Convex @convex.dev is quickly becoming the go-to sync engine for startups building AI and agentic apps. At Sync Conf 2025, co-founder James Cowling @jamescowling.dev will share how theyβre making it happen.
Convex @convex.dev is quickly becoming the go-to sync engine for startups building AI and agentic apps. At Sync Conf 2025, co-founder James Cowling @jamescowling.dev will share how theyβre making it happen.
Convex Chef supports Gemini 2.5 Pro! It's sometimes too verbose and occasionally forgets to call tools but produces way better UIs and seems able to build more fully-featured apps than Claude. Try it out!
Chef is fresh out the oven so let us know what you think and hit Submit Feedback with any issues!
chef.convex.dev
LLMs love writing @convex_dev code so we made an app-builder to let them do it.
Convex Chef: The AI builder that knows backend.
Get cooking...
Convex Vibe Week Day 3:
Get started with Convex without signing up for anything: stack.convex.dev/anonymous-de...
Ain't no one got time to log in to stuff. Now you can use @convex.dev without an account! Just npm i convex and npx convex dev.
Convex Vibe Week Day 2:
Agent framework
www.convex.dev/components/a...
stack.convex.dev/ai-agents
Durable workflow engine
www.convex.dev/components/w...
stack.convex.dev/durable-work...
Building an AI agent or need a durable workflow? Day 2 of @convex.dev Vibe Week is dedicated to you.
New agent framework and durable workflow engine ready to go, check out links in the reply.
Convex Vibe Week: www.convex.dev/vibe-week
Day 1:
Convex MCP in GA: stack.convex.dev/mcp-ga
Day 1 of @convex.dev Vibe Week is a go! Announcing official release of our MCP server folks have been trialing. Check out the vid and the links in the reply. Stay tuned for more tomorrow!
Convex has a referral program! Double your free quota by referring someone or applying a referral. Refer up to 5 people and get a bump each time.
Find your referral link under Team Settings in your dashboard, or if you're fast you can use mine: www.convex.dev/referral/JAM....
Convex storage, search, pagination and optimistic updates are now officially "non-beta". They're reliable to use and have been for a long time.
We'll continue to make improvements as always, especially to pagination and optimistic updates.
Remember when a beta label actually meant something then Gmail was beta for half a decade and ruined it for everyone?
We just took a bunch of @convex.dev features out of beta. What's the point of long-term beta if customers depend on it already?
Your coding agent talking directly to your production database sounds sketchy as hell... unless it's via an MCP server on a db that supports sandboxed code execution, in which case it sounds cool af.
"Constant time" -> constant in the number of rows in the db. Convex keeps a running inferred schema for each table to perform optimizations like this.
Ok this is legitimately cool. Cursor talking to @convex.dev via MCP and figuring out schema improvements based on the rows actually stored in the database, performed in constant time!
3. Design strong abstractions.
Agents aren't perfect so focus them on building your app, not trying to solve hard systems problems like managing distributed state. Use libraries or frameworks with good high level abstractions that prevent an agent from wandering off the path.
2. Express as much as possible in standard, procedural code.
Agents are great reasoning about code but less good dealing with declarative languages or configs outside the codebase. Keep as much as possible in the code itself and as little as possible in SQL or external configs.
1. Provide tight, automatic feedback loops.
Coding agents are able to iterate, try things, correct mistakes. They are only able to do this when able to efficiently determine if a step is correct. Type safety is hugely effective at providing guardrails for agents.
We launched Fullstack-Bench yesterday to test autonomous agent performance on fullstack coding tasks. Starting with Convex, FastAPI, Supabase but more backends to come.
stack.convex.dev/introducing-...
Three design patterns stood out at producing high quality code with minimal interventions:
π§΅
Thanks @brancz.com! The team loves using Polar Signals so itβs certainly a bonus that itβs made by awesome people.
Do you like TypeScript? Do you like databases? Do you like neither but just want to build apps using regular old functions?
My @db.cs.cmu.edu talk on TypeScript Stored Procedures for app developers is online now.
www.youtube.com/watch?v=zpgV...
Convex also now members of the Open Source Pledge @opensourcepledge.com, committing to financially support development of open software we all benefit from.
Our initial contribution was towards @tanstack.com! Really glad to be supporting @tannerlinsley.com and crew.
news.convex.dev/open-source-...
If you're new to Convex start off with the tutorial in our docs. If you're a Convex OG and want to give self-hosting a spin, you can get started over at github.com/get-convex/c...
Cloud-hosted Convex is still the best place to start for most folks getting started, plus includes a very usable free tier, but we're really glad to have the self-hosted option available for people whose needs fall outside our hosted plans.
When people say they want open source they rarely mean they just want to read the source code. They mean they want unrestricted access, an active community, ease of use, and yes, often something for free.
news.convex.dev/self-hosting/
We just launched self-hosted @convex.dev! We open sourced last year but realized the code was open but not so useful on its own. This time we've included the Convex dashboard, bindings for Postgres, SQLite (and MySQL v. soon), packaged up in a docker container ready for real use
CMU Databases talk streaming live 4:30EST Monday.
Come learn how every good idea you have was already invented by someone in the 80s and why we should be running typescript functions in our databases.
db.cs.cmu.edu/events/sql-d...
Summary of how well each model is performing at writing
@convex.dev code?
I used AI to write a script to help me write AI evals
to help AI write Convex code. Very meta
π‘Each step uses RAG from previous evals
Flow:
1. π§βπ»oneliner -> π€ task description -> π§βπ» audit
2. π€ answer ->π§βπ» audit
3. π€ tests for answer ->π§βπ» audit
π to evals π & my script in π§΅