Q3.5 reflections
Hello Convex…ers? Converts? Convicts? We haven’t figured out a collective noun yet.
It’s the start of a new development sprint and time to share new product plans. Let’s first reflect on the past six weeks and how we went.
<see thread for details>
Overall I feel like we got quite a lot done, spent a bit more time on scaling than anticipated, and honed some of our priorities. Going into the next sprint we want to tackle slightly fewer but bigger projects, some of which will spill into subsequent sprints.
10 Replies
Core platform
- Better scheduled function management: ability to cancel all running scheduled functions, pause toggle for deployments that you want to stop quickly.
✅ Done
- Programmatic access to metadata so you can directly query and subscribe to storage metadata, scheduled functions and status, etc.
✅ Releasing soon
- Redesign pagination to split pages that grow too large, and potentially also allow paginating over multiple database queries in a single function.
✅ Done, although we decided to skip multi-query pagination
- Lots of small ergonomic improvements like unifying our filter/index syntax, allowing unions in validators, relaxing constraints on what characters are allowed in a field name, etc.
🤷🏼 Partially done but we didn’t find these issues sufficiently pressing
- Full-text search improvements, e.g., prefix search, fuzzy search, snippets.
❌ Most effort was spent on vector search instead, stay tuned for the coming sprint
Scale
- Scale out function execution to allow greater concurrency and dynamic memory allocation.
✅ Lots of scale improvements done, more to come
- Support larger code modules and npm modules that aren’t able to be bundled client-side.
✅ Done
Enterprise/Pro
- Add preview deployments for GitHub branches.
✅ Done
- Export streams for logs and function metrics, plus exception reporting to Sentry.
✅ Releasing soon
- Complete our SOC-2 audit.
🤷🏼 Proceeding well, won’t be finished just yet
- Edge acceleration.
❌ Skipped for now, didn’t feel pressing since performance has been good
AI
- Develop an AI Kit for bootstrapping AI apps on Convex.
✅ Convex is now on Langchain and we’ve improved our AI support, let’s call this done because we’re happy where we landed.
- Release Game Kit which encapsulates a game engine for apps on Convex.
✅ The work is done but hasn’t been marketed yet
- Scale up our vector database to support up to 1M vectors per index.
✅ Done
Developer engagement
- Better onboarding flow.
✅ Done
- First-class python client library including subscriptions.
✅ Releasing soon
- Additional client libraries for Node and Svelte.
🤷🏼 Node is done, Svelte not yet
It's neat to see this detailed retrospective, and impressive how much you all have accomplished in the last three months six weeks! :convex: 🧑💻👨💻👩💻
+1, appreciate the transparency and thoroughness! Mourning the search part, that's the one thing I'm waiting for (fuzzy, multi-field) 🙈 Keeping the faith tho 🙂
keep the faith! we'll get there 🙂
that was actually six weeks! we'll try to keep the pace up
Love it all, amazing. I like Converts since generally speaking, the market has an ingrained traditional, messier idea of how backends should be done, so there's a need to adopt the new, with evangelization paving the way.
And you can reserve Convicts for that special class that builds the next FTX on Convex. 😵💫
good to see the updates, thanks for all the hard work convex team!
fantastic transparrency, thanks guys
will all the client librarys be open sourced?
Our client libraries are actually already open source!
https://github.com/get-convex/convex-js
https://github.com/get-convex/convex-py
https://github.com/get-convex/convex-rs
future clients will also be open source
oh last time i checked they were not, thanks