A desktop app that runs bioinformatics tools, AI, and pipelines locally. Your data never leaves your machine — Rust-powered native speed, even on multi-gigabyte files.
See the codeSee what people are saying
README
Liatir
Liatir is a local-first desktop environment for bioinformatics, built with Rust,
Tauri 2 and SvelteKit. It brings native tools, locally managed AI Models,
.lia Plugins, visual pipelines and scientific viewers into one application,
with API Connectors, external workflow integrations and a controlled local MCP
interface for connecting other applications.
Local analyses keep working offline once their dependencies are installed.
Workspaces retain inputs, execution logs, results and provenance. Shared types
and input/output contracts live in packages/liatir-core.
Product documentation · Downloads and platform status · Plugin development
Scientific Showcases
Liatir Scientific Showcases document scientific questions, methods, measured results and limitations, with technical packages in showcases/.
Single-cell foundation models vs established baselines
A completed comparison of pretrained Geneformer and scGPT representations with PCA, Harmony and scVI on PBMC and Pancreas single-cell data. Ten configurations completed; two UCE configurations remain blocked with documented causes.
Observed result: pretrained models did not show a uniform advantage. scGPT was competitive on PBMC (logistic macro-F1 0.95685), while PCA and scVI remained strong on Pancreas (0.97788 and 0.97184; scGPT 0.89987). Macro-F1 measures cell-type prediction with equal weight per type. This is a two-dataset, one-seed study; scVI was trained on the evaluation datasets.
Original validated figure. Pancreas scGPT ran on a PC GPU; the other completed representations used Mac CPU. These times do not establish a same-host speed ranking.
Read the study · Source, results and validation evidence · Complete reproducibility artifacts and citation: Zenodo DOI
The source, small results, figures and evidence are tracked here. The complete approximately 1.8 GB archive is kept outside Git and linked through the Zenodo record.
Repository layout
| Path | Purpose |
|---|---|
| packages/liatir-core | Shared types and scientific input/output contracts. |
| src-tauri | Native app, bridge commands and process management. |
| src-ts | TypeScript bridge and Plugin runtime. |
| frontend | SvelteKit interface, pipelines, tools and viewers. |
| runtime-boxes | Installable runtime definitions, catalog and release evidence. |
| docs | Public product documentation and Scientific Showcases. |
| showcases | Canonical scientific study packages. |
| .context | Shared architecture, decisions and current project status. |
Development
Read AGENTS.md and .context/index.md before substantial changes. Install Node.js, Rust and the Tauri system prerequisites, then use the repository scripts:
npm ci
npm ci --prefix frontend
npm run localdevconf
npm run dev
Use npm run dev:frontend for the interface alone. npm run test:fast runs
unit and contract checks; npm run test:verify is the normal completion gate.
Native app changes also require the relevant npm run test:ui suites. Heavy AI
tests are explicitly opt-in.
npm run native-tools:build prepares the bundled Native Tools for this host;
ordinary development does not build them automatically. See
AGENTS.md for platform constraints and the complete build/test commands.
Build the public documentation with npm run docs:build.
Liatir's application is licensed under GNU GPL v3.
Similar repos
A desktop app that runs bioinformatics tools, AI, and pipelines locally. Your data never leaves your machine — Rust-powered native speed, even on multi-gigabyte files.
See the codeSee what people are saying
README
Liatir
Liatir is a local-first desktop environment for bioinformatics, built with Rust,
Tauri 2 and SvelteKit. It brings native tools, locally managed AI Models,
.lia Plugins, visual pipelines and scientific viewers into one application,
with API Connectors, external workflow integrations and a controlled local MCP
interface for connecting other applications.
Local analyses keep working offline once their dependencies are installed.
Workspaces retain inputs, execution logs, results and provenance. Shared types
and input/output contracts live in packages/liatir-core.
Product documentation · Downloads and platform status · Plugin development
Scientific Showcases
Liatir Scientific Showcases document scientific questions, methods, measured results and limitations, with technical packages in showcases/.
Single-cell foundation models vs established baselines
A completed comparison of pretrained Geneformer and scGPT representations with PCA, Harmony and scVI on PBMC and Pancreas single-cell data. Ten configurations completed; two UCE configurations remain blocked with documented causes.
Observed result: pretrained models did not show a uniform advantage. scGPT was competitive on PBMC (logistic macro-F1 0.95685), while PCA and scVI remained strong on Pancreas (0.97788 and 0.97184; scGPT 0.89987). Macro-F1 measures cell-type prediction with equal weight per type. This is a two-dataset, one-seed study; scVI was trained on the evaluation datasets.
Original validated figure. Pancreas scGPT ran on a PC GPU; the other completed representations used Mac CPU. These times do not establish a same-host speed ranking.
Read the study · Source, results and validation evidence · Complete reproducibility artifacts and citation: Zenodo DOI
The source, small results, figures and evidence are tracked here. The complete approximately 1.8 GB archive is kept outside Git and linked through the Zenodo record.
Repository layout
| Path | Purpose |
|---|---|
| packages/liatir-core | Shared types and scientific input/output contracts. |
| src-tauri | Native app, bridge commands and process management. |
| src-ts | TypeScript bridge and Plugin runtime. |
| frontend | SvelteKit interface, pipelines, tools and viewers. |
| runtime-boxes | Installable runtime definitions, catalog and release evidence. |
| docs | Public product documentation and Scientific Showcases. |
| showcases | Canonical scientific study packages. |
| .context | Shared architecture, decisions and current project status. |
Development
Read AGENTS.md and .context/index.md before substantial changes. Install Node.js, Rust and the Tauri system prerequisites, then use the repository scripts:
npm ci
npm ci --prefix frontend
npm run localdevconf
npm run dev
Use npm run dev:frontend for the interface alone. npm run test:fast runs
unit and contract checks; npm run test:verify is the normal completion gate.
Native app changes also require the relevant npm run test:ui suites. Heavy AI
tests are explicitly opt-in.
npm run native-tools:build prepares the bundled Native Tools for this host;
ordinary development does not build them automatically. See
AGENTS.md for platform constraints and the complete build/test commands.
Build the public documentation with npm run docs:build.
Liatir's application is licensed under GNU GPL v3.