

























The Napster API lets you create, manage, and deploy AI agents with persistent identity, memory, knowledge, real-world actions, and built-in safety — across text, voice, and video.
Define your agent once, deploy it to web, WebSocket, or phone, and it maintains context across every interaction. Full documentation can be found at developers.napster.com. This section covers the key steps to get building quickly.
Every agent you build has five core capabilities:
Define your agent's role, persona, voice, and visual appearance. These persist across every session and surface.
Give your agent domain expertise by ingesting files, FAQs, and structured data. It draws on this to answer questions accurately.
Your agent remembers who it's talking to and what was discussed — across sessions and channels. When a user returns, the conversation picks up where it left off.
Connect your agent to real systems through tools and functions so it can trigger workflows and generate content, not just talk.
Content moderation, crisis detection, topic restrictions, and full session auditability are always active at the platform level.
The API also supports real-time video avatars with lip sync, ~50 languages, and 2,000+ pre-built companion personas.
Once you register, your API key will be waiting in your inbox. Keep your API key private. Never expose it in client-side code, only use it server-side.
All API requests use the X-Api-Key header:
The companion defines your agent's identity, how it looks and behaves. You can browse the catalog of pre-built companions or create a custom one.
Save the companion id from the response — you'll need it in Step 4.
Knowledge gives your agent domain expertise. Tools let it take real actions.
Bring everything together into a single persistent agent:
The id in the response is what you use to deploy the agent to any channel.
The same agent can run in the browser over WebRTC (audio and video), over WebSockets (audio or text), or on a phone number through VoIP or SIP. For WebRTC and WebSocket, you start a session by creating a connection from your backend.:
The Web SDK works with React, Vue, Angular, and vanilla JavaScript. See the full channel guides at developers.napster.com/docs for WebSocket and SIP setup.
Your submission requires two things: a hosted, runnable build (a live URL or deployable repository that judges can interact with directly) and a demo video of up to 3 minutes. Both are required. A few best practices:
These are meant to spark thinking, not constrain it. The best submission will probably be something we didn't put on this list. But if you're staring at a blank page, start here.
Deploy one Omniagent across a VS Code extension (via WebSocket) and a browser-based code review interface (via WebRTC), sharing memory of what the developer is working on. The same agent that answered a question in the IDE already knows the context when the developer opens the review view. The integration with real developer tooling is the differentiator here.
Job interviews, salary negotiations, difficult feedback conversations. The agent plays the other person and uses memory to track how the user’s performance changes across practice sessions. It’s a use case where video presence and persistent memory both do real work.
An agent that joins a live stream or virtual event as a visible participant — answering audience questions via WebRTC in real time, with a knowledge base scoped to that event. Think of it as the layer between a live host and a large audience that the host can’t individually respond to. Production focus: latency, handling simultaneous sessions, and graceful fallback when questions fall outside the knowledge base.
A screenplay, a funding round, a product launch — something with many moving parts over weeks. The agent holds the full context of the project across every session: decisions made, open questions, next steps. Each conversation continues from where the last one ended. The demo should show the agent referencing something specific from a much earlier session unprompted — that’s what makes the memory capability tangible.