HubScope

Availability monitoring and quality evaluation for AI hubs (model gateways).
One Go binary. Embedded Vue dashboard with light and dark themes. Embedded SQLite. No runtime dependencies.

简体中文

---

What it does

• Availability monitoring — Probes every endpoint (model × protocol) every 5 minutes, over both non-streaming and streaming requests. Records success, latency, TTFT and token usage, and derives a green / yellow / red status with 24-hour availability trends.
• Model evaluation — 5 built-in capability suites (instruction following, reasoning, coding, knowledge Q&A, language understanding & generation) with hybrid rule-based + LLM-judge scoring, on a weekly schedule or on demand. Absolute scores stay comparable across time — silent vendor downgrades get caught.
• Alerting — Pushes to a Lark (Feishu) group bot when an endpoint keeps failing and when it recovers.
• Admin console — Hubs, models, webhooks and judge models are all configured online. Multi-user with per-hub roles and isolation. Light and dark themes included.

Get started

One-command install (Linux — downloads the latest release binary, verifies its sha256, and installs a hardened systemd service; no Go/Node toolchain needed):

curl -fsSL https://raw.githubusercontent.com/taliove/hubscope/main/scripts/install.sh | sudo bash
sudo DATA_PATH=/var/lib/hubscope/app.db /usr/local/bin/hubscope admin create --username admin --password 'your-strong-password'

Pin a version with HUBSCOPE_VERSION=v0.5.1. Overridable via env: HUBSCOPE_PREFIX, HUBSCOPE_DATA_DIR, HUBSCOPE_PORT — see the script header.

Docker (no Go/Node toolchain needed — the image builds itself):

git clone https://github.com/taliove/hubscope.git && cd hubscope
docker compose up -d --build
docker compose exec hubscope hubscope admin create --username admin --password 'your-strong-password'

Behind an HTTP proxy? The build pulls base images, Go modules and npm packages, so both the Docker daemon and the build itself need the proxy:

# 1. daemon: /etc/systemd/system/docker.service.d/proxy.conf
[Service]
Environment="HTTP_PROXY=http://<proxy-host>:<port>"
Environment="HTTPS_PROXY=http://<proxy-host>:<port>"
Environment="NO_PROXY=localhost,127.0.0.1,::1"
# then: sudo systemctl daemon-reload && sudo systemctl restart docker

# 2. build containers:
HTTP_PROXY=http://<proxy-host>:<port> HTTPS_PROXY=http://<proxy-host>:<port> \
  docker compose up -d --build

Prebuilt binary, no service (Linux / macOS, amd64 / arm64 — from Releases; good for a quick look):

curl -LO https://github.com/taliove/hubscope/releases/download/v0.5.1/hubscope_v0.5.1_linux_amd64.tar.gz
tar xzf hubscope_v0.5.1_linux_amd64.tar.gz
./hubscope_v0.5.1_linux_amd64
./hubscope_v0.5.1_linux_amd64 admin create --username admin --password 'your-strong-password'

Then open http://localhost:8080, log in, add a hub — models are discovered automatically.

Production deployment (systemd / nginx / reverse proxy): docs/deployment.md. Building from source or hacking on the code: sudo scripts/install.sh --build-from-source (requires Go + pnpm).

Build from source

Requires Go 1.26+ and pnpm:

make build          # frontend (vite) + single binary → bin/hubscope
./bin/hubscope      # serves :8080, data in ./data/app.db

Configuration

Everything is injected via environment variables; hub credentials live in the database (managed from the admin console), never in env or git.

Variable	Default	Purpose
ADDR	:8080	Listen address
DATA_PATH	./data/app.db	SQLite data file

Full table: docs/deployment.md.

License

MIT