Getting started.
From an empty terminal to an analysis whose every step you can trace. About fifteen minutes, most of which is a download.
Install
Downloads a prebuilt binary. You do not need Node, Bun, or a Python environment — but you do need Docker or Podman, and Inflexa will not start without one.
curl -fsSL https://inflexa.ai/install.sh | bashOther install methods are on the install page.
Provision the sandbox
This pulls the sandbox image and starts the local database. The image already contains R, Python, Node, and the bioconda toolchain, which is why you never install an analysis package by hand.
inflexa setupThe first pull is a few gigabytes. It happens once.
Choose your model
Inflexa does not ship an intelligence. You bring one. It works with Anthropic and with any OpenAI-compatible endpoint, which includes a model running locally on your own hardware.
Your API key stays in your config on your machine. Inflexa has no account, and there is no Inflexa server for it to travel to.
inflexa auth loginStart it
This opens the terminal interface: a chat pane, a sidebar tracking your session, analysis, data profile, and runs, and a command palette.
inflexaPoint it at a dataset and describe what you want to know. It reads the relevant literature, plans the analysis, writes the code, runs it in the sandbox, and reports back. You can inspect and override the plan before anything executes.
Ask where a result came from
This is the part that matters. Every file Inflexa produces carries a record of the commands and inputs that produced it, and which model reasoned about each step.
inflexa prov lineage results/volcano.pngExport the record as W3C PROV, and hand it to someone who does not trust you. They can verify it offline, with no account and no server.
inflexa prov export my-analysis --format jsoninflexa prov verify-file provenance.jsonThe rest of it.
The terminal interface
The chat pane, the sidebar (session, analysis, data profile, runs), the command palette, and the slash commands for switching analyses, managing projects, and exporting provenance.
Analyses and runs
Runs are checkpointed. If the process dies, the machine reboots, or you close the laptop, the run resumes from where it stopped rather than starting over. Long analyses are expected to survive interruption.
Skills
Twenty-plus packs covering bulk and single-cell transcriptomics, spatial, chromatin, methylation, variants, proteomics, metabolomics, microbiome, multi-omics integration, enrichment, cheminformatics, drug repurposing, immune profiling, PK/PD, translational safety, and statistical modeling. They teach the agent a real tool, with its actual interface and failure modes.
The sandbox
Every step runs non-root, with all Linux capabilities dropped, no new privileges, memory and CPU limits, read-only inputs, and no network access unless you grant it. Your analysis cannot phone anywhere, and neither can Inflexa.
Provenance
Records are built on the W3C PROV data model and export as PROV-JSON or PROV-N. Each is signed with Ed25519 over a SHA-256 hash chain, so any change after signing is detectable. Verification needs nothing but the file and its signature.
Databases
The agent queries the scientific record directly: PubMed, Open Targets, ChEMBL, PubChem, GEO, ClinVar, the GWAS Catalog, ClinicalTrials.gov, STRING, Reactome, FAERS, PharmGKB, and more. A few need an API key of your own.
Stuck, or think something is wrong?
The code is open. If the answer is not here, it is in the source, and if the tool is wrong, we would rather hear about it than not.