Lincei
Lincei is an agentic clinical research environment. You direct coding agents (Claude Code,
Codex, Cursor, OpenCode, and others) at clinical datasets, stay in the loop while they work, and
review what they produce as visual research cards instead of walls of chat. See
www.linceilabs.com.
Lincei is built on T3 Code and tracks it closely: threads,
terminals, diffs, the built-in browser, remote access, and provider support all come from upstream.
Lincei adds the clinical layer:
- A managed clinical Python environment. M4 for clinical
datasets such as MIMIC-IV, Vitrine for research cards,
and the usual analysis stack (pandas, DuckDB, scikit-learn, lifelines, statsmodels). Every agent
is told where it is, so nothing has to be set up per project.
- Clinical skills. Validated definitions the agents read before the matching task: SOFA,
Sepsis-3, KDIGO AKI staging, comorbidity scores, cohort pitfalls, and more. They are in the
composer's skill picker next to your own skills.
- A dataset ready on first run. The open-access MIMIC-IV demo is downloaded during setup.
- Your own datasets. Settings → Clinical Data lists the datasets M4 supports. It
downloads one with your PhysioNet account, or prepares it from a folder of files you already
downloaded, with progress and no terminal. Your PhysioNet password is used for that download
and never stored, and agents never see it.
- Private by default. No telemetry. No hosted relay. Data stays on your machine unless you
send it to a model provider yourself.
[!WARNING]
Lincei is an early alpha. Agents send the text they read, including query results, to the
model provider you chose. Use the open MIMIC-IV demo, or make sure your data use agreement
covers that provider before pointing Lincei at credentialed data.
Install
- Install and sign in to at least one provider, for example
Claude Code (
claude auth login) or
Codex CLI (codex login).
- Download Lincei from www.linceilabs.com or from
Releases, choosing the build for your
computer:
On Windows, the installer is not signed yet, so Windows SmartScreen says it protected your PC.
Choose More info, then Run anyway.
On a Mac, Lincei is not signed with an Apple Developer ID yet, so macOS blocks the first launch and
says it cannot verify the app. Open Lincei once and choose Done, then go to System Settings →
Privacy & Security, scroll down to the message about Lincei, and choose Open Anyway. The
button shows for about an hour after macOS blocked the app; if it is missing, open Lincei once
more. On macOS 14 or earlier, right-click the app and choose Open instead. You do this once per
download, and the installer window shows the same steps.
Lincei does not update itself yet. To update, quit Lincei, download the new build from
Releases, and replace the app in Applications, run
the new installer on Windows, or replace the AppImage on Linux; your conversations and datasets
stay. On a Mac, allow it in Privacy & Security again, as above. On its first launch macOS asks
whether Lincei may use "lincei Safe Storage" in your keychain, where Lincei keeps its sign-ins
encrypted: enter your Mac password and choose Always Allow.
On first launch Lincei sets up the clinical environment in the background: Python, M4, Vitrine,
the clinical skills, and the demo dataset. It takes a few minutes and needs a network connection,
but no terminal. The last step of the first-run wizard and Settings → Clinical Data show how
far it is and let you retry a step that failed. You can also ask the agent "is the clinical
runtime ready?"; it can check and repair the setup itself.
The clinical skills are linked into ~/.claude/skills and ~/.agents/skills (on Windows, the
.claude\skills and .agents\skills folders in your user folder), where Claude Code and Codex
look for skills, so they are available to those tools outside Lincei too. A skill of your own with
the same name is never replaced.
Try it
Open a folder as a project, start a thread, and ask:
Using the MIMIC-IV demo, build a Sepsis-3 cohort, describe it, and show me the result as cards.
The agent reads the relevant skills, queries the data through M4, and publishes tables and figures
to Vitrine. When the first card arrives, Lincei opens the study beside the conversation for you to
review. The flask button at the top right of a thread opens that study at any time; while Lincei
is still setting up, it shows what it is working on.
Run Lincei in your browser
Lincei also runs as a server on your computer that you open in a web browser, for example on
Linux. Agents, the clinical environment, and your data stay on that computer, as with the app. It
needs an Apple Silicon Mac, or Linux on x64 or arm64, and a provider as in step 1 of Install.
curl -fsSL https://raw.githubusercontent.com/hannesill/lincei/main/scripts/install.sh | sh
lincei
lincei starts Lincei and opens it in your browser. The first start sets up the clinical
environment, as the app does. The server only accepts connections from the same computer.
The research board opens in a new browser tab instead of beside the conversation. When the
first card of a conversation arrives, Lincei asks before it opens that tab.
lincei update moves to the newest release, and lincei uninstall removes Lincei but keeps your
conversations and datasets. The app and lincei share those, so quit the app before you start
lincei.
Develop
See docs/lincei/README.md for how the fork is organized, how to pull in
upstream changes, and how to cut a release. Everything else in docs/ is
upstream's documentation and applies unchanged, with "T3 Code" read as "Lincei".
Credits
Lincei is built on T3 Code by T3 Tools Inc. Almost all of
the app is their work: Lincei adds a clinical layer and follows their releases. Thank you to T3
Code's maintainers and contributors for building it in the open. If you want an agent app for
software work rather than clinical research, use T3 Code itself.
Lincei is MIT licensed, as is T3 Code. LICENSE is T3 Code's, and the app shows it
under Settings → Open source licenses.
