hyle - Observability Dashboard

> hyle

Live

364

Tests

20Hz

TUI Refresh

35+

Models

00:00:00

Uptime

Token Usage (last 24h) healthy

Prompt tokens 124,892

Completion tokens 48,231

Context efficiency 78%

Cost (today) $0.42

Latency p99 <2s

340ms

TTFT p50

1.2s

TTFT p99

Tok/sec

Total (p50) 2.8s

API calls (24h) 847

Active Model

claude-3.5-sonnet $3/M in

gpt-4o $5/M in

llama-3.1-70b free

deepseek-coder $0.14/M in

Event Log streaming

14:23:01 TOOL read src/agent.rs (2,847 bytes)

14:23:02 API stream_completion started, model=claude-3.5-sonnet

14:23:04 INFO context compressed: 45,231 -> 12,847 tokens (72% reduction)

14:23:05 TOOL write src/agent.rs (atomic, backup created)

14:23:05 TOOL bash: cargo test --lib agent

14:23:12 INFO tests passed: 47/47 in 6.8s

14:23:13 API stream complete: 1,847 tokens, TTFT=340ms

Sessions

abc123 now

def456 2h ago

ghi789 yesterday

jkl012 2d ago

Thread Architecture

TUI Event Loop (50ms, ~20Hz)
├── telemetry_rx.try_recv() ← OS thread
├── process_bg_responses()  ← BgWorker
├── rx.try_recv()           ← API stream
├── event::poll()           ← Keyboard
└── render_tui()            ← Pure CPU

Click a component to see details. Non-blocking I/O throughout. All file operations via spawn_blocking.

Live Terminal

> add error handling to parse_config

[read] src/config.rs

[patch] src/config.rs +15 lines

[bash] cargo test config

test config::tests::parse_valid ... ok

test config::tests::parse_missing ... ok

test config::tests::parse_invalid ... ok