lobehub / lobe-chat

What This Repo Claims

Run a full-featured AI chat interface locally, connected to ollama models, with a single Docker command. No API key. No cloud dependency. No database required in standalone mode.

70.5k stars. Marketed as a modern, open-source alternative to ChatGPT that runs entirely on your machine.

What Was Tested

Standalone mode — the documented path for local ollama use.

Environment:

• macOS, Docker 28.3.2
• ollama 0.15.4
• Models available: llama3.2:latest, llama3.1:8b, deepseek-r1:1.5b

Install

docker run -d \
  --name lobechat-test \
  -p 3210:3210 \
  -e OLLAMA_PROXY_URL=http://host.docker.internal:11434 \
  -e OLLAMA_MODEL_LIST="+llama3.1:8b,+llama3.2:latest" \
  lobehub/lobe-chat

Image pulled cleanly. Container started. UI live at localhost:3210 within seconds.

What Worked

• UI loaded immediately — clean, polished interface
• Ollama detected as a provider automatically
• Model list populated from local ollama instance
• Sent "Hello, what model are you?" — coherent response received from llama3.2:latest
• Zero API keys required
• Zero database setup required
• Zero auth configuration required

One Friction Point

Default model picker showed llama3.1 but local model is tagged llama3.1:8b — caused a 404 from ollama on first attempt. Fixed by passing OLLAMA_MODEL_LIST with exact model names as an environment variable. Documented path exists in the official docs. Minor friction, not a blocker.

What Was Not Tested

• Full server mode (Docker Compose + PostgreSQL + Minio + Casdoor auth)
• Multi-user setup
• File upload / RAG features
• MCP marketplace integration

Verdict: SOLID

The core claim holds. One Docker command, ollama running locally, no API key — chat works. The model name tag mismatch is a minor UX issue with a documented fix. Standalone mode is production-ready for personal local AI use.