Refine is a deep-review engine for technical documents. It uses an agentic pipeline of frontier AI models to read a draft end-to-end, track detailed arguments, and surface the kinds of substantive issues an expert reviewer would flag.
- Expert-level depth — Feedback comparable to a careful 4–6 hour expert read-through, delivered in minutes
- Order of magnitude deeper — Coverage far more thorough than frontier models, even with expert use
- Built for complexity — Especially strong on long, technical, or complex documents where precision and internal consistency matter
- Multimodal — Reviews figures, charts, and tables alongside the prose, not just the text
Our core users are researchers at leading universities, think tanks, and central banks — working in fields like medicine, economics, computer science, and law.
Backend — Python, FastAPI, Azure Functions, Redis for orchestrating multi-stage agentic pipelines at scale
Frontend — TypeScript, TanStack Start, React
Infrastructure — Azure, Pulumi IaC, containerized deployments
AI — Custom routing and orchestration across frontier LLMs — OpenAI, Claude, Gemini, and others
Zero data retention with LLM providers — your documents are never stored by the models processing them, and never used for training. SOC 2 and ISO 27001 audits in progress. See our trust center.
Refine ships an MCP server and a REST API, so you can run deep reviews directly from your editor or wire them into your own pipelines.
Get started → — setup, REST API reference, and MCP server config (Claude, Cursor, VS Code).
The developer APIs are in Alpha; contracts may change without versioning.