The builds
Build your own Gong
Conversation intelligence: transcribe calls, read Recap talk ratio and signals, summarize, and score eligible calls. Uses Hear, Recap, and Trace.
Build your own Wispr Flow
A voice dictation app: push-to-talk mic capture, streaming partials rendered at the cursor, finals committed as you speak. Uses Hear streaming.
Build an AI receptionist
24/7 phone answering: greets, answers from your knowledge base, books appointments, transfers to a human. Uses Omni, telephony, and tools.
Build call-center QA
Score eligible calls against compliance rule packs instead of manually sampling a few. Uses Trace.
More you can build on the same APIs
- A meeting notetaker. Record or stream the meeting into Hear, let Recap write the summary and action items when it ends. Same pipeline as the conversation-intelligence build, pointed at meetings instead of sales calls.
- An outbound dialer that knows when a human answers. AMD tells you human vs. voicemail in a fraction of the dead-air dwell, then Omni takes the live ones. See the AMD guide.
- A voiceover studio. Speak turns scripts into natural speech with stock or
designed voices; Clone enrolls a consented custom
voice; Cast handles multi-line, expressive renders under
/v1/cast/*. - An IVR replacement. Omni answers with speech, not a phone tree, and its
built-in call control (
transfer_to_human,collect,send_dtmf) covers the routing. See the Twilio guide.
Where to start
1
Get an instant sandbox key
No signup, email, or card:
POST /v1/sandbox/keys returns a working
pyai_test_ key in one curl. The quickstart has the
copy-paste version.2
Pick a build above
Each guide lists the exact scopes your key needs and the order to build in.
3
Go live on a funded key
Create a
pyai_live_ key in the console when
you’re ready for production traffic.