Common practice
When a TPM, FPKM or log table has a known batch the design can hold, put the batch in the limma design. The test never takes adjusted values.
Inflexa plans, runs and records your analyses with any model, including open weights on your own GPUs. Its Knowledge Plane grounds every plan in published methods, so an open model plans like a frontier one. Run your own model and your data and prompts stay on your network.
Free to start: the open-source CLI, or the Cloud Personal plan.
Question: Which genes and pathways distinguish responders to pembrolizumab in GSE78220?
Plan: Anti-PD-1 responder DE and GSEA. Steps: Cohort assembly, Responder DE, Immune and stromal mix, GSEA on hallmark sets, Survival association.
Findings: Gene selection repeats inside every fold; Recorded sex is checked against the transcriptome.
Sessions built from public datasets: GSE78220, the Hugo et al. anti-PD-1 cohort.
Inflexa planning benchmark, 54 bulk RNA-seq planning tasks, 14 Sep 2026. Single run, scored by a blind AI judge.
The lower-cost and open models gained the most. Qwen 3.8 27B is open weights, so a team can host it on its own GPUs and keep every prompt inside its network.
Grounding lifted every model we tested.
Plan quality, 0 to 100
| Model | Alone | With the plane | Cost per plan |
|---|---|---|---|
| Claude Opus 5 | 81.7 | 90.9 | $0.72 alone |
| GLM 5.3 | 69.8 | 89.2 | $0.14 with the plane |
| Qwen 3.8 27B | 62.9 | 83.4 | $0.09 with the plane |
$0.72
Claude Opus 5 alone
$0.14
GLM 5.3 with the plane
up to 80% lower cost per plan
with a model you choose on your machine
Qwen 3.8 27B with the plane: $0.09 per plan
0 invented references in any plan made with the plane
Inflexa planning benchmark, 54 bulk RNA-seq planning tasks, 14 Sep 2026. Single run, scored by a blind AI judge.
The Knowledge Plane holds rules for when each method applies, drawn from the literature, with tested script templates. Inflexa matches your study design to those rules before it writes a plan, then checks the plan against them.
Your machine
Never sent
sample rowsfile pathscolumn namesidentifiersKnowledge Plane
Common practice
When a TPM, FPKM or log table has a known batch the design can hold, put the batch in the limma design. The test never takes adjusted values.
Consensus
When the values are TPM, FPKM or RPKM, use limma-trend on the log values. DESeq2, edgeR and voom need raw counts.
Back on your machine
Plan checked: no rule broken
Pick the model per analysis: Claude or GPT through their APIs, or any OpenAI-compatible endpoint, including open weights on your own cluster. With the Knowledge Plane an open model outplanned a frontier one, so the GPUs you already own can carry the work.
Model for this analysis
Hosted Inflexa Cloud runs Claude. On Enterprise, deployed in your own cloud account, you choose the model.
What crosses the line
Your model provider
prompts and excerpts, to the one you chose
Public sources
the lookups and web searches it makes for you
Knowledge Plane
counts and categories, paid plans
The Knowledge Plane never receives sample rows, file paths or identifiers.
Point Inflexa at a model you host and the prompts stay on your network too.
Describe the biology in plain language. Inflexa reads your data profile and the literature before it proposes anything.
You decide the question
You get a plan with every method and parameter laid out. Nothing runs until you approve it.
You decide the plan
Each step runs in a sandbox, and independent steps run side by side. Change course at any point.
You decide when to change course
Results arrive with the record of how they were made. Ask any output where it came from.
You decide what the biology means
The planner, the sandbox, the signed record, the skill packs and the database connections are the same in every product. Choose where the work runs and how you like to work.
Not sure where to start?
On your machine
Open source, in your terminal.
Free
curl -fsSL https://inflexa.ai/install.sh | bashOn your machine
A full workspace, served from your machine.
In the cloud
Your team's workspace, with compute.
Personal free · Biotech · Enterprise
Many tools now write and run an analysis for you. The difference is whose model does the thinking, where it runs and who can read the code.
| What it does | A chat assistant | An AI scientist from a model provider | Inflexa |
|---|---|---|---|
| Runs the analysis for you | Some do, on the vendor's servers | Yes, in a sandbox | Yes, in a sandbox on your machine, your cloud or ours |
| Models | The vendor's | The vendor's | Any, including open weights you host |
| Model on your own GPUs | No | No | Yes, behind any OpenAI-compatible endpoint |
| Prompts stay on your network | No | No | Yes, with a model you host |
| Source | Closed | Closed | Apache 2.0 engine |
Inflexa adds two things to the table. The Knowledge Plane checks each plan against published rules for when a method applies. And every result carries a signed record that anyone can verify offline with an open-source verifier.
Each is a question on public data, with the plan Inflexa proposes for it. Open one to see the evidence. The last one is yours.
A run on GSE109142, 206 patients: three mechanism-anchored hypotheses, one disqualification the direction analysis forced, and CXCL13 as the candidate the evidence kept returning to.
Read the case studyOnce PK, safety and clinical readiness are scored, rosiglitazone falls from the top biology score and sodium butyrate rises to first.
Install the CLI and pay nobody but your model provider, open a free Cloud account, or talk to us about Pro for your team.