Skip to content

Conversation

@vapi-tasker
Copy link
Contributor

@vapi-tasker vapi-tasker bot commented Jan 15, 2026

Summary

  • Adds comprehensive documentation explaining data flow through Vapi's voice pipeline
  • Documents all pipeline components: Transport, STT, Orchestration, LLM, TTS
  • Includes detailed Mermaid diagrams for pipeline architecture and data flow scenarios
  • Provides "bring your own infrastructure" support matrix for each component
  • Clarifies what data is stored vs ephemeral for each pipeline stage

Changes

File Change
fern/security-and-privacy/data-flow.mdx New documentation page (comprehensive update)
fern/docs.yml Added navigation entry in Security and privacy section

Documentation Covers

Pipeline Components

  1. Transport Layer - SIP, Telephony (Twilio/Telnyx), WebSocket, WebRTC
  2. Speech-to-Text - Supported providers, custom transcriber support
  3. Orchestration Layer - Vapi's proprietary models (endpointing, interruption detection, emotion detection, backchanneling, filler injection)
  4. LLM - Supported providers, custom LLM server support
  5. Text-to-Speech - Supported providers, custom TTS support

Artifacts Documented

  • Call recordings
  • Transcripts
  • Call logs
  • Usage metrics
  • PCAP files

"Bring Your Own" Support Matrix

Component Provider Keys Custom Server
STT Deepgram only ✅ Any via WebSocket
Orchestration ❌ (Vapi proprietary)
LLM ✅ All providers ✅ OpenAI-compatible
TTS ✅ All providers ✅ PCM audio endpoint
Storage N/A ✅ S3/GCP/R2

Diagrams Included

  1. Complete Pipeline Flow - Shows all components from Transport → STT → Orchestration → LLM → TTS → Storage
  2. Default Data Flow - Standard configuration with Vapi storage
  3. Custom Storage Data Flow - With customer cloud buckets
  4. Custom Models Data Flow - With customer STT/LLM/TTS servers

Test plan

  • Verify MDX syntax renders correctly in Fern preview
  • Confirm all 4 Mermaid diagrams display properly
  • Check navigation links work in docs.yml
  • Validate cross-references to other documentation pages
  • Review provider support tables for accuracy

Related

🤖 Generated with Claude Code

vapi-tasker bot and others added 2 commits January 15, 2026 19:17
Adds comprehensive documentation explaining data flow when using
custom bucket storage and custom LLM configurations in Vapi.

Includes:
- Mermaid diagrams for default, custom storage, custom LLM, and combined flows
- Data storage summary table
- Use case recommendations
- Cross-references to related documentation

Addresses customer inquiries about what data is passed and stored
in Vapi vs their own systems.

VAP-11220

Co-Authored-By: Claude <noreply@anthropic.com>
Adds the new data flow documentation page to the Security and
privacy section in the docs navigation.

VAP-11220

Co-Authored-By: Claude <noreply@anthropic.com>
@github-actions
Copy link
Contributor

…ture

- Added complete voice pipeline architecture diagram with all components
- Documented all pipeline stages: Transport, STT, Orchestration, LLM, TTS
- Added "bring your own" support matrix for each component
- Clarified orchestration layer is Vapi-proprietary (ephemeral processing)
- Added custom models data flow diagram
- Expanded artifacts storage summary with all artifact types
- Added provider support tables for STT, LLM, and TTS
- Improved use case recommendations

VAP-11220

Co-Authored-By: Claude <noreply@anthropic.com>
@github-actions
Copy link
Contributor

Changes based on feedback:
1. Simplified "Complete Pipeline Flow" diagram - removed inner boxes from
   STT, LLM sections, removed Post-Processing
2. Updated STT table with correct BYOK support based on credential.types.ts
   (Deepgram, Gladia, AssemblyAI, Speechmatics, Google, Azure all support BYOK)
3. Removed "Custom Server" column from component tables, added notes with
   links to custom integration docs instead
4. Renamed "Custom API Keys" to "Bring your own API Key" throughout
5. Ensured consistent naming across all diagrams (Transport, Transcriber,
   Orchestration, LLM, Voice)
6. Fixed custom models diagram layout - user at top, custom servers at bottom

VAP-11220

Co-Authored-By: Claude <noreply@anthropic.com>
@github-actions
Copy link
Contributor

- Convert Speech-to-Text provider table to bullet list format
- Convert Language Model provider table to bullet list format
- Convert Text-to-Speech provider table to bullet list format
- Maintain custom voice references (confirmed supported via CustomVoice class)

VAP-11220
@github-actions
Copy link
Contributor

- Fix Custom TTS link to customization/custom-voices/custom-tts
- Update Custom Storage Data Flow: rename "Vapi Metadata" to "Vapi Storage" with Product Usage Metrics and System Logs
- Add clear distinction between System Logs (internal) and Call Logs (customer-facing)
- Ensure consistent naming across all diagrams
- Add Vapi Storage to Custom Models Data Flow diagram
- Fix BYOI Summary: Transport BYOK via Twilio/Telnyx/Vonage, Storage in BYOK column
- Fix Artifacts Storage Summary: Custom Storage as Yes/No, add HIPAA disclaimer

Co-Authored-By: Claude <noreply@anthropic.com>
@github-actions
Copy link
Contributor

@stevenbdf stevenbdf merged commit 65e69f2 into main Jan 15, 2026
6 checks passed
@stevenbdf stevenbdf deleted the tasker/VAP-11220-data-flow-docs branch January 15, 2026 20:51
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants