ABI Documentation¶

ABI lets researchers and AI agents run reproducible bioinformatics workflows through one predictable, confirmation-gated interface. Start by choosing your goal, then move from read-only discovery to a reviewed execution plan.

Start here¶

Your goal	Start with
Understand the components and request flow	Components and Architecture
Run an analysis safely	Using ABI: From a Biological Question to Results
Connect an AI agent	Agent Usage Guide
Add or change a workflow	ABI Development Standards and Plugin Development Guide
Deploy queued or production work	ABI Job Service, Runtime and HPC Development Guide, and Release-ready runtime locks

Install¶

ABI supports Python 3.10–3.13.

pip install abi-agent
abi --version

Install optional capabilities only when needed:

pip install "abi-agent[mcp]"       # MCP server
pip install "abi-agent[report]"    # Scientific figures and richer reports

Five-minute example¶

From a source checkout, the bundled metatranscriptomics fixture can build a plan and dry-run without installing analysis tools or a reference index.

abi list-types
abi query --type metatranscriptomics --what stages

abi plan \
  --type metatranscriptomics \
  --config examples/metatranscriptomics/config_demo.yaml \
  --sample-sheet examples/sample_sheet_transcriptomics.tsv \
  --outdir results/docs-plan

abi dry-run \
  --type metatranscriptomics \
  --config examples/metatranscriptomics/config_demo.yaml \
  --sample-sheet examples/sample_sheet_transcriptomics.tsv \
  --outdir results/docs-dry-run

The fixture contains placeholder reference paths. It demonstrates planning and dry-run output, not real biological execution. Continue with Using ABI: From a Biological Question to Results to prepare a real configuration.

Understand ABI¶

ABI separates transport, reusable workflow mechanics, biological plugins, runtime execution, and result publication. This keeps agent integrations thin and makes every transport use the same safety and result contracts.

Components and Architecture

Use ABI¶

The standard lifecycle is query -> plan -> check -> dry-run -> run -> inspect -> report. Real execution requires explicit confirmation.

Usage and Examples

Develop ABI¶

Start from acceptance criteria, choose the owning architectural boundary, add a regression test, and run checks proportional to the affected release surface.

Development Standards

Operate and release¶

Production use requires pinned versions, verified tools and databases, representative biological acceptance criteria, and a strict runtime lock.

Operations and Release

Component summary¶

Layer	User-visible role
Transports	CLI, JSON, MCP, provider descriptors, dispatch, and HTTP jobs
Core	Planning, permissions, diagnostics, contracts, provenance, tables, reports
Plugins	Analysis-specific tools, DAG branches, parsers, assertions, interpretation
Runtimes	Local, Conda, Docker, Nextflow, Snakemake, HPC, and cloud execution
Results	Execution plan, provenance, standard TSV tables, reports, and figures

Lifecycle and discovery commands expose structured JSON, commonly through --output-json. Use abi <command> --help for exact management-command options and abi list-types --output-json for the authoritative plugin list in the current environment. Representative declarative plugins include metagenomic_plasmid, easymetagenome, and viral_viwrap; availability still depends on the installed runtime and resources.