Primeta is the persona layer for AI agents. You pick a character — a 3D avatar with a voice, a personality prompt, and its own reactions — and whichever AI tool you already use can speak through them. Claude Code, Claude Desktop, Cursor, OpenClaw, whatever's next. The agent stays the brain; Primeta handles the face.
Who's building this
One person, one small server, nights and weekends. Every feature shipping right now ships because I decided it needed to exist. If something feels rough, it probably is — tell me and I'll fix it.
What we're building toward
A persona layer that's genuinely tool-agnostic. "Who you talk to" should be a separate concern from "which agent or model you're using." A persona shouldn't belong to Claude or to OpenClaw — it should be yours, and it should follow you to whichever tool you pick up next.
Today that works in two directions — MCP sessions for pull-style clients, channels for push-style agents. See MCP sessions vs channels for how they differ. There's a lot more we want to do. The current focus:
- More animations and reactions. Hook-triggered behaviour for test runs, tool calls, deploys — the avatar should visibly react to what the agent is doing.
- More 3D models and a smoother import path. Bring your own character; we'll keep making that easier. See import a VRM.
- Longer-lived personas. Memory, preferences, voices that evolve over time.
- More integrations. Anywhere an agent lives, Primeta should have a way in.
What you can expect
- Honest status. If a feature is rough, we say so in the docs. If something breaks, we say that too.
- Small-team responsiveness. No ticketing queue. Discord or GitHub will usually get you a reply from someone who can actually fix the thing.
- No lock-in on personas. Your personas and conversations are exportable from Settings. You own your character.