Bring your own LLM and voice
The customer operates ASR, LLM, TTS, retrieval, tools, policy, and application state. The avatar product receives the final speech output and presents it visually.
Architecture comparison
BYO LLM avatar API vs bundled conversational avatar
Choose BYO LLM when model behavior, tools, knowledge, and orchestration are already strategic product assets. Choose a bundled conversational avatar when reducing integration work matters more than controlling each AI component.
Verified Aug 3, 2026Decision-ready guidePrimary sources
Decision matrix
Compare the complete path.
Do not compare isolated numbers unless the definitions, inputs, environment, and included services match.
| Decision area | Bring your own LLM and voice | Bundled conversational avatar |
|---|---|---|
| Model ownership | Customer chooses and operates models | Platform-supported models and settings |
| Tools and knowledge | Customer-defined | Vendor framework or integrations |
| Latency ownership | Shared across customer services | More of the path is vendor-managed |
| Voice choice | Customer TTS | Bundled or supported voices |
| Migration | Avatar layer can be changed independently | Components may be coupled |
| Implementation | More integration work | Faster initial assembly |
| Best fit | Differentiated AI product | Managed conversational experience |
How it works
Two different operating models.
Architecture decides which team owns rendering, transport, recovery, and the surrounding AI product.
Bundled conversational avatar
The vendor connects conversation, voice, and avatar components into a managed experience. Configuration is simpler, while model and data choices follow the platform boundary.
Best fit
Choose for the system you can operate.
The best option is the one whose responsibilities match your product, client, network, and team.
Choose Bring your own LLM and voice when…
- Prompt and tool logic are proprietary
- Need a specific LLM or TTS vendor
- Already operate observability
- Want independent component replacement
Choose Bundled conversational avatar when…
- Need a fast managed prototype
- Supported models meet requirements
- One support owner is preferred
- Team does not want to run the voice stack
Limitations and unknowns
BYO architecture moves more operational responsibility to the customer. Bundled architecture can limit optimization or migration options. Evaluate both initial speed and year-two flexibility.
Build a controlled evaluation.
Use one workload and record both user experience and operational responsibility.
1. Evaluation stepList mandatory model and voice providers.
2. Evaluation stepTrace every data boundary and retention setting.
3. Evaluation stepBuild a latency budget per component.
4. Evaluation stepRun a replacement exercise for one bundled dependency.
- Define the user job and acceptable fallback.
- Use the same input, session duration, and client.
- Record latency, traffic, compute, errors, and recovery.
- Compare total operating cost, not only list price.
Evidence
Sources and freshness.
Last verified Aug 3, 2026. Recheck implementation details when SDKs or plan terms change.
Related decisions