Agents do the legwork.You keep the judgment.
Epoetic is an agentic research and development environment: one desktop app where the agent you choose works across your code, papers, notebooks, data and notes. You see every change it makes, and you decide what stands.
Invite-only beta · Mac · Windows · Android
A look inside
Agents, knowledge and the work itself.
Screenshots from a development build. Open any image for a closer look.
See how your knowledge connects.
The Brain turns a vault of notes into a linked graph. Open a concept, follow its connections, and give your agents the same source material.
Epoetic · Development build
Choose the agent. Keep your workspace.
Run different agent providers from one interface, using your existing subscriptions.
Epoetic · Development build
Find the context your agents need.
Search the vault by concept, then open the source note and follow its links. Agents can search the same knowledge base.
Epoetic · Development build
Agents
Bring the agent you already pay for.
Run Claude Code, Codex, Cursor, Grok Build, OpenCode and other agents from one app, on the subscriptions you already hold. Epoetic has no model of its own, so there is no second bill. Work that should stay on your computer can run on a local model through Ollama.
Switch agents partway through a thread and its context carries over. Start a task at your desk, then follow and steer it from a browser or an Android phone while it runs.
Research
Papers, notebooks, data and the draft, in one place.
References, analysis and drafts usually live in separate tools, so an assistant sees one piece at a time. In Epoetic they sit beside the conversation, and the agent works across all of them.
References.
Agents read and search the PDFs in your Zotero library and quote them by page. Paste a DOI or arXiv ID to add a paper, and review bibliography entries before they go into your document.
Notebooks.
Python and R notebooks that you and the agent edit and run together, on your computer or a machine you connect to. Installed Julia, MATLAB and Wolfram kernels appear too, and a thread can connect to Google Colab.
Data.
HDF5, NetCDF, Parquet and FITS files open as a structure with a sample of rows, and large CSV files are sampled where they sit. Agents inspect files the same way instead of reading them whole.
Documents.
Write in TeX and build the PDF beside the thread. Send the build errors to the agent in one step, and pull from or push to Overleaf. A figure saved from a notebook keeps the cell and the run it came from.
Claim checks.
A small model on your own machine tells the agent whether a source supports a sentence in your draft, refutes it or doesn’t address it. It is a reason to re-read, not a verdict.
Development
Every turn leaves a record.
Epoetic began as an environment for coding agents, and all of that is still here: a git worktree for any task that needs one, a terminal, a browser, pull requests on GitHub, GitLab and others, and an iOS Simulator or Android Emulator the agent can drive.
With git on, each turn ends with a snapshot of the folder and a line-by-line diff of what it changed. When an agent runs notebook cells, the snapshot also records the kernel’s language version and installed packages. Usage adds up tokens and estimated cost across every agent, by model.
Knowledge
Knowledge that shows where it came from.
The Brain.
A folder of markdown notes becomes pages to read and a map of how they link. Agents search it, and past conversations, before deciding something that has already been decided. After each turn a review proposes the pages worth keeping, and a person accepts them.
Provenance.
Every page shows who wrote it, when, and the conversation it came from. Marking a page reviewed records the exact version you checked, and a later edit clears the mark. The notes stay plain markdown you own.
Learn.
Opt-inSay what you want to understand. Learn finds where your understanding stops, plans the shortest route to the goal and checks each step with a question. What you establish is kept in a vault of your own, so later lessons build on it.
Your data
Your machine. Your accounts.
Threads, settings and logs stay on your computer. Epoetic sends no usage data and no crash reports.
An agent’s prompts, and the files it reads, go to the provider you chose, under your account and on that provider’s terms, as they would if you ran the agent on its own. With a local model they go to Ollama on your computer instead.
We don’t claim that nothing leaves your machine. The app checks for updates, fetches your git remotes and contacts the services you connect. The guide every tester gets names each one.
Who it’s for
Researchers first.
We are building Epoetic first for computational researchers: PhD students, postdocs and lab members who already use Claude or ChatGPT and work across papers, notebooks and code. It fits any team that works from a body of knowledge and has to show how it reached a result, in development, operations, finance or architecture.
Beta
The beta is open by invitation.
It runs on Macs with Apple silicon and on 64-bit Windows 10 and 11, with an Android app for following your agents from your phone. Each person needs a Claude plan that includes Claude Code, a ChatGPT plan that includes Codex, or an API key.
Tell us what you research or build, and which agents you use now. We reply within one working day.
Rules per project for which models may run and which files agents may not read, and an export of a thread’s models, versions and steps for a Methods section or an AI-use declaration.