buremba

Theoretically yes, but CF workers or this project doesn't support it. Indeed none of the cloud providers support WASM as first-party support yet.

I wonder why V8 is considered as superior compared to WASM for sandboxing.

I tried this and it broke the conversation. :(

After Claude Code couldn't find the relevant operation neither in CLI nor the public API, it went through its Chrome integration to open up the app in Chrome.

It grabbed my access tokens from cookies and curl into the app's private API for their UI. What an amazing time to be alive, can't wait for the future!

but at least you get a fraction of CPU and 1GB memory.

Exactly. I have also been playing with DuckDB for streaming use cases, but it feels hacky to issue micro-batching queries on streaming data in short intervals.

DuckDB has everything that streaming engines such as Flink have; it just needs to support managing intermediate aggregate states and scheduling the materialized views itself.

PGlite is embedded so you can't really connect to PGlite using the PG clients. Dbfor.dev runs the PG protocol in the server and use Websockets & PGlite so any PG client can connect to your browser using PG protocol.

The hard work is done by PGlite and we use PGlite, it just enables PGlite to be accessible from everywhere.

PGlite is fantastic. I use it for my in-browser PostgreSQL server for development. It implements the PG protocol on the server; when clients connect, we forward queries to the user's browser, which runs PGlite under the hood.

The result is a PG server that fully lives in your browser: https://dbfor.dev

Just wait for the people to update their LinkedIn titles to TEO expert. :)

So essentially all Claude users are going to surface the "coding agent", making it more suitable even for generic-purpose agents. That makes sense right after their blog post explaining the context bloating for MCPs.

I have been trying a similar idea that takes your MCP configs and runs WASM JavaScript in case you're building a browser-based agent: https://github.com/buremba/1mcp

Oh that’s great! I have been experimenting a similar approach with WASM, I convert MCP tools into Typescript files and expose a single tool to run JS at runtime.

https://github.com/buremba/1mcp

I was more optimistic before bur if 95% of the all software is written by these two languages, it will be very hard for any (better) alternative to disrupt them. The only way will likely to make better profiling & debugging tools to help maintain existing codebase.

It's hard to believe customers are happy with it. I find it shady if you were to try all the cards I added to the product without my permission.

I'm okay with you sending me an email before you try, as long as explicit consent is given. However, if I don't care to change my primary card after the first attempt, that means I don't really care about continuing the subscription.

Happy to be wrong if you prove it with the data, though.

Only if it would work. I think they miss a big opportunity here by (1) not caring about security at all, (2) trying to develop their own model and only make it available in the cloud.

Funny enough, Perplexity blocks the bots themselves. Imagine I develop an "agent" called Merplexity, which simulates an anonymous client browsing on Perplexity and injects my ads into the output without paying for the Sonar API. Would that be OK with Perplexity?

I often need a PostgreSQL server for testing an application, vibe coding, or troubleshooting a PG client. Inspired by Neon's architecture, I implemented PostgreSQL WASM (PGlite) with a cloud proxy (via Websockets) at https://postgresql.dbfor.dev

All the PostgreSQL data lives in your browser, and you have unlimited PostgreSQL servers that persist the data locally without installing anything.

Queued would be more useful as we can also stop the conversation and prompt right away.

This is pretty cool, here are few feedbacks:

- I switch in between planning and execution in the middle of the conversation using Terminal a lot, it would be nice to have it here as well rather than defining how I want Claude to think in advance.

- Entering messages when the agents can result in task lists to save time.

I’m also experimenting different UX with CC, here is Claude Code running in Slack if anybody is interested. https://peerbot.ai

I doubt that there are many inactive subscribers, as Claude Code / Max Plan is relatively new. They might be hoping that way in a couple of months, though.

How do you switch to o3 to review the plan?

I'm okay with free VC money; it's their problem to make it profitable.

They're likely burning money so I can't be pissed off yet, but we see the same Cursor as well; the pricing is not transparent.

I'm paying for Max, and when I use the tooling to calculate the spend returned by the API, I can see it's almost $1k! I have no idea how much quota I have left until the next block. The pricing returned by the API doesn't make any sense.

I'm in the same boat; HA is making a considerable effort, but connectivity is challenging. I was a bit frustrated when I found out that the antennas don't support both Zigbee and Matter simultaneously, despite the claim. You can only support one at a time, so apparently we will need the second antenna.

> The cursor assistant operates the Supabase database with elevated access via the service_role, which bypasses all row-level security (RLS) protections.

This should never happen; it's too risky to expose your production database to the AI agents. Always use read replicas for raw SQL access and expose API endpoints from your production database for write access. We will not be able to reliably solve the prompt injection attacks in the next 1-2 years.

We will likely see more middleware layers between the AI Agents and the production databases that can automate the data replication & security rules. I was just prototyping something for the weekend on https://dbfor.dev/

The clone now has more stars on GitHub than the original work, CheatingDaddy. What's funny is that in a week, most likely nobody will remember that the code was stolen, thus Pickle will probably be fine with their new, shiny, popular project, which will be featured in GitHub Trends.

It's the same with another Soham, who was moonlighting for years. I would not be surprised if he starts a company soon, given the fame he has gained.

Marketing wins.

But “we are family”

I think making the identity piece regional hurts the UX a lot. I like GCP's approach, where you manage multiple regions with a single identity, but I'm not sure how they can make it resilient to regional failures.

The web console in ht appears to be quite helpful for debugging sessions. Great work!

In a world where LLMs can write code fairly well and make use of browsers, I'm not sure if MCP is truly the "USB-C port of AI applications."

The more MCP tools I expose to the LLM, the harder it becomes for the LLM to get the job done. Instead, a single run_python tool works much better and faster. This is especially true for the reasoning models where context matters more.

Amazing, this is what I was trying to find for the last few weeks! I wonder if it's possible to run WASM directly from Python instead of the subprocess approach, though?