@mitchellh Can’t wait to try it out! Happy to beta test it. I am currently using tmux to make Claude code multi user through a web app that connects through a websocket curious If that will apply.
@karrisaarinen@hariiiiiishd I am planning to use the linear agents api to trigger the tasks from linear but then let people manage the tasks from our native interface with the runners
I took sometime to put together a video of the setup we have been working on internally for our AI factory.
Some of the things I think are going to be hard to replicate:
* Networking
* Resizing of sandboxes mid-workstream
* Pre-cache sizes. (we need more than 30GB of drive size) and are reaching for custom drives attachment in those sandbox providers.
I hope this helps
https://t.co/DbbWZpgXQy
Over the past few weeks, I have been building an AI Software factory at 514. It has definitely been significantly increasing my output.
I used to be skeptical about companies rolling their own, and now I am more than convinced.
In a couple of weeks, we have:
* Fully owned and switchable sandbox environments (@e2b and @vercel )
* Support for multiple agents (@claudeai Code and @opencode ) is always on the latest and greatest
* The sandboxed environment is network accessible to engineers on the company network through @Tailscale
* Live Collaboration - Alice can start the agent, Bob can prompt the agent, and steer it, anytime.
* Drag and drop files; they get auto-uploaded and are referenced by the agent.
* 24 / 7 up time - devs don't need to be in front of their laptop for work to move.
* Scoped and secure integration with our internal secrets store, @1Password .
* Pre-packaged skills for everyone in the company to benefit.
* A self-improving system, we have been using those agents to build and improve this system, like russian dolls agents.
Please let me know if you have any questions. I would be happy to answer!
#aifactory #ai #agents #llms
Over the past few weeks, I have been building an AI Software factory at 514. It has definitely been significantly increasing my output.
I used to be skeptical about companies rolling their own, and now I am more than convinced.
In a couple of weeks, we have:
* Fully owned and switchable sandbox environments (@e2b and @vercel )
* Support for multiple agents (@claudeai Code and @opencode ) is always on the latest and greatest
* The sandboxed environment is network accessible to engineers on the company network through @Tailscale
* Live Collaboration - Alice can start the agent, Bob can prompt the agent, and steer it, anytime.
* Drag and drop files; they get auto-uploaded and are referenced by the agent.
* 24 / 7 up time - devs don't need to be in front of their laptop for work to move.
* Scoped and secure integration with our internal secrets store, @1Password .
* Pre-packaged skills for everyone in the company to benefit.
* A self-improving system, we have been using those agents to build and improve this system, like russian dolls agents.
Please let me know if you have any questions. I would be happy to answer!
#aifactory #ai #agents #llms
Just had another run into this. I am building an internal coding platform on top of vercel sandboxes. The platform can run on vercel sanboxes but needs an api key to spin up sandboxes during testing.
It doesn't look like I have a way to create a token that's only scoped to that.
I have to give full production access to my team to the llm that's in yolo mode. Which obviously I don't want to do
Hey @rauchg, from a security perspective being able to have service accounts / scoped access to some apis would be Nice. Current use case: I want a token to access only the cost information for AI gateway and I have to expose all of the Vercel apis with a token for this .
With the Cursor SDK, you can build your own agents with Composer 2.5. It's now available in Python and TypeScript.
This long weekend, Composer usage is 90% off in the SDK. We're excited to see what you build!
I was having dinner with another CTO last night, and I was both surprised and unsurprised to hear that some engineers are still attached to crafting code.
Code can be beautiful, most of the time it is ephemeral, it serves its purpose, and gets replaced. Code that doesn't change is most likely to eventually get thrown away.
What's your take?
I went to the MIT museum this weekend and I loved those 4 essentials.
1. Asking questions others are not willing to ask
2. Trying the unexpected in pursuit of a greater solution
3. A willingness to work hard, even in situations of uncertainty
4. Embracing distinctive skills and a combination of talents
Engineering, when starting a product, is NOT about avoiding technical debt. It is about taking debt and managing it to go fast, make mistakes, keep some, throw out unnecessary things when you can, and iterate.