Stacks/SaaS / agent platform

ODSEA's LangGraph + Next.js + Supabase production stack

ODSEA published the stack it actually runs: Bun, Next.js 15, Supabase PostgreSQL, Drizzle, LangGraph (over CrewAI), Infisical, and Vercel. The post is written as ADRs — what they chose, what they rejected, and what they gave up — for a platform that runs agent workflows around the clock. Models named in the architecture wrap-up are Claude and GPT.

Use case

A production platform that 'serves thousands of users' and 'processes AI agent workflows 24 hours a day', documented as a set of public architecture decision records

Category

SaaS / agent platform

Team size

not stated

Orchestrator / framework

LangGraph is the AI agent orchestration layer (chosen over CrewAI). The rest of the application is Next.js 15 App Router running on Bun, deployed on Vercel.

Memory

Supabase (PostgreSQL) is the database and auth layer, accessed through Drizzle ORM, with a schema the authors say spans over 200 tables and 16 schema files.

Evals

not stated

Observability

not stated

Monthly cost

not stated

Models

ClaudeGPT

Observability tools

not stated

Other tools

BunSupabaseVercelInfisical (secrets)Supavisor connection pooler

Source

https://odsea.com/blog/production-ai-stack-langgraph-nextjs-supabase

Alex Chen, CTO and co-founder · May 25, 2026