HELIXA is a research studio for changing an agent’s scientific workflow without trusting a model to invent biology. Genomic pipelines can report success while the reference build, allele, or metadata is wrong. A language model can propose a DNA edit, but it cannot reliably decide the exact sequence change that would produce an arbitrary body feature. HELIXA separates those problems. A precise substitution is drafted only when the request names a coordinate and a base the loaded reference can check. A trait wish, such as greater ultraviolet tolerance, becomes an inspectable research proposal. No allele is invented, and nothing is saved until a validated edit is confirmed. The body model does not change shape just because a base changed. The hard part is recursive harness improvement. When a scientific check fails, HELIXA treats that as a gap in the agent’s policy, not as a reason to retry the same broken run. It proposes a bounded, typed change to the harness, evaluates the candidate against the parent on the original incident, negative controls, and sealed held-out cases, and promotes it only if an immutable policy envelope allows the change. The decision is stored as a cryptographic evidence receipt. A candidate cannot rewrite the evaluator, widen its own permissions, or inflate false positives. Worker leases use fencing tokens so a stalled worker cannot overwrite a newer result. MongoDB holds the durable record of that process: runs, monotonic events, checkpoints, harness versions, and evidence receipts. Raindrop is the observability path for why a harness run failed and whether a later policy was better. ElevenLabs turns a verified receipt into a spoken briefing, with on-screen captions when speech is not configured. The four tabs are the application of that mechanism: a Genome Editor, Genome and Harness Evolution, an Earth Evolution Sandbox, and a Harness Library. The Earth view plays back a stored population-genetic model. A horizon of up to two million years is a model parameter, not a reconstruction of Earth, and a genotype is not claimed to guarantee a phenotype. Research use only. HELIXA does not diagnose, treat, or design a clinical sequence.
More from The Harness Engineering & Model Wrangling #SFTechWeek Hackathon
MongoMation is an autonomous PLC engineer. You describe a machine in plain English, and Claude agents write IEC 61131-3 PLC code, compile and deploy it to a…

LabRat is a self-improving harness for a physical lab agent, running on a real bench: two SO-101 arms, a 3018 CNC carrying a servo pipette I designed during…

Recursive Harness Scientist: an AI scientist that improves its own scientific method, and must replicate every gain on data it has never seen. Why it…

Super Bid is a self-improving AI estimator that teaches itself to price bids the way a specific company would. It practices on the company’s real past bids…

A robot that gets better at handling objects it’s never seen, without retraining. After every round, an AI coach reviews what broke and rewrites the robot’s…
WORLDSTATE gives physical spaces a memory. Any phone becomes an observer: open a link, point the camera, and WORLDSTATE tracks what's inside fridges, shelves…
