​
Using Biomni Tools in BioCortex

BioCortex uses Biomni’s 200+ domain tools (genomics, single-cell, pharmacology, etc.) so the model (Planner + Executor) can call Biomni functions inside the analysis sandbox. Two ways: embedded (install and use) or custom path.

​
Default: embedded tools (Plan A)

If Biomni is present at vendor/Biomni/biomni (e.g. added as a git submodule), BioCortex loads it automatically. No configuration needed.
  • After cloning BioCortex, run: git submodule update --init --recursive to fetch the embedded Biomni tools.
  • See vendor/README.md in the repo for how to add the Biomni submodule.

​
Optional: custom Biomni path

To use a different Biomni clone (e.g. your own fork or another path on the server):
  1. Open Settings → Biomni path.
  2. Enter the absolute or relative path to the Biomni biomni directory (e.g. /path/to/Biomni/biomni).
  3. Save. This overrides the embedded path and is sent with every new analysis.
If you leave Biomni path empty and vendor/Biomni/biomni exists, the embedded path is used. If neither exists, analysis runs with BioCortex built-in domains only (ecology, neuroscience, plant, marine, browser, etc.).

​
How it works

  1. Tool descriptions — Biomni’s tool_description/*.py are imported into BioCortex’s tool registry and used by the hybrid retriever (vector + KG + LLM) so the Planner picks the right tools.
  2. Tool bridge — Biomni’s actual tool/*.py functions are exposed to the Executor via a preamble and docs. The Executor generates code that imports and calls them, e.g. from biomni.tool.cell_biology import annotate_celltype_scRNA.
  3. Know-how — Biomni’s know_how/*.md best-practice docs are loaded and injected into the Planner context.

​
Run analysis

In Chat, describe your task in natural language (e.g. “Analyze single-cell RNA-seq data from pbmc.h5ad with clustering and cell type annotation”). The Planner will select Biomni tools (and other BioCortex tools) via the hybrid retriever; the Executor will generate and run code that calls Biomni functions. No change to prompts is required.

​
Verify Biomni is loaded

  • Status API: GET /api/v1/status — when an agent with Biomni is loaded, the response can include biomni_bridge (e.g. number of public functions, domains).
  • Resources page: Biomni tools appear in the tools list with domain labels (e.g. genomics, cell_biology).

​
Summary

CaseAction
EmbeddedEnsure vendor/Biomni/biomni exists (git submodule update --init --recursive). No Settings needed.
Custom pathSet Biomni path in Settings (or in analysis config when using the API).
No BiomniLeave path empty and do not init the submodule; analysis uses built-in domains only.