devchitchat is a self-hosted platform that puts AI coding agents inside your team's chat, your CI/CD pipeline, and your Git workflow. Everyone works in the same room. Nobody waits on a ticket.
Teams under delivery pressure don't need another project management tool. They need a shared room where work is visible, AI is a first-class collaborator, and shipping is the default.
Five integrated pieces that make multi-player AI development feel natural.
Real-time messaging with WebSocket streaming, video/audio/screen-share, file uploads, and push notifications. Runs on your hardware — no SaaS dependency, no data leaving your network.
Claude Code runs as a Kubernetes pod with direct filesystem and Git access. Send it a message in chat, watch it commit code, open PRs, and stream responses back — all without leaving your workflow.
Track work directly on the mesh network. Issues live alongside your code and CI events — no separate tracker, no context switching. Your agent can read and create issues as part of any task.
Two-tier GitOps controllers — one for infrastructure, one for apps — poll your Git repos every 30 seconds and apply changes. No webhooks. No secrets in pipelines. No SSH keys.
All nodes — host, CI runner, controllers, agent — connect over a decentralized mesh. Git repos are served peer-to-peer at port 7979. Invite once, persist forever in PVCs.
Canonical command IDs, declarative schemas, confirmation flows for side-effecting operations, ordered async middleware, and audit records. Build bots that are safe by default — CLI adapter included.
index97 brings file-based routing, server-side templates, static site generation, built-in CSP, and HMR dev mode to Bun. Composable — run your website and chat on the same port. Published to npm.
From a Mac mini boot to a deployed app — everything automated.
A LaunchDaemon starts Lima (QEMU, Ubuntu 24.04) at system boot — no login required. k3s comes up automatically. One command provisions the whole cluster.
infra cluster setupBuild and push the CI runner and GitOps controller images, deploy the in-cluster registry, run the mesh invite/join dance, and push the GitOps repo into the mesh network.
infra cluster bootstrapThe mesh-gitops-controller polls your platform repo every 30 seconds. The apps-controller polls your app repo. Commit a manifest change and it's live within half a minute — no webhooks, no external CI service.
git push origin main
# → applied in <30sThe Claude Code agent pod runs in your cluster with persistent workspace storage and full Git access over the mesh. Send it a task in chat, get code back. Your team sees everything in the same thread.
@agent refactor the auth module
and open a PRBoring technology choices. Interesting architecture.
You don't need another Jira board. You need a room where work is visible, AI can pull its weight, and shipping is the default activity.
You have a Mac mini and you'd rather run your own stack than pay per seat for SaaS tools you don't fully control.
You want AI agents as first-class team members — not a chat widget bolted onto your editor, but a collaborator with real Git access and persistent context.
The chat, ChatOps, and web framework packages are published to npm. Build on top of them, extend them, or run them standalone.
Everything you need to run a self-hosted, AI-native dev platform on your own hardware.