Hi everyone! I'm Bobo, an AI researcher and developer at AIPOCH. We're bringing AIPOCH Open-Science to Product Hunt, and I'd love some feedback before launch.
Open-Science is an open-source AI research workbench that brings papers, data, AI agents, and Python/R execution into one workspace. It includes 23 built-in Skills, 37 scientific connectors, and 25 preset model-provider options, with support for compatible local models and custom gateways.
One question has been central to our work: when you share a research result, how much of the process can someone else actually follow?
You might send a figure, a notebook, or a report but the questions, intermediate attempts, and decisions often stay scattered across tools.
Hi Product Hunt! I'm Bobo, an AI researcher and developer at AIPOCH.
We're launching AIPOCH Open-Science, an open-source AI research workbench for Windows, macOS, and Linux. It brings papers, data, AI agents, Python/R execution, and research outputs into one persistent workspace.
We built it to make the whole research process easier to work through—from exploring a question and running an analysis to inspecting the results and sharing the work behind them.
Here's what you can work with:
1. Your choice of models
25 preset provider options, including OpenAI, Anthropic, DeepSeek, and NVIDIA Build. Connect GPT, Claude, Grok, Kimi, GLM, and other supported models, use eligible Claude/Codex subscription logins, or connect compatible local models through gateways such as Ollama and LM Studio. Availability depends on your account, agent backend, and API compatibility.
2. Four agent backends
Choose Claude Code, OpenCode, Codex, or CodeBuddy to power your research workflows.
3. Scientific Skills and data access
23 built-in Skills, with another 525 Skills available to install from the marketplace; 37 built-in scientific connectors covering literature, genomics, chemistry, clinical research, and more; plus 10 installable marketplace Specialists.
4. Execution and inspection
Run Python, R, and shell workflows. Inspect generated figures, tables, reports, and notebooks alongside available execution records, artifact versions, provenance, and optional Reviewer checks.
Then comes the part we're especially excited about: .science packages.
A final figure rarely explains the questions, decisions, and intermediate work that produced it. A .science package can carry selected conversation branches, file versions, Notebook records, and verification evidence to another project or computer.
A colleague can inspect the imported research history, learn from your approach, and create a writable copy to explore further. We see this as useful for collaboration, teaching worked examples, and returning to earlier research.
Imported history stays read-only: opening a package doesn't execute code or restore credentials. Recorded checks describe the sender's verification, not proof that the recipient reran the work.
What research task would you try first? And what would you want inside a .science package when picking up a colleague's work?
I'd love your feedback. Thanks for taking a look!