Chat with your personas directly — no MCP client needed

Until this week, talking to a Primeta persona required a Model Context Protocol client in the middle — Claude Code, Claude Desktop, Cursor, or the OpenClaw gateway — to route your message from a local process into our server and back. Great if you already lived in one of those tools. A meaningful setup step if you didnt.

Now you can skip it. Bring your own LLM API key, and chat with a persona directly from the browser.

How it works

In Settings, theres a new "LLM Connections" section. Add a connection — name it, pick a provider, pick a model. Supported providers today:

• OpenAI (GPT-4, GPT-5, reasoning models — whatevers live on your key)
• Anthropic (Claude 4 and later)
• xAI (Grok)
• Ollama Cloud (open-source models running on Ollamas hosted infra)

Once the connection exists, it appears as a card on your dashboard alongside any MCP sessions or bridge channels you already have. Click it, start a conversation, and your persona responds using your key on your provider. Voice, face, emotion, animations — same 3D avatar experience as with an MCP client, just without the client.

What fell out of the work

A handful of smaller improvements came along for the ride:

• Model dropdown is always current. Rather than a hand-curated list that drifts, the dropdown pulls from each providers live catalog via ruby_llm and filters to chat-capable models. You see exactly whats available on your key today.
• The setup form reveals itself progressively. Name your connection, pick a provider, pick a model. One step at a time so the form doesnt look like a wall on first visit.
• Persona switches greet you mid-conversation. If you change characters during a chat, the new persona speaks up immediately with a short in-character greeting, rather than sitting silent until your next message.
• Composer always shows. The message input was previously hidden on MCP-only conversations; that routing now correctly reveals the composer on direct-LLM connections.
• Connection records split from provider-key records. One key can back multiple connections — useful if you want separate named conversations on the same account.

Why this matters

The MCP flow stays the first-class experience if you already write code with Claude Code or a similar agent — your coding session IS the chat, and the avatar reacts to everything your agent does. Thats still the deepest integration.

But MCP clients are developer tools. Direct-LLM chat opens Primeta to anyone whos paying for an LLM subscription and wants to talk to a 3D character version of that LLM without installing anything. The avatar, the voice, the personality prompts, the hook animations — all of it available with a key and a dropdown.

It also removes a friction point for existing users: if youre on a non-dev machine without Claude Code set up, or you just want a quick conversation without switching contexts, direct chat is there.

Try it

1. Grab an API key from OpenAI, Anthropic, xAI, or Ollama Cloud.
2. Open Settings → LLM Connections and click "Add connection".
3. Name it, pick your provider, pick a model.
4. New connection shows on your dashboard. Click through to chat.

Connection keys are encrypted at rest — we store them via Rails encrypted attributes, not in plaintext. You can remove a key at any time from the same settings panel.