"Push-to-talk AI voice companion for Discord channels, powered by your own machine."
Real-Time AI Voice Chat
For Discord Communities
Oracle relays 48 kHz Opus voice turns between your Discord channel and your desktop companion app. Speech-to-Text, AI inference, and Kokoro TTS run on your machine inside an isolated sandbox.
How Oracle Works
Push-to-talk voice conversation in three steps.
Push-to-Talk in Discord
Call /join and hold your push-to-talk key to speak. Oracle only relays audio while your key is held—never ambiently.
Sandboxed Local Processing
Your paired companion client transcribes speech in memory and queries a sandboxed local AGY worker or your local Groq key with zero host file access.
Direct Opus Voice Reply
Synthesized audio is encoded to 48 kHz Opus on your client and played directly into the Discord voice channel.
Privacy & Host Isolation
Built so voice chat never exposes your files, credentials, or local network.
Dedicated Windows Worker Sandbox
Oracle runs under a dedicated low-privilege local user account (OracleWorker) with OS ACLs denying access to user profiles, documents, and credentials.
Firewall Local Network Shield
Outbound Windows Firewall rules block RFC1918 private LAN ranges, link-local addresses, and localhost loopbacks from the worker user.
Live Session Canary Verification
Every session verifies isolation against a host canary file before answering. If the canary can be read, the client refuses to process queries.
Memory-Only Audio Processing
Voice audio is processed in memory and discarded after each turn. No voice audio is written to disk on the client or server.
Core Capabilities
Direct Opus relay, outbound client pairing, and live persona switching.
Push-to-Talk Only
Oracle enforces push-to-talk while in a channel and restores exact channel permissions on teardown and startup recovery.
Outbound Client Bridge
Your companion app connects outbound over TLS WebSocket using an Argon2id-hashed pairing token. No port forwarding or inbound listeners.
Live Persona Switching
Create custom personas and switch active personalities mid-session with /persona without leaving the voice channel.
Zero Server Transcoding
The server relays raw 48 kHz Opus frames without decoding or inspecting audio content. All STT and TTS run on your client.
Local AGY or Local Groq
Run the full local AGY bundle or use your own Groq API key stored exclusively in your Windows Credential Locker.
Role & Channel Controls
Restrict command access with /setroles and rely on mandatory spoken and text channel disclosures when users join.
Pricing
Start on the Free tier or upgrade for unlimited daily voice time and custom personas.
Free Tier
- ✅ 10 Daily Voice Minutes (resets 0:00 UTC)
- ✅ Default System Persona
- ✅ Push-to-Talk 48 kHz Opus Relay
- ✅ Outbound Companion App Pairing
Oracle Premium
- ✅ Unlimited Daily Voice Minutes
- ✅ Unlimited Custom AI Personas
- ✅ Oracle Bundle & Groq Listener Support
- ✅ Live Mid-Session Persona Updates
- ✅ Priority Session Capacity Pool
