visionforge4d/eddy-llm-workbench

Eddy LLM Workbench

Local-first switchboard for six providers: Groq • *** • OpenAI • Google • LM Studio • *** BYO keys • Zero telemetry • One-command run • Compare outputs side-by-side

Quickstart

Run the container (multi-arch image: linux/amd64 and arm64):

bash
docker run -d --name llm-workbench --restart unless-stopped \
  -p 127.0.0.1:8080:8080 \
  -e NODE_ENV=production \
  -e PORT=5000 \
  -v "$HOME/.config/llm-workbench:/home/llm/.config/llm-workbench:rw" \
  visionforge4d/eddy-llm-workbench:latest

# Health (Nginx + API)
curl -fsS http://127.0.0.1:8080/nginx-healthz && echo "OK"
curl -fsS http://127.0.0.1:8080/health && echo

Fedora/SELinux note: add :Z to the -v mount if you see permission warnings.

Prefer GHCR? Use ghcr.io/<your-gh-username>/llm-workbench:latest instead of the Docker Hub image.

Open: http://localhost:8080

Click a tile to choose a provider. Run pro***s, compare outputs, export sessions.

What it is

• Switchboard UI Six providers live with distinct tiles & icons (Groq, ***, OpenAI, Google, LM Studio, ***).

• Side-by-side compare Run the same pro*** across multiple providers and view outputs together, with basic latency notes.

• Local session log Every run is saved to JSONL under your config directory.

• Export One-click export of a session to JSONL or Markdown (includes provider, model, and timing).

• Local-first ***ds to localhost, strict CORS, zero telemetry, BYO keys.

Architecture & Ports

• Nginx serves the UI and proxies at :8080.

• Node API listens on :5000 (proxied at /api/*).

• Health endpoints:

  • Nginx: GET /nginx-healthz → ok
  • API: GET /health → JSON

Configuration

Config file (created on first run):

~/.config/llm-workbench/config.yml

Example:

yaml
keys:
  openai: ""          # sk-...
  groq: ""            # gsk_...
  anthropic: ""       # claude key
  google: ""          # Gemini key
  mistral: ""         # Mistral key
  lmstudio_base_url: "http://127.0.0.1:1234"  # LM Studio local endpoint

server:
  bind_host: "0.0.0.0"
  port: 5000   # API port inside the container (Nginx proxies 8080 -> 5000)

security:
  cors_origins: ["http://localhost:8080"]
  rate_limit_rpm: 60

ui:
  default_provider: "lmstudio"

• Missing provider key? Its tile is greyed out until configured.
• LM Studio can be used as the fallback provider when keys aren’t set.

Docker Compose (optional)

yaml
services:
  llm-workbench:
    image: visionforge4d/eddy-llm-workbench:latest
    environment:
      NODE_ENV: production
      PORT: "5000"
    ports:
      - "127.0.0.1:8080:8080"
    volumes:
      - ${HOME}/.config/llm-workbench:/home/llm/.config/llm-workbench:rw
    restart: unless-stopped

Upgrades (safe pattern)

Zero(ish) downtime: verify on a side port, then switch.

bash
# Spin up new on 8081 and verify
docker run -d --name llmwb_new \
  -p 127.0.0.1:8081:8080 \
  -e NODE_ENV=production -e PORT=5000 \
  visionforge4d/eddy-llm-workbench:latest

curl -fsS 127.0.0.1:8081/nginx-healthz && curl -fsS 127.0.0.1:8081/health

# Promote to 8080
docker rm -f llm-workbench || true
docker rename llmwb_new llm-workbench
docker run -d --name llm-workbench --restart unless-stopped \
  -p 127.0.0.1:8080:8080 \
  -e NODE_ENV=production -e PORT=5000 \
  visionforge4d/eddy-llm-workbench:latest

Troubleshooting

• “port is already allocated” — Something else has 8080. Free it or use -p 127.0.0.1:8081:8080.

• TCP reset on curl — Usually too early after start; try again, or check logs:

  bash
  docker logs --tail=100 llm-workbench
  docker exec -it llm-workbench sh -lc 'busybox netstat -ltn'
  

• map directive is not allowed here — You’re loading a server file as the master. The image already ships a valid master.

• SELinux “permission denied” — Add :Z to ***d mounts.

Tags & Platforms

• latest – rolling stable
• X.Y.Z – immutable release
• X.Y – rolling minor
• Platforms: linux/amd64, linux/arm64

Build from source (multi-arch)

bash
docker buildx create --use || true
docker buildx build --platform linux/amd64,linux/arm64 \
  -t visionforge4d/eddy-llm-workbench:1.7.6 \
  -t visionforge4d/eddy-llm-workbench:latest \
  --push .

Security posture

• Runs as non-root user llm

• No writes to /var/lib/nginx; temps under /tmp

• Logs to stdout/stderr (12-factor style)

• Safe defaults (no secrets in image)

• Optional runtime hardening:

  bash
  docker run -d --name llm-workbench \
    --read-only --tmpfs /tmp:rw,noexec,nosuid,nodev \
    -p 127.0.0.1:8080:8080 \
    -e NODE_ENV=production -e PORT=5000 \
    visionforge4d/eddy-llm-workbench:latest
  

Roadmap

• Add Ollama provider
• Fedora COPR / RPM packaging (systemd service)
• Flatpak for Linux desktop
• Preset Compare pairs (Groq ↔ ***, OpenAI ↔ ***)

Contributing

• See CONTRIBUTING.md for guidelines.
• Provider ***s are small modules—follow the Provider interface to add more.
• Please open issues for bugs, feature requests, or provider quirks.

License

MIT © VisionForge4d

Feedback, issues, and PRs very welcome.