Althea: a research assistant grounded in the commonsLacuna, the open commons: 733,000+ papers organized into research directions and proposals. Open without an account., connected to your peers
It reads the literature, runs the code, keeps the figures, and stays up to date with the field of machine learning research.
Althea is Tiptree's research assistant. It works over Lacuna, an open commons of machine learning research, and over the parts of your research the commons cannot see: the open web, your code, your data, your machine learning research stack. And when a question needs a person rather than a paper, it can carry it to other researchers' agents, with your consent each time. One rule runs through all of it. Every claim traces to a source.
What Althea can do
Each capability has its own page describing what it is and how it works.
| Ask it to | What happens | The page |
|---|---|---|
| Review the literature | A Careful run searches, reads, and synthesizes a structured, source-linked review | Deep research |
| Check a claim in code | It clones the repository and runs it — sandboxed, or on your cluster through SLURM — answering with files, line numbers, and outputs | Code agents |
| Summarize with the figures | Replies keep the paper's own figures, each linked to its exact location in the source PDF | Summaries with figures |
| Make it interactive | The code agent builds a widget in the chat; filter it, hover it, share it by link | Widgets |
| Remember your field | A readable, editable memory plus a researcher profile built from your public work | Memory and Profile |
| Watch for movement | Background monitors that interrupt you only past a relevance bar | News briefings |
| Meet you anywhere | Web, WhatsApp, and email; one agent, one memory, every channel | Channels |
| Run on your stack | W&B and GitHub connections; launch runs, track them, open pull requests | Integrations |
| Reach another researcher | It carries your question to a peer's agent through a consent-first network | The network |
Interacting with Althea
Althea shares its thought process as it works, so you see exactly what it is up to: what it is searching, what it is reading, what it made of what it found. You also pick how hard it tries. Quick mode asks Althea to be succinct in its effort; Careful mode lets it proceed as usual, often going through multiple rounds with the sources and following up on the new ones each round turns up. For the deepest dives, Althea can hand the work to asynchronous jobs that keep running while you do something else; when they finish, it reaches back out. Your first five minutes shows how to use both modes.
Althea also does not exist only while you are talking to it. Between conversations it watches the literature for movement in your field, consolidates its memory of your work, and finishes the jobs it started with you. The chat is where you meet it, not where it lives.
Reaching other researchers
Some questions are not literature questions. The answer sits with a researcher in another lab. Althea can reach that researcher's agent through a consent-first network. Nothing you said travels without your consent, and Althea asks each time.
When to reach for Althea
Althea is built for machine learning research: questions about the literature, claims worth checking against code, a field moving fast enough to need watching. It handles general questions too; a frontier model sits underneath. What sets it apart is the agent harness built around that model — the commons, the code sandbox, the provenance trail, the network — all aimed at research, which is where it shines. The comparison page states the boundaries plainly.