Configure your AI models and voice before your first browser or phone conversation. Your workspace model settings provide the defaults for agents; you can also set per-agent overrides in agent settings.
Open Model Configurations#
Open Models in the workspace sidebar, or choose Configure models on the overview page. Both open Model Configurations at /model-configurations.
- Speech to Speech: a realtime speech-to-speech model handles the live conversation; you still configure an LLM for extraction, QA, and other non-realtime tasks.
- Menace Voice: one Menace Voice Service Key selects Menace Voice-managed LLM, voice, and transcriber models. Usage is billed through Menace Voice credits on the account that owns the Service Key.
- BYOK: bring your own provider accounts and API keys for LLM, voice, transcriber, and embedding models. You pay those providers directly for inference usage.
Menace Voice-managed inference#
When managed services are available, sign-in may configure a default Service Key and managed models for your workspace. If no configuration appears, open Model Configurations and add a Service Key or your own provider keys.
Managed usage costs
Managed inference uses Menace Voice credits on the account that owns the Service Key. Check your balance and available pricing in the workspace before testing.
Service Keys vs workspace API keys#
- Service Keys authenticate Menace Voice-managed inference in Model Configurations. Create them from Developers → API Keys, choose Service Key, and paste the value into the Menace Voice section. They are not interchangeable with workspace API keys.
- Workspace API keys authenticate your organization to the Menace Voice API and MCP endpoint — for example triggering agents, listing workflows, or connecting Codex or Claude to /api/v1/mcp/. They do not replace a Service Key in model settings.
Bring your own keys#
BYOK is available on every plan, including Free. Supply valid provider credentials for the model categories you enable. Your selected providers bill inference usage separately from the Menace Voice platform plan.
Browser testing vs hosted telephony#
- Free includes WebRTC browser testing (Web Call), one voice agent, and BYOK. It does not enable hosted telephony or outbound campaigns.
- Starter and Pro enable hosted telephony (inbound and outbound). Pro also enables outbound campaigns. Telephony still requires a configured provider or SIP route, an active phone number or trunk where applicable, and provider-side charges separate from Menace Voice credits or subscription fees.
Separate cost layers
Budget for the platform subscription, your chosen inference service (managed credits or BYOK provider usage), and any carrier or SIP charges. Start with a browser test before connecting a phone line.
With models configured, continue to Build your first voice agent and run your first Web Call before adding telephony.