about rollback. **aipoch/open-science** — Open Science is an open-source, local-first, model-agnostic AI research workbench for scientific discovery. - Repository: https://github.com/aipoch/open-science - Website: https://www.aipoch.com/open-science - Stars: 4,931 · Forks: 416 - Language: TypeScript - License: Apache-2.0 - Published: 2026-08-08 · Updated: 2026-08-18 · Language: en - Canonical page: https://hysenlabs.com/projects/aipoch-open-science ## The gap Open Science fills between a chat window and a notebook Most AI coding tools assume the artifact is code. Scientific work produces something else: a figure, a table, a fitted model, and an argument about whether the result holds. Open Science targets that second category. The README describes it as an open-source, local-first, model-agnostic AI research workbench for scientists and researchers, with support for machine learning, statistics, life sciences, chemistry, materials science, physics and environmental science. The intended user is someone who already has data on a laptop or workstation and wants an agent to read files, search the web, run code, query scientific data sources, and produce reports, tables and figures with traceable provenance. The provenance emphasis is the differentiator. A chat assistant will happily produce a plot and forget which CSV, which filter and which model produced it. Open Science organizes work into projects and sessions and attaches a Provenance view to each artifact, listing its versions and the available evidence behind the selected result. That is a research-integrity feature, not a convenience feature, and it is the reason the project exists in this form. ## How the agent loop, runtimes and notebook kernels fit together Open Science is an Electron application. The repository carries electron.vite.config.ts, electron-builder.yml, a src directory, a packages directory and a Prisma schema, and package.json names ./out/main/index.js as the entry point. That layout says the app is a packaged desktop binary rather than a web service, which matches the local-first claim: your files stay on disk and the storage root is chosen during setup. The first launch walks through five steps. Environment checks compatibility, app storage, secure credential storage and network access. Agent runtime selects and prepares Claude Code, OpenCode, Codex or CodeBuddy. Model provider connects and tests a model, either a built-in provider, a custom gateway, or an existing Claude or Codex subscription login. Notebook runtime optionally prepares app-managed Python and R environments, or uses detected and manually registered interpreters. Data location chooses where large artifacts, notebooks, uploads and environments live. Two details matter here. First, the app can install its own runtimes without Node.js, npm or an administrator password, which removes the usual dependency chain for a scientist who does not maintain a toolchain. Second, notebook execution is optional, so you can use the agent loop on files without ever starting a kernel. When a kernel is running, a Variables view shows the live Python or R namespace with names, types, shapes and previews, read-only, refreshed after each execution. That read-only choice is deliberate: inspection without letting the UI mutate your state. ## Installing Open Science on macOS, Windows or Linux There is no package-manager install documented in the README. The Quick Start says to open the latest release page, expand Assets, and pick the installer for your machine: the macOS DMG for Apple Silicon or for Intel, the Windows x64 installer, or the Linux x64 AppImage or Debian package. The release page is also where verification information is published, and the README points at SECURITY.md for validating a package before installation. If macOS or Windows shows an unidentified-developer or unknown-publisher warning, the README's instruction is to confirm the package came from the official Releases page before continuing. Note that package.json also declares a bin entry named open-science pointing at cli/index.mjs, and the repository contains a cli directory and a REMOTE_CONTROL.md file. The README does not document a CLI workflow, so treat that as an internal surface rather than a supported install route. After the binary is in place, the first real use is a project and a session. The README gives this sequence: click New project, give it a stable research name and optional description, open a session, then describe the goal, input data, constraints, desired outputs and how the result should be checked. Within a session, @ references an existing project file and / explicitly selects an enabled skill. Approval mode controls which sensitive actions the agent may take without asking, and generated artifacts open in a preview panel. Editing an earlier user message and resending it creates a new branch, and the message revision controls let you return to either path. That branching behaviour is the part worth trying first, because it is what separates an exploratory session from a linear chat log. ## What the README does not tell you about updates and rollback The README documents installation and first-run setup in detail and says nothing about rollback. It also does not document a downgrade path, a migration guide between releases, or what happens to an existing project database when the app version changes. Given that the repository contains a Prisma schema and the .env.example notes that at runtime the app supplies the SQLite path programmatically under the storage root, an upgrade that touches the schema is a real risk to in-flight work, and the documentation is silent on how that is handled. The changelog for v0.22.0 mentions safer update downloads and runtime installs, which suggests the update path has been hardened, but hardening downloads is not the same as guaranteeing reversibility. The practical consequence: back up the storage root you chose in step five before accepting an update, and do not assume you can step back to an earlier release with your projects intact. This is a documentation gap, not a demonstrated defect, and it is the first thing you should test on a spare machine rather than on the workstation holding your only copy of a dataset. ## Where Open Science is the wrong tool Three cases. If your research runs on a shared cluster or a lab server and collaborators connect over SSH, Open Science is a desktop application with a local storage root, and the README describes no headless or multi-user deployment. Running it on one person's laptop does not give a group a shared workspace. If you need a stable programmatic API to script runs from CI, nothing in the README describes one; the bin entry and the cli directory exist in the repository, but the README does not document their interface, so you would be building on an undocumented surface. If you want a hosted service with zero local setup, this is the opposite design: you choose where artifacts and environments are stored, and you are responsible for that disk. There is also a subtler limit. The provenance view records versions and available evidence for artifacts the app produced. It cannot vouch for data you imported from elsewhere, and the README does not describe any validation of external inputs. Treat provenance as a record of what the agent did, not as a guarantee that the input was correct. ## Open Science compared with the Open Science Framework The name collides with OSF, the Open Science Framework, and the search data shows people conflate them. They solve different problems. OSF is a hosted platform for registering studies, archiving materials and sharing preprints, and its core value is a public, citable record. Open Science by AIPOCH is a local desktop workbench whose core value is running agents over your own data and keeping provenance attached to the artifacts they generate. One is a place to publish and register; the other is a place to compute. If your need is a DOI-backed preregistration or a repository for supplementary files, OSF is the right tool and Open Science does not attempt that job. If your need is an agent that can read a directory of instrument output, run Python or R against it, and leave a trail explaining how a figure was made, OSF does not do that. The overlap is only the phrase. The README's What This Is Not section exists precisely because the project expects this confusion, and it is worth reading before you decide which one you actually need. ## Licence, maintenance and what an upgrade actually costs Open Science is Apache-2.0, the licence declared in package.json and in the repository. Apache-2.0 permits commercial and academic use, modification and redistribution, and it includes an explicit patent grant, which matters if your institution cares about that. It also requires that you preserve notices and state significant changes if you redistribute. That is a summary of the licence text, not legal advice; if you plan to ship a modified build, have your institution's office read the actual LICENSE file. On maintenance: the repository is not archived, and the last push was on 2026-08-29, which is recent enough that active development is a fair description. The release cadence visible in the release list is fast, with v0.21.0 on 2026-08-28 and v0.22.0 the next day, plus a nightly build tracking main. A nightly channel is useful and also a signal that main moves quickly. The upgrade cost for a user is therefore not the download; it is re-validating your agent runtime, your model provider connection and your notebook interpreters after each version bump, because those are the three moving parts the first-run flow checks and the three most likely to change. Pin to a tagged release rather than nightly if your analysis has to be reproducible months later. ## Conclusion Adopt Open Science if you want agent-driven analysis of local scientific data where every figure keeps a provenance trail and the model provider is your choice. Skip it if you need a headless server deployment, a stable documented API, or a project with a long release history, since v0.22.0 is the newest tagged release and the README does not document rollback. Before trusting it with real data, verify the published checksums against SECURITY.md, confirm that your chosen agent runtime and model provider both pass the first-run checks, and test the update path on a spare machine. ## FAQ ### What is the concept of open science? The README frames Open Science as an open-source, local-first, model-agnostic AI research workbench for scientists and researchers, built for reproducible, inspectable research. It runs scientific AI agents that read files, search the web, execute Python and R, query scientific data sources, and produce reports, tables and figures with traceable provenance. ### What is open science research? The README states that Open Science supports computational and data-intensive research across disciplines including machine learning, statistics, life sciences, chemistry, materials science, physics and environmental science. It covers the process from literature review and hypothesis development through code execution, data analysis, simulation and visualization to traceable research outputs. ### How do I use the Open Science Framework? Open Science by AIPOCH is not the Open Science Framework. The README's Quick Start says to download the installer for your platform from the latest release page, complete the five-step first-run setup, then click New project, open a session and describe your research goal, input data, constraints and desired outputs. ### What is the Open Science Framework OSF? The README distinguishes Open Science by AIPOCH from OSF by describing itself as a local-first desktop workbench rather than a hosted platform, and it includes a What This Is Not section for exactly this confusion. OSF is a separate hosted service for registering studies and sharing materials; Open Science runs agents over your own files on your own machine. ### What is open science 101? The README does not describe any training course, curriculum or introductory programme under that name, and it does not use the phrase. The material covers the desktop application only, so this question cannot be answered from the project's own documentation. ## Sources - [Official documentation](https://www.aipoch.com/open-science) - [Official README](https://github.com/aipoch/open-science#readme) - [Project repository](https://github.com/aipoch/open-science) - [Release notes](https://github.com/aipoch/open-science/releases) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/aipoch-open-science