Why multiple models
Different families excel at different tasks: some are extremely fast for brainstorming, others are stronger at long-context reasoning or multilingual answers. WebVoice lets you pick from the catalogue enabled by your administrator — typically including high-throughput hosts (such as Groq-accelerated stacks), frontier assistants from providers like Moonshot (Kimi), MiniMax, and other ChatAIModel entries configured in the control plane.
You can start a thread with one model, open another thread with a different model, or change the default for new conversations so teams can standardise on a safeguard-oriented profile for customer-facing replies and a lighter model for internal drafts.
Providers and defaults
Each row in the model list shows display name, provider, and credits per request where applicable. Groq-backed options often cost fewer credits per turn while maintaining low latency; other providers may charge more per message but add capabilities (longer memory windows, specific tool formats, etc.). Superusers curate which model IDs are visible so obsolete endpoints disappear without client updates.
Optional chat tones and agent personas (when enabled) layer system instructions on top of the base model, so the same backbone can sound formal, concise, or educational without swapping weights.
Together with voice
Chat is not isolated: you can move from transcription to summarisation, then send polished text to TTS, or attach voice memos alongside written prompts. Credits for chat, TTS, and STT share one wallet, which simplifies budgeting for mixed-media projects.
Summary
- Selectable models from an admin-curated catalogue (Groq, Moonshot, MiniMax, …)
- Per-request credits shown before you send
- Multiple parallel threads with different models
- Works alongside TTS/STT and API-based automation