G-Meter

Open-source observability for OpenAI-compatible Gonka API brokers.

G-Meter runs scheduled probe suites against configured brokers and presents a public, read-only dashboard for availability, performance, capabilities, model limits, trends, and pricing signals.

• Multi-broker catalog with per-provider model IDs and display aliases
• English and Russian UI
• Overview gauges, provider drill-downs, trend charts, history, and limit checks
• Probe logs with captured request/response details for debugging
• Pricing extraction from /pricing endpoints, completion response bodies, or billing headers when brokers expose it
• Public read-only mode for hosted dashboards

• API uptime
• Latency
• Failed probe percentage
• Output speed and stream speed
• Capability pass rate: JSON mode, tool calling, multimodality, input/output probes
• Maximum context and maximum output tokens
• Estimated spend per 1M tokens when pricing data is available

cp .env.example .env
# edit .env and add at least one broker API key
docker compose up --build

Open:

• Frontend: http://localhost:5173
• API health: http://localhost:8000/api/health

Broker entries without API keys are seeded as disabled. Add keys in .env, or provide your own broker catalog via BROKERS_CONFIG_JSON or BROKERS_CONFIG_PATH.

The backend includes backend/brokers.json as the default broker catalog. It is safe to commit because it only references environment variables.

Useful variables:

Example broker catalog:

{
  "brokers": [
    {
      "name": "proxy.gonka.gg",
      "base_url": "https://proxy.gonka.gg/v1",
      "api_key": "${GONKA_API_KEY}",
      "models": [
        { "id": "moonshotai/Kimi-K2.6", "alias": "kimi-k2.6" },
        { "id": "Qwen/Qwen3-235B-A22B-Instruct-2507-FP8", "alias": "qwen3-235b" }
      ]
    }
  ]
}

models may also be an alias-to-model object:

{
  "models": {
    "kimi-k2.6": "moonshotai/Kimi-K2.6",
    "qwen3-235b": "Qwen/Qwen3-235B-A22B-Instruct-2507-FP8"
  }
}

Environment references like ${GONKA_API_KEY} and ${VAR:-fallback} are expanded before parsing.

Backend:

cd backend
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

Frontend:

cd frontend
npm install
npm run dev

backend/    FastAPI, SQLite, APScheduler, probe runner
frontend/   React + Vite dashboard served by nginx

Do not commit real API keys. Use .env, deployment secrets, or BROKERS_CONFIG_JSON supplied by your runtime environment.

MIT