Skip to main content
r5d.chat is a self-hosted chat application for AI agents that actually do work. Each user chats with an OpenAI-compatible model while the agent operates inside a dedicated Kubernetes pod: it reads and writes files, runs shell commands, searches documents, and installs tools. The pod scales to zero when idle and keeps its home directory on a retained volume, so the next conversation resumes with the same files.

Quickstart

Install the chart, claim the administrator account, and run a first agent task.

Core concepts

Control plane, workspaces, providers, and the idle lifecycle.

Install with Helm

Install from the Helm repository or directly from the OCI registry.

Configuration

Environment variables, model providers, SSO, branding, and MCP.

What you get

Every user maps to one deterministic Deployment, Service, identity Secret, and retained ReadWriteOnce PVC. The Deployment scales between zero and one and the complete PVC mounts at /home/r5d. The browser never receives workspace credentials.
Users configure any OpenAI-compatible endpoint with their own API key, custom headers, model IDs, context windows, and output limits. Keys and headers are encrypted at rest with the installation key.
The workspace image ships Bun, Node.js, Python with the scientific stack, native build tools, FFmpeg, ImageMagick, libvips, OCR, LibreOffice, Pandoc, Recoll full-text search, and Git/GitHub tooling.
Administrators set the product name, tagline, logo, favicon, and primary colors at runtime. No rebuild required.
Conversations are stored in PostgreSQL with explicit fresh-context boundaries. The agent can re-read its own history through the r5dchat CLI inside the workspace.

Install in one command

Add the Helm repository and install into a dedicated namespace:
Both the chart and its container images are published to GHCR on every release tag. A chart version always resolves to the images built from the same commit. See Releases and publishing.

Where to go next

Workspace lifecycle

How workspaces start, lease activity, and scale to zero.

Security model

Trust boundaries, isolation guarantees, and their limits.

Agent tools

Every tool the model can call, with its parameters.

Upgrades

Rolling out new versions and existing user workspaces.