biotope 0.7.1

CLI integration for BioCypher ecosystem packages

biotope

Package
Meta

CLI for the BioCypher ecosystem: Croissant-described data → BioCypher knowledge graph, with git-like metadata version control.

Status: pre-alpha, developer-facing. APIs and CLI will change. Not yet suitable for end users.

Start here

The fastest way in is the tutorial — a 15-minute end-to-end walk-through that builds a real knowledge graph from public airport/flight data. It is the canonical, most up-to-date onboarding path; the snippet below is just a flavour preview.

Install

uv add biotope                     # in a uv-managed venv
pipx install biotope               # global install
uvx biotope init my-kg             # no install: ephemeral venv for the scaffolder
uv pip install -e ".[dev]"         # editable, with test deps (for biotope itself)

From init to a knowledge graph

# 1. Scaffold a project (uvx works fine here — no local install needed).
uvx biotope init my-kg --purpose "What approved drugs target genes in T2D?" --no-prompt
cd my-kg && uv sync

# 2. Bring in data — biotope get downloads + tracks; biotope add stages a
#    local file/folder and runs croissant-baker over it.
uv run biotope get https://example.org/opentargets.parquet --output-dir data/ot --no-add
uv run biotope add data/ot --license CC-BY-4.0 --creator "Open Targets"

# 3. Inspect the pipeline state at any time.
uv run biotope queue        # raw / processed / mapped, with provenance footer

# 4. Declare intent — what entities and relations the graph must contain.
#    Non-interactive (agent-friendly); without flags, `biotope map` opens
#    a wizard that captures intent and resolves slots in one flow.
uv run biotope map --entity gene --entity disease --entity drug \
                   --relation gene_associated_with_disease

# 5. Resolve the slots. Two equivalent paths:
#    a) Wizard (humans):  uv run biotope map
#    b) Edit mappings/*.mapping.yaml directly, then validate (agents):
#       uv run biotope map scaffold .biotope/datasets/data/ot.jsonld
#       uv run biotope map inspect  <croissant> --json   # field catalogue
#       uv run biotope map preview  --json               # projected schema + tuples

# 6. Build a runnable BioCypher project. Strict: rejects unresolved slots.
uv run biotope build
uv run python build/create_knowledge_graph.py
uv run biotope view

biotope init is a pure scaffolder. All non-autogeneratable metadata is supplied as CLI flags — by a user or an agent reading AGENTS.md. Semantic decisions (which record set, which fields, which transforms) are made by the human or copilot; biotope only enumerates options, validates, and previews.

Architecture

Two layers, both in this repo:

Layer	Module	Role
Project & VCS	biotope.commands.*	init, add, commit, status, log, push, pull, mv, check-data — git-like metadata workflow
KG construction	biotope.croissant.*	spec, codegen, acquisition, mapping, alignment, scaffold, registry — Croissant → BioCypher project

biotope.croissant.api exposes scaffold_mapping, propose_alignment, materialize, discover_sources as pure functions; the CLI verbs are thin wrappers. The mapping authoring surface lives under biotope.commands.map (Click group) and biotope.commands.map_wizard (Rich-based guided flow).

The agent surface is AGENTS.md (template lives at biotope/templates/AGENTS.md, copied into every project by init). No MCP server — agents drive the same CLI a human uses.

Commands

init                                project scaffolding
get add mv rm                       acquisition + tracking (baker writes croissants)
queue mark                          pipeline-state dashboard + manual transitions
map (inspect|scaffold|preview)      semantic mapping (intent + wizard + agent path)
propose-alignment                   cross-mapping same_node equivalences
build view                          build + inspect a graph
status commit log push pull         git-like metadata VCS
check-data                          checksum verification
annotate config                     field-level annotation + project config

discover benchmark                  scaffolded but not yet wired into the standard flow
read chat search                    promises / auxiliary

biotope describe and the heuristic biotope propose-mapping were removed/deprecated. Intent capture is now biotope map --entity ... --relation ...; scaffolding is biotope map scaffold. propose-mapping remains as a deprecated alias for the scaffold subcommand.

See docs/architecture.md for the data-flow diagram and docs/api-docs/ for per-command reference.

Development

uv sync --extra dev
uv run pytest
uv run ruff check biotope tests

Build backend is hatchling; lockfile is uv.lock. CI runs uv sync + uv run pytest on Python 3.10 and 3.12.

Copyright

Copyright © 2025–2026 BioCypher Team. Apache 2.0.