Skip to content

Mostafa-M-Hussein/memos-proxy

Repository files navigation

memos-proxy

Use MemOS hosted models (qwen3-32b, deepseek-r1, qwen2.5-72b-instruct) through Open WebUI, OpenClaw, the Ollama CLI, or any OpenAI-compatible tool. No local GPU needed.

Prerequisites

Install these manually before running the setup script:

# What How
1 Docker docs.docker.com/get-docker
2 Ollama ollama.com/download
3 OpenClaw After Ollama is installed: ollama launch openclaw
4 MemOS API key Sign up at memos-dashboard.openmem.net

Everything else is handled by the setup script. Stuck? quick tutorial

Quick start

git clone https://github.com/Mostafa-M-Hussein/memos-proxy.git
cd memos-proxy

macOS / Linux

cp .env.example .env        # paste your MEMOS_API_KEY
./setup.sh

Windows (PowerShell)

copy .env.example .env      # paste your MEMOS_API_KEY
.\setup.ps1

First run? If PowerShell blocks the script: Set-ExecutionPolicy -Scope CurrentUser RemoteSigned

That's it. The script will:

  1. Start all Docker services
  2. Install and start OpenClaw (via ollama launch openclaw), pre-configured to use Memos models

Services

Service URL What it does
open-webui http://localhost:3333 Browser chat UI
ollama-mcp-bridge http://localhost:11436 Memos models + MCP tool support
memos-proxy http://localhost:11435 Raw Ollama-compatible API

Your local Ollama stays on 11434 — no conflicts.

Architecture

open-webui (browser)
      │
      ▼
ollama-mcp-bridge :11436  ◄── OpenClaw (messaging: WhatsApp/Telegram/Slack)
      │
      ▼
memos-proxy :11435
      │
      ▼
MemOS Cloud  (qwen3-32b · deepseek-r1 · qwen2.5-72b-instruct)

Manual usage

Ollama CLI

OLLAMA_HOST=http://localhost:11436 ollama run qwen3-32b
OLLAMA_HOST=http://localhost:11436 ollama list

aider

OPENAI_API_BASE=http://localhost:11435/v1 OPENAI_API_KEY=dummy aider --model openai/qwen3-32b

curl

curl http://localhost:11435/api/chat \
  -d '{"model":"qwen3-32b","messages":[{"role":"user","content":"hello"}]}'

Stop everything

docker compose down && openclaw gateway stop

Models

Name Notes
qwen3-32b default, good balance
deepseek-r1 reasoning
qwen2.5-72b-instruct largest

About

use memos hosted models (qwen3, deepseek-r1) through ollama cli — no local gpu

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors