Your LLM has never met your organism.
Pheno is the memory layer that introduces them.
A general model knows the chemistry of every organism and the particulars of none. Hand it your organism’s memory and the same model starts proposing experiments you would actually run.
Reasoning is only as good as what the model can see.
Ask which knockout raises a target in your strain and a model answers from textbook pathways and a reaction it remembers from another organism. Fluent, confident, and not about your organism.
Coding agents had the same weakness until someone gave them a memory of the codebase. Biology needs that layer more, because a wrong design goes to the bench and costs months.
“A fluent guess still fails at the bench.”
Pheno hands your LLM the context that matters.
Pheno compiles what is known about your organism into one memory and serves it to the LLM you already use. The suggestion it makes is about your organism, under your conditions, with the evidence attached.
- Your genome
- What the organism can run, reaction by reaction, with direction and evidence.
- Donor organisms
- Who carries the step you are missing, nearest relatives first.
- Metabolism and conditions
- How pools move on your feedstock, under the conditions you measured.
- Literature
- Every paper read whole, every claim tied to its quote and page.
Every investigation ends in a report you can act on.
The answer is written up as a living scientific report: what to do and why, what was found, the route, and the next experiment with its controls. Each line links to its evidence, and the report is pinned to the release it ran on, so it still holds when someone opens it later.
“A finding without its evidence is a rumour. A report carries both.”
Raising a target molecule on a new feedstock
Objective: pick the first edits to test now that the host grows on the new feedstock.
- Cut the drain on the target first; the host runs it on this feedstock.
- Bring in the donor step only after the knockout is measured; one change at a time keeps the result readable.
- Treat the second route as unproven; one reaction has no host enzyme yet.
- Every host step to the intermediate has a gene behind it. evidence
- The drain runs on this feedstock; the donor step is reference chemistry. evidence
- Controls
- Parent strain on the same feedstock
- Measurements
- Target and intermediate across the run
- Decision
- Donor step only if the target rises
A memory for every engineered organism.
We are starting with strain engineering because the cost of forgetting is highest there. The model is drawn wider: a memory for every organism anyone engineers, kept current by agents that read each new paper the night it lands.
If you engineer an organism, we would like to build its memory with you.
“The next strain programme should start where the last one ended.”
Read next
How it works