Local-first agentic AI orchestration for reproducible biological data analysis.

---

Inflexa turns a plain-language analysis request into runnable code, executes it in an isolated sandbox, and records exactly what happened so the result can be reproduced and audited. It runs entirely on your machine and works with the model provider of your choice, including local models for a fully offline workflow.

It is built for scientists, bioinformaticians, and engineers who need analysis they can trust and re-run, not just an answer in a chat window.

Why Inflexa

• Local-first. Your data, code, and results stay on your machine. No account required.
• Reproducible by construction. Every run records its full provenance and lineage in a local SQLite database, with export and replay.
• Sandboxed execution. Generated code runs in an isolated, unprivileged, resource-limited sandbox with no network access by default. The full isolation model is described in SECURITY.md.
• Bring your own model. Use any supported LLM provider via API key, or run local models end to end, offline.
• Open source, in full. The CLI is a complete product under Apache-2.0, not a limited trial. See Open source and commercial.

Quick start

Install with Homebrew (macOS, or Linux on x86_64):

brew install inflexa-ai/tap/inflexa

Or with the install script (same platforms, no Homebrew needed) — it downloads the latest release, verifies its checksum, and installs to ~/.local/bin:

curl -fsSL https://inflexa.ai/install.sh | bash

On Windows (PowerShell):

irm https://inflexa.ai/install.ps1 | iex

Or from npm (any platform above, Node.js 18+) — also runnable without installing via npx @inflexa-ai/inflexa:

npm install -g @inflexa-ai/inflexa

On other platforms, download a binary from the latest release, or run from source.

All you need is Docker, running locally — analyses execute in the sandbox image. The inflexa CLI itself is self-contained.

inflexa setup            # one-time: connect a model provider, pull the sandbox image, start local services
cd path/to/your/data     # go where your data lives
inflexa                  # launch the TUI

Prefer to run from source? See Running from source.

Configuration

inflexa setup walks you through the model connection: sign in to a provider through the local proxy, or point Inflexa at your own endpoint — including a local model, for a fully offline workflow. It prompts for what it needs and tells you how to supply your key.

In one pass it provisions everything an analysis needs:

• Model connection — sign in to Claude, OpenAI, Gemini, Qwen, or iFlow through the managed local proxy, or point at your own endpoint.
• Postgres + pgvector — the database the harness runs on, provisioned as a local container via Docker Compose.
• Analysis resource allowance — the share of your machine's CPU and memory that analyses may use in total.
• Embeddings — an in-process local model, your own API key, or off.
• Sandbox image — pulls the Docker image analyses execute in (Python, or Python + R).

To change any of it later, run inflexa config.

Running from source

Developing, or building the binary yourself, additionally requires Bun. The CLI lives in cli/:

git clone https://github.com/inflexa-ai/inflexa.git
cd inflexa/cli
bun install

bun run dev                 # launch the TUI from source
bun run build               # compile a standalone dist/inflexa-<os>-<arch>

See cli/README.md for the full CLI developer guide, and CONTRIBUTING.md for the development environment and contribution workflow.

Repository layout

This repository is a monorepo of independent subsystems — work inside the one you are changing; each has its own dependencies and tooling.

Directory	What it is
cli/	The local-first TUI/CLI — this product. SQLite storage, auth, the chat UI. Start here to run from source.
harness/	@inflexa-ai/harness, the host-agnostic agent harness: agent loop, durable workflows, sandbox protocol, providers. The execution model and its design decisions live here — see harness/CONTEXT.md and the specs in harness/openspec/specs/.
skills/	Shared bioinformatics skill packs the agent loads at runtime.
templates/	Report-rendering templates.
images/	The sandbox images: the base image with its Go execution server, and the published python / python-r variants with the analysis packages baked in.
scripts/	Build, validation, and publishing tooling for the sandbox library store.

Open source and commercial

Inflexa is developed by Inflexa, Inc. The open-source CLI in this repository is a complete, standalone product under Apache-2.0. A separate commercial hosted platform adds team- and organization-scale capabilities such as collaboration, governance, managed infrastructure, and compliance support. The boundary, and the commitments behind it, are described in COMMERCIAL.md.

Contributing

Contributions of all kinds are welcome: code, example datasets, documentation, validation reports, and triage. Start with CONTRIBUTING.md. The project follows a Code of Conduct and the governance model in GOVERNANCE.md.

Security

Please do not report security vulnerabilities in public issues. Follow the private disclosure process in SECURITY.md.

License and trademarks

Code is licensed under Apache-2.0. The Inflexa name and logo are trademarks, handled separately in TRADEMARK.md: the code is yours to use and fork under the license, but a redistributed fork should be renamed.