Napster Omniagent API is a useful new cost signal for teams that want an AI agent to show up as a voice, video avatar, web companion, or future phone agent without building the whole real-time stack themselves. For Makefun readers, the key question is not whether another agent platform exists. It is whether a video-enabled agent can be budgeted cleanly enough for creator support, product demos, avatar-led onboarding, and AI video workflow handoffs.
The short answer: Napster’s public developer page and Microsoft Marketplace listing make the pricing unusually concrete. The BYO LLM path starts around $0.01 per active minute, while the Napster Hosted tier is listed around $0.058 per active minute with a managed LLM included. That makes it worth comparing against per-minute voice-agent APIs and audio-token realtime models before a team commits to a video agent design.
What Napster Omniagent API includes
The official Napster developer page positions Omniagent API as a multimodal agent layer with video, audio, telephony, memory, knowledge, tools, WebRTC, WebSockets, REST APIs, safety, and monitoring. The developer docs describe an Omniagent as one persistent agent that can keep identity, voice, knowledge, and memory across channels.
The Microsoft Marketplace listing for Napster Companion API adds an important deployment detail: teams can connect their own OpenAI real-time deployment on Microsoft Foundry, or use a Napster Hosted tier with the LLM included. That split is the core budget decision.
Pricing comparison for video and voice agents
| Workflow | Public price anchor | Best fit | Watch-outs |
|---|---|---|---|
| Napster Omniagent API, BYO LLM | About $0.01/min for the platform layer; LLM billed separately | Azure/OpenAI teams that already control realtime model deployment | Azure inference, token usage, tool calls, storage, and monitoring are separate from the platform minute |
| Napster Hosted | About $0.058/min with a managed LLM included | Teams that want one line item for video/audio agent sessions | Hosted convenience can hide model mix, session quality limits, and enterprise terms |
| Deepgram Voice Agent API | Deepgram pricing lists Voice Agent API Standard at $0.075/min | Voice-first support agents and phone-style assistants | Video avatar, visual identity, storage, telephony, and downstream LLM/TTS choices still matter |
| GPT-Realtime-style voice workflow | Audio-token pricing depends on input, cached input, and output tokens | High-control custom realtime agents with strong model reasoning | Long output audio, retries, translation, priority modes, and tool loops can move cost quickly |
Cost calculator template
Use this lightweight calculator before choosing a stack:
- Monthly active minutes: sessions per day x average session length x active days.
- Platform cost: active minutes x Napster BYO or Hosted minute rate.
- Model cost: add Azure/OpenAI realtime usage for BYO deployments, or validate what the Hosted tier includes.
- Failure budget: multiply by expected retries, dropped sessions, QA runs, and rejected avatar interactions.
- Operations budget: add memory storage, transcripts, knowledge refreshes, moderation review, telephony, and human handoff time.
Example: 100 five-minute sessions per day is 500 active minutes. At $0.01/min, the platform line is about $5/day before LLM and operations. At $0.058/min hosted, the same usage is about $29/day before any extra enterprise or workflow costs. That difference is large enough to matter, but not large enough to ignore implementation risk, latency, and quality.
Where this fits in a Makefun workflow
Napster is closest to the customer-facing agent layer: a companion that can speak, appear on screen, remember a user, and trigger tools. Makefun planning often starts one step earlier or later: creating avatar media, generating video assets, testing an AI video API, or deciding whether an agentic video generator should produce a finished clip or guide a user interactively.
A practical path is to prototype the agent’s script and handoff logic first, then decide whether each moment should be live video, voice-only, generated video, or a human review queue. For voice-heavy support flows, compare against the GPT-Realtime-2 voice agent API cost model. For demos, onboarding, and creator sales pages, video presence may justify the extra implementation work.
Decision checklist
- Use BYO LLM if your team already has Azure realtime infrastructure, compliance requirements, and model cost observability.
- Use hosted pricing for faster pilots where predictable per-minute budgeting matters more than model-level control.
- Avoid overbuilding live video for tasks that a generated clip, static avatar, or voice-only assistant can handle.
- Track retry rate, average session length, handoff rate, and unresolved questions before scaling traffic.
- Keep private data, model provider terms, transcript retention, and user consent in the launch checklist.
The SEO opportunity is the same as the product opportunity: video agents are moving from demos into priced, deployable APIs. A useful buying guide has to show the minute cost, the hidden model cost, and the workflow choice instead of treating every embodied agent as the same product.
