A Rust-native, multi-agent AI framework for omics analysis and bioinformatics research.
Prerequisites:
- Rust 1.75+ (install rustup)
- Pixi v0.70+ (install pixi)
Build:
# Clone repository
git clone https://github.com/Minions-Land/AutOmicScience.git
cd AutOmicScience
# Install dependencies with Pixi
pixi install -e default
# Build (debug mode)
cargo build
AutOmicScience (AOSE) runs autonomous AI agents that plan and execute real bioinformatics analyses — single-cell RNA-seq, ATAC-seq, multiome, and spatial transcriptomics — through a reproducible Python (scverse) execution layer, with every cited result traceable back to a real tool call rather than model recall. The runtime, agent loop, tool registry, workflow engine, and terminal UI are all Rust; the scientific compute is delegated to a pinned scverse environment over a typed bridge.
- Real analysis, not hallucinated results. Agents call into a versioned
scverse runtime (scanpy, snapatac2, squidpy, decoupler, …). The
citesubsystem can prove every source in a report came from an actual tool call. - Rust-owned runtime. Agent loop, model chain, memory, compaction, self-verification, tool dispatch, plugins, skills, workflows, sessions, and permissions are all native Rust — fast, typed, and reproducible.
- Multi-agent orchestration. Team patterns, sub-agent spawning, and a JSON/Rhai workflow engine coordinate work across independent agents.
- Provider-agnostic. OpenAI-compatible, Anthropic, and Gemini providers with streaming, plus a mock fallback for offline testing.
- Terminal-first. A Ratatui TUI and a structured NDJSON backend for scripting and integration.
# 1. Build the workspace (release)
cargo build --release
# 2. Configure API keys (writes ~/.aose/.env)
cargo run --bin aose-cli -- setup
# 3. Launch the terminal UI
cargo run --bin aose-cli -- tui
# …or run a one-shot prompt
cargo run --bin aose-cli -- run "Summarize the QC steps for a 10x scRNA-seq run"
# …or an interactive REPL
cargo run --bin aose-cli -- cliA convenience installer is also provided:
./setup.shIt creates ~/.aose/ (config, logs, plugins, skills, commands, workflows),
scans common locations for existing API keys, and writes ~/.aose/.env.
Scientific compute runs in pinned Pixi environments declared in pixi.toml.
Set it up and inspect runtime resolution with:
cargo run --bin aose-cli -- pixi --help # manage Pixi workspace environments
cargo run --bin aose-cli -- env --help # see how AOSE resolves Python/R/Node/Julia/Pixi
pixi install -e task1 # install the scRNA-seq environmentPixi environment aliases: sc-rna -> task1, spatial -> task2,
sc-multiome -> task3, sc-atac -> task4, and sc-atac-r -> r-env.
Modality skills live under skills/omics/ and dispatch to the
scverse runtime:
| Modality | Skill dir | Examples of supported methods |
|---|---|---|
| scRNA-seq | skills/omics/scrna |
QC, normalization, clustering, marker tables, annotation |
| scATAC-seq | skills/omics/scatac |
ATAC QC, peak/feature calling, gene activity, motif/TF |
| Multiome | skills/omics/multiome |
Joint embedding, multiome dynamics, peak-to-gene linkage |
| Spatial | skills/omics/spatial |
Spatial domains, cross-modality integration |
Method references are documented under
docs/guide/methods/ (ATAC QC, GRN, motif/TF,
joint embedding, gene activity, regulation, and more).
The built-in annotation pipeline is also exposed directly:
cargo run --bin aose-cli -- annotate --helpaose-cli is the single entry point. Key subcommands:
| Command | Purpose |
|---|---|
run |
Run a single prompt and print the final text |
tui |
Launch the Ratatui terminal UI (--resume, --continue) |
cli |
Interactive REPL with a default AOSE agent |
setup |
Interactive API-key setup -> ~/.aose/.env |
annotate |
Built-in single-cell annotation pipeline (Pixi-backed Python modules) |
report |
Generate traceable Markdown/LaTeX/PDF reports |
cite |
Verify data provenance -- prove cited sources came from real tool calls |
workflows |
Discover, load, or run JSON workflows |
teams |
Run multi-agent team orchestration patterns |
knowledge |
Local knowledge-base and practice-library utilities |
plugins |
Discover and inspect plugin manifests |
commands |
Discover, load, or run prompt commands |
templates |
Manage agent/team/skill/prompt templates |
evolution / evolve |
Evolutionary-search optimization utilities |
pixi / env |
Manage Pixi envs / inspect runtime resolution |
sessions |
List, inspect, and manage saved sessions |
update |
Check for updates and rebuild AOSE |
stdio |
NDJSON structured-IO backend (used by the TUI) |
Run cargo run --bin aose-cli -- <command> --help for full options.
AOSE is a Cargo workspace of focused crates around Rust-owned runtime and protocol boundaries. The interaction surfaces are the CLI and TUI; external chat platforms and GUIs are optional, non-gating surfaces.
| Area | Crate | Role |
|---|---|---|
| Agent core | aose-core |
Agent loop, model chain, memory, tools, protocol, compaction, self-verification |
| Tools | aose-tools |
Tool registry, Python/Code/Notebook/R/Julia tools, Bio MAS & annotation wrappers |
| Omics core | aose-omics |
Modality types and the runtime-dispatch contract the scverse layer obeys |
| Providers | aose-providers |
OpenAI-compatible, Anthropic, Gemini streaming + mock fallback |
| Python bridge | aose-bridge |
Typed subprocess bridge to the scverse runtime |
| Knowledge | aose-knowledge |
TF-IDF + vector-store KB, biomedical API tools, practice library |
| Workflows | aose-workflow |
JSON workflow loader/registry/engine + JS/TS sidecars |
| Skills / Commands | aose-skills, aose-commands |
Markdown + JS/TS skill and prompt-command loaders |
| Report / Cite | aose-report, plus cite |
Traceable report generation and provenance verification |
| MetaHarness | aose-metaharness |
Self-improving capability: feature-gap detection, sub-agent spawning |
| TUI / CLI | aose-tui, aose-cli |
Terminal UI and command-line entry point |
Full module map: docs/ARCHITECTURE.md.
- Documentation index -- organized by audience
- Architecture -- system design, module map, runtime boundaries
- Bio-Agent guide / Multi-Agent system
- Traceability & provenance
- LLM providers
- Benchmarking
- Contributing
cargo build # debug build
cargo check --workspace # fast type-check across all crates
cargo test --workspace # run the test suite
cargo run --bin aose-cli -- tui # launch the UISee docs/CONTRIBUTING.md for the contribution workflow.
MIT. (A root LICENSE file is not yet present; the license is declared in
Cargo.toml.)