DPC progress.
Amazing stuff deployed tonight. Devplace code. Claude max experience in your browser. You could go from not having a devplace account to having created AND host a guest book for example.
Devplace Code is in development for several months and vs code integration for two weeks. I have dog fed the project well enough.
It all starts with creating a (private optional) project and click ok the workspaces menu item on the project detail page. More explwinwition is not needed. A blazing fast vs code (this server is five times more powerful than the machine I work on) will start the code agent instant.
The agent is able to do vision, deep research and swarming. 1M (cached) context. The deep research functionallity did beat claude code on literally every category. The deep research backend is developed for over a year by now. It's a personal fetisj and you can go pretty deep into the subject.
DPC does support all the stuff you're used too from claude. Local claude md, global claude md, agents, rules, workflows and skills. DPC's rendering is objectively superior in rendering.
Not that it matters since it's hosted online - but the agent is ten times lighter to use than claude. It's written in NIM what transpiler to C making it a zero dependency binary (if openssl was less anal).
Smart context - source code is indexed in a hybrid of BM25 and vector. It ain't cheap but executed in the most economical way possible. The vector database is a custom design.
The concurrency of DPC is mayhem. Getting bored about claude being a snail and beyond? You won't be bored this system develop software. It goes fast, try to keep up. It's not weird that it opens 5 panes of live streaming terminals at once.
The goal with this platform is giving people a claude max experience ik the hope people will ditch their pro accounts, those should never be taken seriously.
Comments
Since @Lensflare heard about it, he has not been heard from again. All that resource, he can't stop.
@Lensflare flarii? ๐
Flarii is summoned by snek and followed him into the rabbit hole.
@retoor, you say the deep research backend beat Claude Code on every category, but what exact evaluation suite and metrics did you run to make that claim hold up?
My own.