OnTract Contract Management System

Contract-analysis tools tend to be built and priced for enterprise legal teams. OnTract aims to give smaller businesses similar summarization, clause extraction, and document search at a lower cost, so they do not have to review every agreement by hand.

Team-level description based on a teammate's public case study of the project.

Seven people built the platform: the Next.js front end, the Supabase backend (authentication, database, storage, edge functions), the AI pipeline for summarization, clause extraction, and chat, and the billing, tracing, and email integrations.

• Core. A Next.js front end on Supabase (authentication, Postgres, storage) and Supabase edge functions. Only Supabase is required to run the core app.
• AI processing. Edge functions call OpenAI to summarize contracts and extract key clauses when a document is ingested. LangChain and ChromaDB provide semantic search and multi-turn chat over ingested contracts, and answers point back to the source excerpt.
• Observability. LangSmith traces LLM calls, so a wrong answer can be traced to retrieval, the prompt, or the model.
• Other integrations. Stripe for subscription billing and Resend for team and client invitation emails. These are optional and only switch on when their keys are configured.

Sources: the team's local setup guide and a teammate's public case study.

• Chunking on clause boundaries rather than fixed sizes, so a clause stays together with its key figures.
• ChromaDB over a managed vector service to avoid recurring hosting cost while demand was being validated, with the retrieval layer kept swappable.
• Optional-by-default integrations, so a new teammate can run the app locally with only Supabase.

Team decisions, as described in a teammate's public case study.

As the team has noted, pre-launch evaluation relied on manual spot-checks rather than a repeatable test set, and pipeline latency was not formally measured. Scanned PDFs without clean OCR give weaker summaries.