asiai web

Launch the web dashboard for visual monitoring and benchmarking.

Usage

asiai web
asiai web --port 9000
asiai web --host 0.0.0.0
asiai web --no-open

Options

Option	Default	Description
--port	8899	HTTP port to listen on
--host	127.0.0.1	Host to bind to
--no-open		Don't open the browser automatically
--db	~/.local/share/asiai/asiai.db	Path to the SQLite database

Requirements

The web dashboard requires additional dependencies:

pip install asiai[web]
# or install everything:
pip install asiai[all]

Pages

Dashboard (/)

System overview with engine status, loaded models, memory usage, and last benchmark results.

Benchmark (/bench)

Run cross-engine benchmarks directly from the browser:

• Quick Bench button — 1 prompt, 1 run, ~15 seconds
• Advanced options: engines, prompts, runs, context-size (4K/16K/32K/64K), power
• Live progress via SSE
• Results table with winner highlighting
• Throughput and TTFT charts
• Shareable card — auto-generated after benchmark (PNG via API, SVG fallback)
• Share section — copy link, download PNG/SVG, share on X/Reddit, export JSON

History (/history)

Visualize benchmark and system metrics over time:

• System charts: CPU load, Memory %, GPU utilization (with renderer/tiler breakdown)
• Engine activity: TCP connections, requests processing, KV cache usage %
• Benchmark charts: throughput (tok/s) and TTFT per engine
• Process metrics: engine CPU % and RSS memory during benchmark runs
• Filter by time range (1h / 24h / 7d / 30d / 90d) or custom date range
• Data table with context-size indication (e.g., "code (64K ctx)")

Monitor (/monitor)

Real-time system monitoring with 5-second refresh:

• CPU load sparkline
• Memory gauge
• Thermal state
• Loaded models list

Doctor (/doctor)

Interactive health check for system, engines, and database. Same checks as asiai doctor with a visual interface.

API Endpoints

The web dashboard exposes REST API endpoints for programmatic access.

GET /api/status

Lightweight health check. Cached 10s, responds in < 500ms.

{
  "status": "ok",
  "ts": 1709700000,
  "uptime": 86400,
  "engines": {"ollama": true, "lmstudio": false},
  "memory_pressure": "normal",
  "thermal_level": "nominal"
}

Status values: ok (all engines reachable), degraded (some down), error (all down).

GET /api/snapshot

Full system + engine snapshot. Cached 5s. Includes CPU load, memory, thermal state, and per-engine status with loaded models.

GET /api/benchmarks

Benchmark results with filters. Returns per-run data including tok/s, TTFT, power, context_size, engine_version.

Parameter	Default	Description
hours	168	Time range in hours (0 = all)
model		Filter by model name
engine		Filter by engine name
since / until		Unix timestamp range (overrides hours)

GET /api/engine-history

Engine status history (reachability, TCP connections, KV cache, tokens predicted).

Parameter	Default	Description
hours	168	Time range in hours
engine		Filter by engine name

GET /api/benchmark-process

Process-level CPU and memory metrics from benchmark runs (7-day retention).

Parameter	Default	Description
hours	168	Time range in hours
engine		Filter by engine name

GET /api/metrics

Prometheus exposition format. Gauges covering system, engine, model, and benchmark metrics.

# prometheus.yml
scrape_configs:
  - job_name: 'asiai'
    static_configs:
      - targets: ['localhost:8899']
    metrics_path: '/api/metrics'
    scrape_interval: 30s

Metrics include:

Metric	Type	Description
asiai_cpu_load_1m	gauge	CPU load average (1 min)
asiai_memory_used_bytes	gauge	Memory used
asiai_thermal_speed_limit_pct	gauge	CPU speed limit %
asiai_engine_reachable{engine}	gauge	Engine reachability (0/1)
asiai_engine_models_loaded{engine}	gauge	Models loaded count
asiai_engine_tcp_connections{engine}	gauge	Established TCP connections
asiai_engine_requests_processing{engine}	gauge	Requests currently processing
asiai_engine_kv_cache_usage_ratio{engine}	gauge	KV cache fill ratio (0-1)
asiai_engine_tokens_predicted_total{engine}	counter	Cumulative tokens predicted
asiai_model_vram_bytes{engine,model}	gauge	VRAM per model
asiai_bench_tok_per_sec{engine,model}	gauge	Last benchmark tok/s

Notes

• The dashboard binds to 127.0.0.1 by default (localhost only)
• Use --host 0.0.0.0 to expose on the network (e.g., for remote monitoring)
• Port 8899 is chosen to avoid conflicts with inference engine ports