Use custom models

Connect Vibecape to a Downcity Federation and use local or private text and image models.

Vibecape discovers AI models through a Downcity Federation. You can connect a local or private Federation, sign in, and choose any text or image model published for your account.

Setup flow

Start a Federation Server that publishes your models.
Enter its URL in Vibecape Account settings.
Test the connection and save it.
Sign in to the Federation.
Reload the model catalog and choose a model.

Changing the Federation URL changes the account authority. Sign in again after switching services.

Start a local Federation

The downcity-local-proxy skill can configure a local service for Codex, Claude Code, or both.

Install the skill for your coding agent:

npx skills add https://github.com/wangenius/downcity-local-proxy-skill \
  --skill downcity-local-proxy \
  --agent codex \
  --global \
  --yes

You can then ask the agent:

Create and start a local Downcity Federation Server so Vibecape can use
local Codex or Claude Code models. Use the downcity-local-proxy skill,
create ./downcity/local-proxy, install its dependencies, and start the
model sources I choose. Complete OAuth login if needed. Then give me the
Federation URL and guide me through connecting it in Vibecape.

The generated project is typically started with:

cd ./downcity/local-proxy
pnpm install
pnpm start --codex --claude-code

Use --login for first-time OAuth, for example pnpm start --codex --login. The default local URL is commonly http://127.0.0.1:3000.

How model discovery works

A provider maps a stable public model ID to an upstream model. AIService publishes the catalog, and Vibecape requests the selected model by that public ID.

import { AIService, Federation } from "@downcity/city";

const models = [
  provider.model({
    id: "team-codex",
    name: "Team Codex",
    meta: { upstream_model: "gpt-5-codex" },
  }),
];

const federation = new Federation({ db });
federation.use(new AIService().use(models));