Hey HN!

Inflexa is an OSS-first TUI for agentic-AI reproducible biological analysis with provenance tracked on every step. It's a better designed Claude Science (bar the GUI for now).

Before this, we spent years building a cloud multi-omics analysis platform, working alongside research teams in pharma, biotech, and academia.

Our vision is as follows: - OSS first - Data privacy, and security are of utmost importance - Agents are untrusted actors

Agentic products do not solve a core architectural and design problem: provenance.

How do you build up the body of evidence to support your claims from hundreds of chat messages, scattered artifacts (scripts, intermediary data files, random research files)?

We are one of the first in the biotech space to propose using an established specification for provenance, which its extremely fitting for the world of agentic AI: [W3C PROV](https://www.w3.org/TR/prov-overview/).

Our provenance is deterministic and driven programmatically. Our architecture imposes it, not the whim of agents that decide whether or not to call a tool / or an MCP (such as https://arxiv.org/abs/2508.02866).

The analysis is backed by a PROV document that maintains the lineage of every step, every action, and every file.

Other platforms such as Claude Science are a security disaster waiting to happen: they allow agents to determine which Python & R packages they need to perform an analysis step. Sure, you need to approve the installation, but if you get prompted tens of times, are you really going to vet every package and every version?

We offer batteries-included sandboxed execution. The analysis' code is generated by agents and executed in ephemeral sandboxes. We do not install packages on your system.

Curious to hear your thoughts!