Bowei Liu

[Bowei Liu @boweiliu]

holy shit. this is the true nature of cloud agents.

[Daniel Lee @dylayed]

...and @opencode runs on celld!

Lots of monkey-patching to get this working, but really fun to see a glimpse of what's coming:

✅ Fresh OpenCode session per cell (aka Durable Object)
✅ Filesystem backed by SQLite VFS
✅ Bash tool provided by just-bash
✅ Multiplayer experience out of the box with WebSockets

Far from production-ready, but running full coding agents inside serverless V8 isolates feels like the future.

[Bowei Liu @boweiliu]

project of one's own. concept ive had that ive finally found a name for.

i see this is the missing piece for consumers to become builders. especially for students

[Paul Graham @paulg]

In the essay I just wrote about how to prepare students to be founders, I talked about the importance of working on their own projects. In this essay from 2021, I explain in detail how and why to do that.

A Project of One's Own: https://t.co/tk6msuTKE4

[Bowei Liu @boweiliu]

very cool. what if: multiplayer federated cloud ai coding agents on celld.

[Ryan Dahl @rough__sea]

celld v0.3.0 is released. This version contains what (hopefully) will be the last major architecture change: a replicated write behind log.

I want celld to be scalable enough to handle a site like Discord. Every channel would be a cell. At an estimated 4B messages per day, before this release celld would be doing ~46k PUTs per second to S3. That equates to $600k/month in write ops alone.

In 0.3.0, celld doesn’t write to S3 on each message sent. It buffers before sending off to object store. Cutting it to ~300 PUTs per second, ~$4k per month.

But doesn’t that weaken the durability and correctness guarantee? In the new version of celld, writes aren’t ack’d until they’ve been persisted on three disks. And should a crash happen in the short buffer period, it can restore from peers. RPO=0, surviving even the loss of an availability zone. This is the replicated write-behind log.

Aside from the cost savings, writes get much faster: ~5ms instead of a ~50ms round trip to S3.

This latest release also includes support for the D1 database, crons, azure blob service, in addition to many bug fixes.

celld is effectively a database. We take correctness and durability problems seriously. The software is becoming more robust, but do keep in mind this is still a 3 week old project and is labeled as Alpha software. We’re making good progress and hope to raise that to a Beta standing soon. Your bug reports and patches are much appreciated!

[Bowei Liu @boweiliu]

@hudsonmp10 @imbue_ai Hi Hudson! for starters, i like @concentrateai
as an alternative. Longer term, at Imbue we're hoping that more personal and honest software gets created, so incentives shift and companies with consumer data will build more additive features and less anti-features.

[Bowei Liu @boweiliu]

@hudsonmp10 saw you're in the agentic research space - have you come across Imbue's catalyst blog post and open source github repo? https://t.co/EvcTHwNGH4

[Bowei Liu @boweiliu]

@cipri_tom @imbue_ai @cinxwei Saw you're in the vision space @cipri_tom - do you have a feeling for if folks are moving more of their compute to on-prem?

[Bowei Liu @boweiliu]

@cipri_tom @imbue_ai @cinxwei Thanks for the questions! We had already set up most of the ipv6 + firewalls in the past so pmuch just mirrored those configs. and yup our nodes here are our general-purpose set, so good cpu + ram + disk with some older GPU models installed.

[Bowei Liu @boweiliu]

@rough__sea Just had my agent teach me about celld and durable objects via https://t.co/WteEMwvOWt . Does that seem like an accurate intro? Love the project btw, feels super powerful. Curious if you've seen folks building agentic/AI apps on it.

[Bowei Liu @boweiliu]

Just ran across both of these pieces recently -- super hyped to learn that they work well together! @herdrdev have you folks tried out prime agent internally yet? does it work well using an open model instead of anthropic?

[herdr @herdrdev]

Prime Agent 🤝 herdr

They shipped first-class herdr support out of the box before we even knew about it.

Congrats @PrimeIntellect, this is so cool!

[Bowei Liu @boweiliu]

For any devs following -- feels like the real juice is in the dependencies. pi-agent, monano, yjs, capnweb bidir protocol.

I'm still wary of the infra stack (wrangler and workerd are "open source" but... remember terraform anyone?)

[Kenton Varda @KentonVarda]

Today we are releasing Cloudflare OS, a chatbot with connectors, just like every other tech company is doing.

Except actually, it's different. This is a remake of Sandstorm[.]io, my startup from 10 years ago, except this time built on Cloudflare Workers (the platform I've spent the last 9 years building) and deeply leveraging AI. This is more or less the culmination of my secret 10-year master plan.

This is a full-on personal app vibe coding platform, in which the sandbox is so secure that you can pretty much go wild -- the AI cannot introduce a significant security bug. We believe a company's security team can feel comfortable giving non-technical users permission to vibe code and then sleep soundly at night.

How is that possible? It's the Sandstorm security model, revisited. A "Gadget" is the same thing as a Sandstorm "Grain": a fine-grained app instance. For example, if you have a document editor app, each document runs as a separate instance of the app, in a separate sandbox (one "Gadget").

This means two things, both of which I think are Big Deals:
1. The platform can manage all access control, by controlling who can access the Gadget at all. There is no way the Gadget can accidentally leak itself to an attacker -- even an attacker who has access to other Gadgets based on the same app.
2. Since everyone is running their own copy of the code, everyone can freely *modify* their copy of the code.

Think about #2 a bit more.

What if, when you wanted a new feature in the software you are using, you could just prompt your agent to add it?

This doesn't work in the cloud Software-as-a-Service model, because you are not running your own copy of the app.

Sandstorm tried to change that 10 years ago, but the world wasn't ready, because not enough people had the skills or patience to actually modify their software. AI has changed that. Now you just ask the agent -- the same agent that you are using to help you interact with the Gadget can also modify the code of the Gadget.

And it is so fun.

https://t.co/qzwdrAQhtK

[Bowei Liu @boweiliu]

hey @KentonVarda How married is this to cloudflare's infra and cloudflare workers? This sounds like exactly the sort of future @imbue_ai is working on -- personalized workers, loyal and secure agents -- except Imbue's version is NOT tied to big tech clouds.

[Kenton Varda @KentonVarda]

Today we are releasing Cloudflare OS, a chatbot with connectors, just like every other tech company is doing.

Except actually, it's different. This is a remake of Sandstorm[.]io, my startup from 10 years ago, except this time built on Cloudflare Workers (the platform I've spent the last 9 years building) and deeply leveraging AI. This is more or less the culmination of my secret 10-year master plan.

This is a full-on personal app vibe coding platform, in which the sandbox is so secure that you can pretty much go wild -- the AI cannot introduce a significant security bug. We believe a company's security team can feel comfortable giving non-technical users permission to vibe code and then sleep soundly at night.

How is that possible? It's the Sandstorm security model, revisited. A "Gadget" is the same thing as a Sandstorm "Grain": a fine-grained app instance. For example, if you have a document editor app, each document runs as a separate instance of the app, in a separate sandbox (one "Gadget").

This means two things, both of which I think are Big Deals:
1. The platform can manage all access control, by controlling who can access the Gadget at all. There is no way the Gadget can accidentally leak itself to an attacker -- even an attacker who has access to other Gadgets based on the same app.
2. Since everyone is running their own copy of the code, everyone can freely *modify* their copy of the code.

Think about #2 a bit more.

What if, when you wanted a new feature in the software you are using, you could just prompt your agent to add it?

This doesn't work in the cloud Software-as-a-Service model, because you are not running your own copy of the app.

Sandstorm tried to change that 10 years ago, but the world wasn't ready, because not enough people had the skills or patience to actually modify their software. AI has changed that. Now you just ask the agent -- the same agent that you are using to help you interact with the Gadget can also modify the code of the Gadget.

And it is so fun.

https://t.co/qzwdrAQhtK

[Bowei Liu @boweiliu]

Loved the @imbue_ai event tonight!! photos 5-7/7

[Bowei Liu @boweiliu]

Loved the @imbue_ai event tonight!! photos 1-4/7

[Bowei Liu @boweiliu]

@jjacky Thought about this for a while -- my solution was: swap to pi, and edit the file-edit tool behavior to edit the local files first and then wait for the remote file sync to land. Hopefully you're not locked to claude. viva open source and #punksoftware

[Bowei Liu @boweiliu]

Hoping to hear of anyone building the open stack version of this, i.e. NOT on cloudflare -- wasm is deployable anywhere including in browser, insane to tie that to a tech oligotitan #punksoftware

[Miguel Salinas @Vercantez]

https://t.co/pGb77Ubtdp

[Bowei Liu @boweiliu]

@Vercantez Really liked this but why did you choose cloudflare? IMO the most exciting part of this is deploying completely in-browser coding agents.

[Bowei Liu @boweiliu]

https://t.co/BbhjU2n7xh

[Bowei Liu @boweiliu]

I've started using this agent skill whenever I want an AI to write a doc that isn't draining to read. For example I tell the agent "Write a /human-readable plan to do xyz". (credz to n.bannur) 🧵

[Bowei Liu @boweiliu]

Remember glitch (js), anyone? TIL the same folks (Fog Creek) made: Glitch, Trello, Stackoverflow, and joelonsoftware

[Bowei Liu @boweiliu]

Just a reminder that it's against the Anthropic Terms of Service to share your account credentials with anyone else, or otherwise make your account available to others :)

[Bowei Liu @boweiliu]

TIL `claude setup-token` replaces ANTHROPIC_API_KEY but allows access for max plans, and lasts for 1 year.

No more /login on every new fly sprite/EC2 instance/lima VM/podman pod/docker container.

Just: export CLAUDE_CODE_OAUTH_TOKEN=<token> claude and go.

#punksoftware