Liatir

Appearance

Pipelines ​

Pipelines let you connect analysis steps visually so that the output from one step becomes the input for the next one.

Use pipelines when you want to repeat a workflow, avoid manual file handoffs, or combine built-in tools, .lia plugins, AI Tools, and saved External Workflows in one place.

What a pipeline contains ​

A pipeline is made of nodes and connections:

  • Input nodes provide starting files or values.
  • Tool nodes run built-in tools, .lia plugins, AI Tools, or saved External Workflows.
  • Logic nodes help control simple branching or conditional behavior.
  • Connections pass compatible outputs into later inputs.

Each node exposes only the inputs that make sense for that tool. File pickers are filtered by compatible format whenever possible.

Running a pipeline ​

  1. Open Pipeline from the sidebar.
  2. Add the tools or plugins you want to use.
  3. Connect outputs to compatible inputs.
  4. Fill in required fields.
  5. Click Run pipeline.

While a pipeline is running, it is locked to prevent accidental edits. You can still work in other pipelines and elsewhere in the app.

Starting from a preset ​

The Pipelines page includes small, editable workflows with their artifact handoffs already connected. Single-cell embedding and preview asks only for an AnnData file and an installed AI Model, then validates the input, creates an immutable embedded AnnData result, opens its bounded preview, and leaves the artifact in Data for later nodes.

Jobs and Results ​

Long-running steps appear in Jobs while they run. When the pipeline finishes, the completed run appears in Results with:

  • the pipeline name;
  • each step that ran;
  • logs and errors;
  • output files;
  • metrics and summaries;
  • provenance for tools and AI Models.

If a step fails, Liatir keeps the logs and shows which part of the pipeline failed so you can fix the input or settings and run again.

Example workflows ​

FASTQ quality control ​

  1. Start from FASTQ files in Data.
  2. Run FastQC.
  3. Run fastp to trim reads.
  4. Add trimmed files back to Data or pass them to the next step.

Variant filtering ​

  1. Start from a VCF or BCF file.
  2. Run BCFtools stats to inspect the callset.
  3. Run BCFtools filter with a quality expression.
  4. Review the filtered VCF in Results.

AI-assisted workflows ​

  1. Install a compatible AI Model.
  2. Add an AI Tool to the pipeline.
  3. Select the model inside the tool.
  4. Connect the generated output to viewers, reports, or later tools.

Example AI pipelines:

Read Local AI for bioinformatics before interpreting AI outputs.

External workflow integration ​

A saved External Workflow appears as a reusable pipeline node. Its declared files and parameters become node inputs, and only its exact declared outputs are available to downstream nodes. The external engine keeps ownership of its internal scheduling while Liatir keeps the parent Pipeline Run, Job, Result and cross-tool provenance.

Saving and reusing workflows ​

Pipelines are meant to be reusable. A saved workflow keeps its structure and settings so you can return to it later, adjust inputs, and run it again.

An MCP-compatible client can also request an exact saved revision through Local MCP. This is off by default: you allow each saved revision in Settings and approve every individual run in Liatir.

For custom steps, use .lia plugins. A plugin can wrap a script or command-line tool through the Liatir API bridge and still behave like a normal node in the pipeline.