Ian Macartney

friendly engineer. works at infra lead for previews @ Dropbox
🦋

[Ian Macartney @ianmacartney]

@noname_oni @theo @mikeysee @convex Yes!
You can now require env vars and also configure how they propagate to components in convex.config.ts.
New deployments inherit default env vars, now for local dev too.
You can also read/write env vars via files or scripts via CLI - including defaults.

[Ian Macartney @ianmacartney]

@tobeycodes good call - I opened an issue

[Ian Macartney @ianmacartney]

@JcCastaneda05 @mikeysee @robelestifanos_ @convex have you tried setting the env var CONVEX_AGENT_MODE=anonymous?

[Ian Macartney @ianmacartney]

@MeshanKhosla @huntjordan22 Thanks! @ballingt did a lot of the heavy lifting too -esp. on the bundling and internal capabilities

[Ian Macartney @ianmacartney]

@GorkaCesium "Cooling the ASICs would be trivial" -> you sure?
Lacking conduction/convection leaves only radiation, which sounds tricky for heat dumping

[Ian Macartney @ianmacartney]

@tiagoefreitas @kiwicopple I went with the prosemirror OT model over Yjs CRDTs. I started exploring CRDT sync in Convex via the automerge-sync component, but for prosemirror-sync I found the rebase approach to be more correct. But if automerge has a lighter weight server impl that would be great to know

[Ian Macartney @ianmacartney]

@tiagoefreitas @mdhruvill @Cloudflare @tan_stack @convex @coderabbitai @Sentry I've been thinking about what a git-compatible VFS Component might look like, cool to see this impl
Not sure how hard it'd be to do with Convex Workflow / etc.
Why not embrace the DO approach?
@Cloudflare is very complimentary to @Convex IMO
esp. on the hosting/r2 front

[Ian Macartney @ianmacartney]

Long time coming, but you can now do
v.nullable(...)
as shorthand for v.union(..., v.null())

[Ian Macartney @ianmacartney]

https://t.co/LtudqcOjpQ

[Ian Macartney @ianmacartney]

Handy validation utilities landed in @convex 1.29:

v.object({name: v.string(), age: v.string()})
.pick("name")
.extend({id: https://t.co/KsYsHoB04M("users")})
.partial();

I'll be working on the core API in addition to convex-helpers & components

[Ian Macartney @ianmacartney]

Grok did a decent job, but to make it more concrete I've built two components to do this.
Workpool - durable functions with steps, suspension until time/event & more
Workpool - general-purpose queues w/ parallelism limits, retry configs, exactly-once execution of onComplete

[gary IH fung @garyfung]

@convex Does Convex have ergonomic queues and durable functions @grok ?

[Ian Macartney @ianmacartney]

@essonam_m @convex @jamesacowling @jamwt @rasmic I pushed a new version that supports the string version of models - previously you had to use the gateway provider

[Ian Macartney @ianmacartney]

@lgrammel I'm adding to @​convex-dev/workflow * to improve HIL, incl. AI SDK loops- adding pause/resume- akin to promises / signals / condition variables for Convex fns.

Planning to ship soon - would love to chat / coordinate!

*Durable async serverless functions akin to termporal

[Ian Macartney @ianmacartney]

..of course sometimes skipping meaningless stack frames is good, and nothing is absolute

[Ian Macartney @ianmacartney]

I feel clever subbing `return await fn()` with `return fn()`

But gotchas don't seem worth it anymore
• Breaks try/catch/finally expectations
• Loses stack trace context

My default is now to always await. Change my mind.

[Ian Macartney @ianmacartney]

try {
return foo(); // async function
} finally {
// executes before foo finishes*
}

* code in foo prior to the first await may run first

[Ian Macartney @ianmacartney]

@ZubairIbnZamir @convex It's up to date! There's also an ongoing effort to support zod4, which will be backwards compatible (using the same clever trick zod used of a separate /zod4 entrypoint)

[Ian Macartney @ianmacartney]

@grinich The slides from the talk on Async are here: https://t.co/lwjZgZ6h5V

[Ian Macartney @ianmacartney]

Big updates to the Agent Component:
- Abort an async stream remotely
- Pending, failed, and aborted streamed messages auto-save
- AI SDK v5 support
- Easier to use alongside AI SDK vs. wrapping
- Type-safe custom context passed to tools
- <SmoothText>msg.text</SmoothText>

[Ian Macartney @ianmacartney]

@vishyfishy2 You can do either - http streaming has always worked via the `toUIMessageStream` & co.
Delta streaming is opt-in and streams only the changes to all subscribers (and can be resumed).
Doing both is possible right now - simplest is to read both the deltas & HTTP & de-dupe tho