GATEWAY · IDENTITY · PARADIGM
ai.kungen.pro
API wires
OpenAI chat and responses, plus Anthropic messages
Paradigm
Delivery engine on the gateway after a model miss
providers
ollama, opencode, qwen, azure, kimi, xai, openrouter
$Signing in leaves this page for Chat. The other rooms follow role and plan. The gateway behind them is ai.kungen.pro.
SSE chat for every signed-in account. The gateway prepends the Kungen system prompt on the way to the provider.
Canvas for image, video, text, and stitch, plus a durable queue. A run with a source photo keeps that person. Admin on the website.
Reactive companion. It sends X-Kungen-No-Identity so the persona stays Dream. Separate from the standalone Telegram companion service.
Remote OpenCode on the runner. Visible on Pro and above.
Voice calls through the external Twilio suite. Admin only. Twilio is not a service inside this repo.
Native Compose app, sideloaded. Chat for everyone. Studio PRO, Dream, and Calls for admins. Console on Pro+.
Admin holds files, memory, sessions, learning, and diagnostics. Open Memory from Admin rather than a separate product nav item.
curl install.sh, or irm install.ps1 on Windows. Launches the coding tools against ai.kungen.pro. Does not start the gateway.
Notes from the website nav, the remote CLI, and Studio PRO. Not customer quotes.
“A signed-in session leaves this page for Chat. Studio PRO, Dream, and Calls are admin. Console is Pro and above. Memory is a tab inside Admin.”
kungen.pro
Website rooms“The remote installer writes a launcher that calls ai.kungen.pro. It can start Claude Code, Codex, Aider, Continue, or native OpenCode. It does not run kungen serve.”
Remote CLI
install.sh · install.ps1“Studio PRO is the media workspace: image, video, text, and stitch on a canvas, with a durable queue. A run that includes a source photo keeps that person. The Android drawer has the same admin gate.”
Studio PRO
Web and Androidinstall.sh and install.ps1 install the remote launcher. It talks to the hosted gateway and can open Claude Code, Codex, Aider, Continue, or the OpenCode reskin. kungen serve starts FastAPI from the repo; the remote CLI does not have that command.
$ curl -fsSL https://ai.kungen.pro/install.sh | bashUnix uses bash. Windows uses the native PowerShell installer. Do not pipe install.sh into iex. Starting the gateway is a different program: kungen serve in the repo.
The FastAPI process at ai.kungen.pro injects identity, routes aliases, and runs Paradigm. Sith stays on the Railway host.
KUNGEN_SYSTEM_PROMPT is required at startup. Chat requests get that system prompt unless the client sends X-Kungen-No-Identity, which also skips Paradigm.
Paradigm is the delivery engine in kungen/paradigm, configured from config.yaml. Chat, stream, and agents share it. A miss recovers on the gateway before the turn is dropped. Dream opts out.
POST /v1/chat/completions and POST /v1/responses on the OpenAI wire. POST /v1/messages on the Anthropic wire. claude and gpt alias to opencode/qwen3.7-max. grok aliases to xai/grok-4.7.
POST /v1/agent/run returns JSON. POST /v1/agent/stream emits tool progress as SSE. Short greeting turns drop tools so a hello does not open the loop.
Sith is the tools island beside the gateway in the Railway project. The perimeter lives on that host. This website does not run those tools.
The router builds ollama, opencode, qwen, azure, kimi, xai, and openrouter from config.yaml. Other provider classes stay registered in code until they are listed there.
The gateway speaks OpenAI and Anthropic. Aliases in config.yaml choose the model. These are the live clients and the seven configured providers.
Claude Code
Anthropic base URL on the gateway. Alias claude routes to opencode/qwen3.7-max.Codex
OpenAI-compatible base URL. The remote CLI launches it after you pick a model.Aider
Points at the same gateway key. Remote command: kungen aider.Continue
Editor config printed by the CLI. Same identity injection as chat.OpenCode native
kungen native is OpenCode reskinned in Canary Matte, with its own config directory.Android
Sideload APK from /app. Not a Play Store listing. Wake word on device.Ollama
Ollama Cloud is a live provider in config.yaml. Local Ollama is the container path.OpenCode
Most legacy aliases land here, including claude, gpt, and hf.Qwen
DashScope provider. Pass qwen/… directly or use an alias that targets it.Azure
AI Foundry catalog. Only deployed models answer, and quota is per model.Kimi
Kimi Code Plan at api.kimi.com. Not the Moonshot platform API.xAI
Grok at api.x.ai. The grok alias resolves to xai/grok-4.7.OpenRouter
Ids keep the vendor prefix. The models block is a whitelist.The website is the rooms. ai.kungen.pro is the FastAPI gateway. The remote CLI installs with the commands above and does not start kungen serve.
Create Account