How vexa-ai/vexa works
Vexa is a meeting-bot and real-time transcription platform with a self-hosted agent lane that turns meeting knowledge into workspace files. The principal workflow is a user request through the gateway, bot capture and transcription, then a live transcript in the Terminal; users can also ask an agent to work with workspace context and commit knowledge. README documents the overall runtime and domain split; sampled source confirms key pipeline and rendering behavior. Several unsampled subsystems are represented at their existing directory paths, so their internal links are limited to documented relationships.
Main entry points: app.py API gateway · pipeline.ts Bot service · join Meeting platforms · control_plane Agent control plane
Components 16
Meeting capture
- Bot service
core/meetings/services/bot/src/pipeline.ts - Meeting platforms
core/meetings/modules/join - Audio capture
core/meetings/modules/recording/src/audio-pipeline.ts
Agent workspace
- Agent control plane
core/agent/control_plane - Sandbox runtime
core/runtime/src/runtime_kernel - Agent turn engine
core/agent/worker/engine.py - Workspace files
core/workspaces/shared - Workspace store
core/workspaces/store/client.py
Gateway and interfaces
- API gateway
core/gateway/services/gateway/src/gateway/app.py - MCP service
core/gateway/services/mcp/src/vexa_mcp/app.py - Terminal workbench
clients/terminal/src/app
Transcription and display
- Channel pipeline
core/meetings/modules/gmeet-pipeline/src/gmeet-pipeline.ts - Mixed pipeline
core/meetings/modules/mixed-pipeline/src/teams-csrc-gmeet-pipeline.ts - Transcription service
core/meetings/services/transcription/src/transcription/main.py - Live transcript view
clients/terminal/src/canvas/LiveTranscriptEngine.tsx
Other components
- User
Connections 20
- User → API gateway: requests bot
- API gateway → Bot service: dispatches capture
- Bot service → Meeting platforms: joins meeting
- Meeting platforms → Audio capture: provides audio
- Bot service → Channel pipeline: selects pipeline
core/meetings/services/bot/src/pipeline.ts - Bot service → Mixed pipeline: selects pipeline
core/meetings/services/bot/src/pipeline.ts - Channel pipeline → Transcription service: requests transcription
- Mixed pipeline → Transcription service: requests transcription
- Bot service → API gateway: streams transcript
- User → Terminal workbench: views transcript
- Terminal workbench → Live transcript view: renders segments
clients/terminal/src/canvas/LiveTranscriptEngine.tsx - Live transcript view → API gateway: reads live stream
- User → API gateway: asks agent
- API gateway → Agent control plane: routes agent request
- Agent control plane → Sandbox runtime: dispatches workload
- Sandbox runtime → Agent turn engine: runs isolated turn
- Agent turn engine → Workspace files: reads and writes
core/agent/worker/engine.py - Workspace store → Workspace files: manages files
core/workspaces/store/client.py - User → MCP service: connects tools
- MCP service → API gateway: calls API
core/gateway/services/mcp/src/vexa_mcp/app.py