LLM Meta Chat

A Rails chat application that runs on top of the LLM Meta infrastructure. It embeds the llm_meta_client Rails Engine and talks over REST to a running llm_meta_server, which in turn fans out to OpenAI, Anthropic, Google, or a local Ollama.

┌─────────────────┐    REST + Google ID token    ┌──────────────────┐    Provider SDKs    ┌────────────────────────┐
│  llm_meta_chat  │ ───────────────────────────▶ │  llm_meta_server │ ──────────────────▶ │  OpenAI / Anthropic /  │
│  (this app)     │                              │                  │                     │  Google / Ollama       │
└─────────────────┘                              └──────────────────┘                     └────────────────────────┘

This app never talks to a provider API directly — it always goes through llm_meta_server.

System Requirements

• Ruby 3.4.9 (see .ruby-version)
• Rails 8.1.2
• PostgreSQL
• Node.js (for asset compilation)
• A running instance of llm_meta_server — this app cannot function without a reachable backend. Set it up first.

Installation Steps

1. Clone the repository

   git clone https://github.com/jdkim/llm_meta_chat.git
   cd llm_meta_chat

2. Install dependencies

   bundle install

3. Configure Rails credentials

   This app uses Rails credentials (not environment variables) for secrets:

   EDITOR="vim" bin/rails credentials:edit

   Add the following keys:

   google:
     client_id: <Google OAuth 2.0 client ID>
     client_secret: <Google OAuth 2.0 client secret>
   
   llm_service:
     base_url: http://localhost:3000        # URL of your running llm_meta_server
     summarize_conversation_count: 10        # optional; default 10

   Key	Purpose	Default
   google.client_id	Google Sign-In for user auth (Devise + OmniAuth)	Required
   google.client_secret	Google Sign-In for user auth	Required
   llm_service.base_url	Base URL of your llm_meta_server instance	http://localhost:3000
   llm_service.summarize_conversation_count	How many recent turns to include when summarizing context	10

   Google OAuth 2.0 setup. If you don't already have OAuth credentials, follow steps 1–5 of the server's Google OAuth setup guide, then add the following authorized redirect URIs to the same OAuth client:

   http://localhost:3001/users/auth/google_oauth2/callback     # dev
   https://<your-chat-host>/users/auth/google_oauth2/callback  # production
   

   In addition, this app's google.client_id must appear in the server's ALLOWED_GOOGLE_CLIENT_IDS env var so the server accepts API requests carrying tokens issued to this client.

4. Set up the database

   bin/rails db:setup

   Creates the database, runs migrations, and loads any seed data. Requires a running PostgreSQL server; connection details are configured in config/database.yml.

5. Start the application

   The Rails default port is 3000, but llm_meta_server also runs on 3000. Start this app on a different port — 3001 is used throughout the docs above:

   bin/rails server -p 3001
   # or, for the plain wrapper:
   PORT=3001 bin/dev

   The chat UI is now available at http://localhost:3001.

Verifying Installation

1. Backend reachable. With llm_meta_server running on port 3000:

   curl -sI http://localhost:3000/up
   # → HTTP/1.1 200 OK

2. Chat app reachable. Open http://localhost:3001 — the chat UI should render with an LLM selector in the header.

3. Anonymous Ollama chat (no Google Sign-In required, assumes an Ollama model is registered on the backend). Select "Ollama Local" in the LLM selector, type a short prompt, and send. A response should stream back.

4. Signed-in chat. Click "Sign in with Google". After the OAuth round-trip you should land on the chat page as a signed-in user; any API key you have registered on llm_meta_server will appear in the selector.

If step 2 fails, run RAILS_LOG_LEVEL=debug bin/rails server -p 3001 and re-check. If step 3 fails but the UI loads, the llm_service.base_url credential is likely wrong, or the backend has no Ollama model registered.

Testing

bin/rails test              # unit + integration
bin/rails test:system       # system tests (browser)
bin/ci                      # full CI pipeline: rubocop, audits, brakeman, tests, system tests

Citation

If you use this software in your research, please cite:

Kim, J.-D. (2026). AIbranch: A platform for branched multi-model LLM conversations. SoftwareX. https://doi.org/10.1016/j.softx.2026.102983

Related repositories

• llm_meta_server — the backend that this app talks to (install this first)
• llm_meta_client — the Rails Engine gem that provides the chat scaffold used here