Skip to main content

Overview

Genomic Intelligence runs transformer language models over DNA sequences for six tasks: promoter, splice site, enhancer, chromatin, expression, and annotation. The same models are exposed two ways: a typed REST API and an MCP server.

Full API reference

ReDoc is the full API reference, generated from the live OpenAPI schema. It documents every endpoint, field, and response. The pages here are guides and orientation — they don't restate the schema.

Two ways to integrate​

Integrate directly​

Call the REST API from any language. Authenticate with an API key, post a sequence to /v1/tasks/<task>/predict, and get back typed results. Use this when you are wiring the API into an application or pipeline.

REST API →

Use it in an AI chat​

Ask questions over DNA in Claude, ChatGPT, or your own agent and let it call the six tasks as tools. Start with no key and no install via the hosted GI connector, then move to a local server for real work.

Use GI in an AI chat →

The contract​

Both paths share one contract:

  • Six tasks, one endpoint each: POST /v1/tasks/<task>/predict for promoter, splice, enhancer, chromatin, expression, annotation — plus two workflows: find-genes-and-predict-expression and genomic-variant-interpretation, which takes a VCF rather than a sequence.
  • Bearer auth on every /v1/* route. Public (no key): /health, /docs, /redoc, /v1/openapi.json, and capability discovery — GET /v1/tasks/{task}/models, so you can read each model's window geometry and request_max_bp before you have a key. Discovery is rate-limited per source IP.
  • One success envelope — {data, meta} on every 2xx and 202. data is task-specific; meta is uniform.
  • One error envelope — {error: {code, message, request_id, details?}}. Switch on error.code, never the message.
  • Sync by default, async on request — add Prefer: respond-async to submit a job, then poll GET /v1/tasks/jobs/{job_id}.
  • Per-task bounds — each task sets its own minimum sequence length (promoter 300 bp … expression 9,198 bp) and its own closed options; the maximum is 500,000 bp everywhere. See Limits.

No key yet? Email contact@genomicintelligence.ai.

The tasks​

TaskWhat it does
PromoterClassify whether a sequence contains a promoter region.
SpliceAnnotate donor and acceptor splice sites across long sequences.
EnhancerScore developmental and housekeeping enhancer activity.
ChromatinPredict chromatin features (DNase, CTCF, histone marks, …) as grouped tracks.
ExpressionPredict gene expression from sequence + an experimental description, in log(TPM+1).
AnnotationFind genes by detecting TSS and PolyA signals, emitting transcript intervals.
Find genes + expressionComposite: find the genes in a region and predict each one's expression in one call.

Integrating from a coding agent​

Point a coding agent at the llms.txt index so it works from the same contract, or drop the single-file agent guide into your repo as AGENTS.md (or CLAUDE.md) — the raw Markdown is served at /AGENTS.md. For MCP clients, the MCP server reaches the same six tasks as typed tools.

Live URLs​

For research and development use, not clinical or diagnostic decisions. Prefer to explore in a browser? Try the web app. Questions: contact@genomicintelligence.ai.