Convex vs Neon: which backend should your ai app sit on?

the full guide

quick answer: for most ai apps it's Neon: #4 of 5 on our backends board at 8.0/10, because it's real Postgres that branches like git, scales to zero when nobody's looking, and doesn't charge per project, so your coding agent can have a database per pull request and your app a database per tenant. Convex is #5 (7.5/10) and it's the better pick for one job: realtime apps written end to end in TypeScript, where the UI updates itself the moment the data changes. different animals, really: one's a database you bring your own code to, the other is the database and the backend code in one.

the short version

ranks and scores come straight from our backend board, and nobody pays to move up it.

why Neon ranks higher

it's Postgres, so you can leave. psql, Prisma, Drizzle, pg_dump, the lot. your coding agent already knows SQL, and so does whoever inherits the codebase. extensions like pgvector and PostGIS are on every plan, so embeddings can sit next to the rows they describe.

branching is the party trick. a child branch is copy-on-write: it starts at $0 of storage and only grows with what you write to it. one per pull request, one per test run, one for the agent that's about to try something daft with a migration. you get 10 branches per project on Free and on Launch, 25 on Scale, then $1.50 a branch-month, billed by the hour.

built for a database per tenant or per agent. there's no per-project fee, the free plan gives you 100 projects, and idle databases cost nothing for compute. Neon's own sum: 100 tenant databases on Launch, each running about an hour a day, comes to roughly $85 a month in total.

the free plan doesn't pause you. it's permanent, needs no card, and Neon says free projects aren't paused for inactivity: compute naps after 5 minutes and wakes in a few hundred milliseconds. you get 100 CU-hours and 1 GB of storage per project (20 GB across the account) and 5 GB of egress per project. run out of CU-hours or egress and compute is suspended until next month; your data isn't deleted.

the catch: pay as you go means no fixed bill. on Launch compute is $0.106 per CU-hour and on Scale it's $0.222, plus $0.35 per GB-month of storage. Neon puts a small database that scales to zero at about $15 a month on Launch. spending notifications on paid plans only email you; to actually stop the meter, cap autoscaling or set consumption limits through the API.

where Convex wins

realtime is the whole point, not a feature. a Convex query is a TypeScript function, and the client subscribes to it. when any data that query read changes, Convex reruns it and pushes the new result down the WebSocket. useQuery in React and the component just updates. no subscription plumbing, no cache invalidation. Neon's realtime is still a beta.

the backend code lives with the data. mutations are transactions with no begin or end to write and retry on conflicts, queries are read-only, and actions are where you call your LLM or send emails. crons, the scheduler, file storage, auth and text and vector search are on every plan. with Neon you bring your own server code (or use Neon's functions) and wire the realtime bit yourself.

it suits ai-written code. end-to-end types and no SQL means your coding agent writes the backend in the same TypeScript as the front end. Convex makes that pitch itself; we'd say it's fair for TypeScript shops, less so if your team thinks in SQL.

the honest bit: it's not Postgres. it's a document-relational database (the cloud runs on PlanetScale MySQL underneath), so your SQL, your ORM and most of your Postgres tooling stay at home. vector search only runs in actions, so those results aren't reactive. on the free plan you get 1M function calls, 0.5 GB of database storage and 1 GB of file storage a month. Convex Professional is $25 per developer a month (25M function calls and 50 GB of database storage included), and Business and Enterprise has a $2,500 monthly minimum, which is a proper cliff if you outgrow it.

which should you pick?

• a SaaS or ai app where you want SQL and an exit door: Neon
• a database per pull request, per tenant or per agent: Neon
• RAG over your own tables with pgvector: Neon, or Supabase if you want auth and storage in the same dashboard
• a collaborative or live-updating app (chat, multiplayer, dashboards) in TypeScript: Convex
• you never want to write SQL or manage subscriptions: Convex
• the whole kit in one box: Supabase is #1 (9.0/10). see Neon vs Supabase and Convex vs Supabase
• a long-running agent or worker that needs a server, not a database: Railway (#3)

the rest of the field

faq

is Convex better than Neon?

not on our board: Neon is #4 (8.0/10) and Convex #5 (7.5/10). Convex is the better pick for realtime TypeScript apps; Neon for apps that want Postgres, branches and a database per tenant.

is Convex built on Postgres like Neon?

no. Convex is a document-relational database you query in TypeScript. its cloud runs on PlanetScale using MySQL for persistence, and the open source version can use SQLite, Postgres or MySQL underneath. you don't write SQL against it either way. Neon is Postgres all the way down.

Convex vs Neon pricing: which is cheaper?

both have a free plan with no fee to start. Neon's paid plans have no monthly minimum: you pay per CU-hour and per GB-month, and a small database that scales to zero is about $15 a month on Launch by Neon's own estimate. Convex Professional is a flat $25 per developer a month with bigger allowances, then per-unit rates. so solo and idle, Neon is usually cheaper; a busy realtime app may fit Convex's included allowances nicely. the Neon sums are in Neon pricing.

does Neon do realtime like Convex?

Neon lists realtime as a beta, free during the beta, with compute and egress billed as usual. in Convex every query is live by default.

can i use pgvector with Convex?

no, pgvector is a Postgres extension. Convex has its own vector search: define a vector index in your schema and query it from an action, with cosine similarity and up to 256 results per search. on Neon, pgvector is one of the extensions on every plan.

sources

checked 2026-11-02

• Neon pricing (Free, Launch and Scale, CU-hour and storage rates, branches, scale to zero, no inactivity pause, auth MAUs, realtime beta, typical monthly cost, database per tenant example, spending limits, Postgres extensions incl. pgvector, from Databricks; page says last updated 2026-10-31): https://neon.com/pricing
• Convex pricing (Free and Starter $0 then pay as you go, Convex Professional $25 per developer a month, Business and Enterprise $2,500 monthly minimum, included resources, vector and text search, preview deployments): https://www.convex.dev/pricing
• Convex overview docs (reactive database, TypeScript queries, mutations as transactions, actions, scheduler, WebSocket clients, PlanetScale MySQL persistence, open source storage options): https://docs.convex.dev/understanding/
• Convex vector search docs (vector indexes, search only in actions, cosine similarity, up to 256 results): https://docs.convex.dev/search/vector-search
• ranks, scores, verdicts and plans: numptr data/tools.json (backend, database and deploy board)