VideoDB Pair Programmer
AI pair programming with real-time screen and audio context. Record your screen and audio, with AI-powered indexing that logs visual and audio events in real-time.
Commands
When user asks for a command, read the corresponding file for instructions:
| Command | Description | Reference |
|---|
| Start screen/audio recording | See commands/record.md |
| Stop the running recording | See commands/stop.md |
| Search recording context (screen, mic, audio) | See commands/search.md |
/pair-programmer what-happened
| Summarize recent activity | See commands/what-happened.md |
| Install deps and configure API key | See commands/setup.md |
| Change indexing and other settings | See commands/config.md |
How It Works
- User runs to install dependencies and set environment variable
- User runs to start recording
- A picker UI appears to select screen and audio sources
- Recording starts and events are logged to
/tmp/videodb_pp_events.jsonl
- User can stop recording from the tray icon (🔴 PP → Stop Recording)
Output Files
| Path | Content |
|---|
| Process ID of the recorder |
/tmp/videodb_pp_events.jsonl
| All WebSocket events (JSONL format) |
/tmp/videodb_pp_info.json
| Current session info (session_id, rtstream_ids) |
Event File Format
Events are written as JSONL (one JSON object per line):
json
{"ts": "2026-03-05T10:15:30.123Z", "unix_ts": 1709374530.12, "channel": "visual_index", "data": {"text": "User is viewing VS Code with auth.ts open"}}
{"ts": "2026-03-05T10:15:31.456Z", "unix_ts": 1709374531.45, "channel": "transcript", "data": {"text": "Let me check the login flow", "is_final": true}}
Environment Variables
The recorder reads these from environment variables:
Reading Context
Events are in
/tmp/videodb_pp_events.jsonl
. Use CLI tools to filter — never read the whole file.
| Channel | Content | Density |
|---|
| Screen descriptions | Dense (~1 every 2s) |
| Mic speech | Sparse (sentences) |
| System audio summaries | Sparse (sentences) |
Channel filter — use grep to filter by channel, pipe to
for recent events:
bash
grep '"channel":"visual_index"' /tmp/videodb_pp_events.jsonl | tail -10
Keyword search — grep across all channels:
bash
grep -i 'keyword' /tmp/videodb_pp_events.jsonl
Time-window filter — filter events from the last N minutes:
- Get current epoch:
- Compute cutoff:
- Filter lines where the JSON field exceeds the cutoff
- Pipe through to narrow by channel
Generate the appropriate filtering command (grep, awk, python3, jq) based on complexity.
For semantic search across indexed content, use
:
bash
node search-rtstream.js --query="your query" --cwd=<PROJECT_ROOT>
is the absolute path to the user's project directory. This is NOT the skill directory — resolve it before running the command.
See commands/search.md for the full search strategy and CLI patterns.