Building Deputies, an open source background-agent control plane
Joined November 2009
- Tweets6K
- Following1.1K
- Followers13.3K
- Likes6.3K
Pinned Tweet
I’m excited to share Deputies, an open source control plane for background coding agents!
The goal is simple: move agentic software development off individual laptops and into a central hub where teams can see what agents are doing, review their outputs, and keep a traceable history of the work.
Inspired by systems like Inspect/Open-Inspect, Minions, and Open SWE, I built Deputies on top of @FredKSchott's Flue Framework.
It’s designed to run in your own infrastructure and trust boundaries, with Slack/GitHub integrations and @daytonaio + Docker sandbox support.
.@AmpCode it would be awesome if on a multiplayer thread (that someone else created) I could explicitly tell the agent to use my git identity.
Sometimes we end up handing off a thread to someone else after creating and it is awkward to have the code ship as if I was authored by the thread creator if someone else was steering.
If your agent cant successfully perform:
> "Run the application and run a user acceptance test that exercises at least one user flow end-to-end"
Stop whatever you are doing and fix that. Like right now.
It will be the highest leverage thing you do all week
I'm joining @TrelisResearch to build the future of AI voice infrastructure!
The goal is to provide the best way to add a robust voice interface to applications. Natural turn-taking, tool calling, professional quality voices, all from a single API
See video for demo:
*This excludes money from consulting work related to setting up and using background agents, that I would not have otherwise gotten without building and talking about @DevDeputies online
If a claude code remote control process dies on a remote server (e.g. is OOMkilled) is there really no way to recover/continue any associated sessions?
I can still see the transcript and any modified files on disk remain, but restarting the cc process produces a separate env 🤔
I'm going to train a model to insert strategic tipos and, grammatical errors to throw off people's slop detection
I have seen a few posts asking for macOS runners from @AmpCode... so I decided to try something:
1. Create a token for @namespacelabs
2. Spawn and Amp thread with the following prompt
3. ???
This also lays the foundation to let people configure most things from within Deputies (instead of primarily through environment variables).
Some teams will prefer to codify everything with env vars and bring their own secret stores, but having the ability to do a 1 click deploy and then plug in the various auth tokens/API keys in-app should reduce the friction for many!
Here is the template used in the video: railway.com/deploy/deputies-…
And before you @ me... yes I collected the API keys ahead of time.
But even if you needed to provision the necesary credentials that is still (hopefully) on the order of 10 minutes 😀
Here is the railway template if you want to deploy your own instance railway.com/deploy/deputies-…
or check the repo for guidance on deploying elsewhere! (e.g. there is Terraform for a fully @awscloud deployment)