Ryan Rasti @ryantypes

Building Typegres - sandboxed client SQL, tables as TS classes Former Google Tech Lead typegres.com Joined July 2025

Tweets

83
Followers

30
Following

72
Likes

75

Ryan Rasti @ryantypes

2 days ago

@nickgeracehacks Haha forgot about Swarm! -- I actually built a whole system on it a long time ago...

1 0 1 13 0

View Details

@JustDeezGuy Yes, and it's very sad to see "the right way" losing in cases like this. There's just this huge network effect to the popular languages that brings in both quality (e.g., TypeScript is unparalleled) and quantity (endless libraries -- every vendor supporting the big guys)

0 0 1 167 0

View Details

Ryan Rasti @ryantypes

3 days ago

@kore_sar Very cool -- I love the callback style to introduce variables! Do types flow through (Typescript) as well?

2 0 1 4 0

View Details

Ryan Rasti @ryantypes

3 days ago

@infinterenders What if we took it a step further and let the client compose the query itself -- full SQL power, executed on the server? (And make it so that nothing bad can happen) No custom RSC per query shape -- client authored & dynamic. (I'm working on this! typegres.com)

0 0 0 5 0

View Details

Ryan Rasti @ryantypes

3 days ago

Thought about this problem a lot -- bits I've converged on: - container/microvm per tool call - rootfs with bind mounts to your working directory and nix store (or even a nixos configuration) - loads your devshell deps upfront -- so each workspace has the tools it needs - network isolation (still have nix cache and can configure) Any feature you think is missing/must have? (I haven't seen a product that has everything here though)

0 0 3 50 1

View Details

Ryan Rasti @ryantypes

3 days ago

@Southclaws @ploeh Yeah, interesting approach -- I actually took it in a round-about way: still implement it in the application layer (TypeScript) but support each dialect fully (all types/functions/operators). I'm genuinely curious, do you have code or a post to learn more?

1 0 0 12 0

View Details

Ryan Rasti @ryantypes

3 days ago

@senro15 @ploeh To give the LLMs immediate feedback if the code is correct -- much faster iterative loop (catch errors in your coding session rather than prod)

1 0 0 12 0

View Details

Ryan Rasti @ryantypes

3 days ago

This is exactly what I've been working on for a year: first-class API and the a capability surface for LLMs Every PG function and operator fully typed, encapsulate tables with classes/methods, then expose the classes directly as your API. Compiles to a single query If interested: typegres.com

0 0 0 123 0

View Details

Ryan Rasti @ryantypes

3 days ago

This list is pretty cool: a nice ergonomic layer for common patterns, so definitely feasible. One cool one that Typegres already happens to share is the method chaining -- e.g., from the playground: `count: orders.id.count()` (if interested: typegres.com/play/)

0 0 0 14 0

View Details

Ryan Rasti @ryantypes

3 days ago

@_rrdelaney @ploeh Curious what DuckDB features you're talking about. Asking because Typegres supports all Postgres builtin functions (and now SQLite too), fully typed. Interesting if we did the same thing for DuckDB too? typegres.com

1 0 1 41 0

View Details

Ryan Rasti @ryantypes

3 months ago

@BuildingWithTom 8s on 500 rows?! Was it doing some roundtrip or something? RLS alternative might interest you: auth is structural, you use typed classes to return what the caller can reach, bound at query construction (i.e., injects the right `where`/`on` clause): typegres.com

0 0 0 15 0

View Details

Ryan Rasti @ryantypes

3 months ago

@StackCurious I built `.live()` into Typegres for exactly this: subscribe to any query. No websocket needed, just a regular HTTP request. Works anywhere you have Postgres typegres.com/play

0 0 1 16 0

View Details

Ryan Rasti @ryantypes

3 months ago

@alexdeliadev @prisma @DrizzleORM Yeah and Drizzle (and every other ORM) doesn't enforce types within query expressions: eq(posts.title, posts.viewCount) (text vs integer) typechecks fine That's what sparked Typegres: every operator and function is typed end-to-end, nullability included. typegres.com/play

0 0 1 103 0

View Details

Ryan Rasti @ryantypes

3 months ago

@heyrapto Yeah — and with many tables, ORMs lack organization. How do you denormalize a relation without breaking clients? My take: the TS class IS the layer. Computed columns + mutations live on the class, all compose to SQL. Schema underneath is refactorable. typegres.com/play

0 0 0 12 0

View Details

Ryan Rasti @ryantypes

3 months ago

Built this because every app I've made ends up rephrasing the same data model four times: schema, ORM, controller, DTO. Typegres collapses that to one class: clients compose typed queries against the methods you expose.

0 0 0 47 0

View Details

Ryan Rasti @ryantypes

3 months ago

Typegres: - Tables as TypeScript classes - Methods are your API: clients compose real SQL (joins, aggregates, subqueries) - Sandboxed via object-capability-based RPC - Live queries built in Try it in your browser 👉 typegres.com/play

1 0 0 40 0

View Details

Ryan Rasti @ryantypes

3 months ago

@hiwinit Thanks! Will do & if let me know if there's anything you'd want to see specifically

0 0 1 1 0

View Details

Ryan Rasti @ryantypes

4 months ago

I'm working on a fifth model, specifically for the agent use-case (a hybrid of 3 & 4): give clients a query builder with full SQL fidelity -- but that can't escape tenancy boundaries. How much would that move the needle on an "untrusted middle layer"? (Link if curious: typegres.com)