Modules
An Elixir framework for building real-time voice AI agents.
Identity, event destination, agreed format, and limits supplied to an adapter.
Explicit pipeline encoding and byte timing for PCM16LE and G.711 μ-law.
An ordered audio fragment. Locally derived timestamps are explicitly labeled.
Contract for application-owned work, independent of the voice provider.
Features supported by the configured backend mapping.
A uniquely identified revision of application-owned work.
Verified backend outcome; cancellation does not imply reversal of side effects.
Runs a fresh Jido AgentServer for each request, observed outside session/audio callbacks.
Optional Legion 0.6.0 execution in isolated Lua agents.
Runs one isolated, bounded Sagents execution per admitted request.
Participant audio and playback controls, independent of device or transport.
Effective channel formats, controls, and sources of playback evidence.
One browser audio connection, terminated at Parlora with ExWebRTC and libopus.
Mountable WebSock signaling for one application-owned session.
CPAL microphone and speaker audio, converted through Rubato to mono PCM16/24 kHz.
Incoming Telnyx bidirectional Media Streaming with raw mono PCMU at 8 kHz.
Bounded incoming-call commands over Mint. The host supplies :api_key.
Mountable signed Telnyx v2 webhook admission. This module starts no listener.
Ed25519 verification of Telnyx v2 webhooks using the original body bytes.
Bidirectional Twilio Media Streams with raw mono μ-law at 8 kHz.
Mountable, signed incoming-call and WebSocket routes. No listener is started.
Twilio HMAC-SHA1 request verification using an explicit public URL.
WebSock transport for an authenticated and authorized Twilio call.
An application proposal validated against the current context revision.
Interprets conversation and application state into backend task decisions.
Immutable context supplied to the application policy, not a persistence format.
The observation that requests context interpretation.
Concise provider context with command and task correlation.
A sanitized failure at a Parlora boundary.
A live observation, independent of business completion or confirmed playback.
Finite per-session bounds. Audio capacities are bytes of the agreed pipeline format.
Continuous full-duplex voice, client delegation and correlated context updates.
Effective voice-provider features; advertise only implemented behavior.
Gemini 3.8 Live with application-owned credentials and backend work.
Grok Voice 3.0 with application-owned backend work over WebSocket.
GPT-Live-1 client delegation over Mint.WebSocket, with explicit application-owned credentials.
Content-free observations in Parlora.status/1 under :health.
Public map returned by Parlora.status/1; this module defines types only.