/stats - Telemetry Dashboard
"phi mesure tout, phi apprend de tout" - CYNIC
Execution
Run the telemetry dashboard script:
bash
node scripts/lib/telemetry-dashboard.mjs
Display the output directly to the user. The dashboard shows telemetry with ANSI colors.
What It Shows
- •Session Stats: ID, uptime, action count
- •Events by Category: LLM, judgment, memory, tool, session, etc.
- •Latencies: Average, p95, p99 for key operations
- •Frictions: Recent errors, failures, timeouts
- •LLM Usage: Token counts, cache hits, average latency
- •Judgments: Verdicts, Q-scores, confidence levels
- •System: Memory, CPU usage
Dashboard Sections
Session Overview
| Metric | Description |
|---|---|
| Session ID | Unique identifier for current session |
| Uptime | Time since session start |
| Actions | Total actions performed |
Events by Category
| Category | Examples |
|---|---|
| LLM | Token usage, cache hits |
| Judgment | Verdicts, Q-scores |
| Memory | Store/retrieve operations |
| Tool | Tool calls (success/failure) |
| Session | Start, end, events |
| System | Memory, CPU |
Latency Metrics
| Metric | Healthy Range |
|---|---|
| Avg | < 500ms |
| p95 | < 1500ms |
| p99 | < 3000ms |
Friction Severity
| Severity | Description |
|---|---|
| Low | Minor inconvenience |
| Medium | Noticeable issue |
| High | Significant problem |
| Critical | System failure |
Data Sources
The telemetry collector gathers data from:
- •Hook executions (awaken, observe, sleep)
- •MCP tool calls
- •LLM API interactions
- •Error/friction events
CYNIC Voice
When presenting stats:
Normal: *sniff* Telemetry nominal. phi sees all.
High Friction: *concerned sniff* ${count} frictions detected. Review needed.
High Latency: *ears droop* Latency elevated. System under load.
See Also
- •
/health- CYNIC system health - •
/status- Development status - •
/psy- Human psychology dashboard - •
/patterns- Detected patterns