gemma-openclaw (jthiruveedula/openclaw-gemma-pro) | MoltPulse

gemma-openclaw

Private agent runtime with Gemma via Ollama, guardrails, fallback routing, and enterprise-safe orchestration patterns for self-hosted GenAI.

jthiruveedula/openclaw-gemma-pro00

MoltPulse

Based on repository activity, growth velocity and community engagement.

Growth3/30

Activity6/25

Popularity2/25

Trust13/20

Stars

High

Sentiment

Votes

README.md

🤖 OpenClaw-Gemma-Pro

Self-hosted GenAI Pro stack — Gemma 4 via Ollama + OpenClaw + WhatsApp/Telegram + daily memory indexing for persistent cross-channel context.

Run a production-quality personal AI assistant for near-zero recurring cost.
No cloud API key required. Runs entirely on your laptop or server.

📺 What It Looks Like

 +-------------------------+
 |  You (WhatsApp/Telegram) |
 +-----------+-------------+
             |
             v
 +-------------------------+     +------------------+
 |  FastAPI Webhook Server  +---->+  Skill Router    |
 +-------------------------+     +--------+---------+
                                          |
                          +--------------+--------------+
                          |                             |
                    LITE tier                      PRO tier
                    gemma4:4b                      gemma4:27b
                  (fast, cheap)                (smart, thorough)
                          |                             |
                          +--------------+--------------+
                                          |
                                          v
                              +---------------------+
                              |  Memory Injector     |
                              |  daily summary       |
                              |  + durable facts     |
                              +---------------------+
                                          |
                                          v
                              +---------------------+
                              |   Your Response      |
                              +---------------------+

# Step 1 – Install Ollama (native Mac app)
brew install ollama
# OR download the .app from https://ollama.com/download/mac

# Step 2 – Pull Gemma 4 model
ollama pull gemma4:27b        # PRO tier (16 GB RAM+)
# ollama pull gemma4:4b       # LITE tier (8 GB RAM)

# Step 3 – Clone & bootstrap
git clone https://github.com/jthiruveedula/openclaw-gemma-pro.git
cd openclaw-gemma-pro
bash scripts/bootstrap.sh

# Step 4 – Copy env and start the multi-agent coordinator
cp .env.example .env
# Edit .env with your Telegram/WhatsApp tokens (see REQUIRED fields)
python -m workers.orchestrator.coordinator "ping"

# Step 1 – Install Ollama
curl -fsSL https://ollama.com/install.sh | sh
sudo systemctl enable ollama
sudo systemctl start ollama

# Step 2 – Pull Gemma 4 model
ollama pull gemma4:27b        # PRO tier
# ollama pull gemma4:4b       # LITE tier

# Step 3 – Clone & bootstrap
git clone https://github.com/jthiruveedula/openclaw-gemma-pro.git
cd openclaw-gemma-pro
bash scripts/bootstrap.sh

# Step 4 – Configure & run
cp .env.example .env
nano .env  # Add your bot tokens (see REQUIRED fields)
python -m workers.orchestrator.coordinator "ping"

# Step 1 – Install Ollama (Windows installer)
# Download from: https://ollama.com/download/windows
# Run OllamaSetup.exe, then open a NEW terminal

# Step 2 – Pull Gemma 4 model
ollama pull gemma4:27b

# Step 3 – Clone & bootstrap
git clone https://github.com/jthiruveedula/openclaw-gemma-pro.git
cd openclaw-gemma-pro

# Use Python venv (PowerShell)
python -m venv .venv
.venv\Scripts\Activate.ps1
pip install -r requirements.txt

# Step 4 – Configure & run
copy .env.example .env
notepad .env  # Add your bot tokens (see REQUIRED fields)
python -m workers.orchestrator.coordinator "ping"

WhatsApp / Telegram
       |
       v
Webhook Server (FastAPI) [coming soon]
       |
       v
[Skill: chat-router]
       |         |
       v         v
   LITE tier   PRO tier
   gemma4:4b  gemma4:27b
  (Ollama :11434)
       |
       v
Memory Context Injection
(daily summary + durable facts)
       |
       v
    Response sent back
       |
       v
Appended to memory/raw/YYYY-MM-DD.jsonl

User Goal
    |
    v
AgentCoordinator  (workers/orchestrator/coordinator.py)
    |
    +-- [1] PlannerAgent   --> calls Gemma 4, decomposes goal into subtasks
    |
    +-- [2] ExecutorAgents --> run subtasks IN PARALLEL (asyncio + semaphore)
    |         each action checked by ActionGuardrail before execution
    |
    +-- [3] MemoryAgent    --> persists results:
    |         raw/YYYY-MM-DD.jsonl
    |         daily/YYYY-MM-DD.md
    |         facts/index.jsonl
    |
    +-- [4] CriticAgent    --> scores output: pass / warn / fail

Any risky action
       |
       v
ActionGuardrail.check()
   |
   +-- ALLOW  --> execute
   +-- WARN   --> log + notify user
   +-- BLOCK  --> hard-stop, reason logged

Blocked by default:
  rm -rf, DROP TABLE, shutil.rmtree memory/,
  shell=True with destructive patterns,
  writes to protected config paths

openclaw-gemma-pro/
├── .github/workflows/    # CI: guardrails, secret-scan, delete-guard
├── config/               # Model routing, OpenClaw config JSON
├── guardrails/           # ActionGuardrail + pre-commit safety hook
├── scripts/
│   ├── bootstrap.sh      # One-shot setup for all platforms (macOS/Linux)
│   ├── doctor.py         # Pre-flight environment validation
│   └── setup_task_scheduler.ps1  # Windows cron equivalent
├── skills/               # Chat-router, daily-memory-indexer
├── workers/
│   ├── agents/           # planner, executor, memory, critic, cloud_fallback
│   ├── memory_indexer/   # Daily memory indexing job
│   └── orchestrator/     # AgentCoordinator (DAG runner)
├── .env.example          # Environment template (REQUIRED/OPTIONAL annotated)
├── CONTRIBUTING.md       # Contribution guide & secrets hygiene
├── requirements.txt
└── README.md

# Ollama endpoint (same for Mac/Linux/Windows)
OLLAMA_BASE_URL=http://localhost:11434

# Model selection (Gemma 4 — updated from Gemma 2)
OLLAMA_MODEL=gemma4:27b       # PRO tier
OLLAMA_LITE_MODEL=gemma4:4b   # LITE tier
OLLAMA_TIMEOUT=300            # Increase for cold-start on 27b

# Cloud fallback (optional)
# CLOUD_FALLBACK_PROVIDER=gemini
# CLOUD_FALLBACK_MODEL=gemini-2.0-flash
# GEMINI_API_KEY=your_key_here

# Messaging (optional – needed for WhatsApp/Telegram)
TELEGRAM_BOT_TOKEN=your_token_here
TWILIO_WHATSAPP_NUMBER=whatsapp:+your_number_here

# Memory paths
MEMORY_BASE_DIR=./memory

# Run a one-off goal via multi-agent coordinator
python -m workers.orchestrator.coordinator "Summarise today's Telegram messages"

# Run pre-flight environment check
python scripts/doctor.py

# Index yesterday's memory manually
python workers/memory_indexer/index_memory.py

# Install pre-commit safety hook
python guardrails/pre_commit_hook.py --install

# Run tests
pytest tests/ -v

# Install the pre-commit hook (runs automatically on git commit)
python guardrails/pre_commit_hook.py --install

# Run linting manually
ruff check . && mypy workers/ guardrails/

gemma-openclaw

MoltPulse

🤖 OpenClaw-Gemma-Pro

📺 What It Looks Like

⚡ Installation

🍎 macOS

🐧 Linux

🪟 Windows

📋 Minimum Requirements

🏗️ Architecture

🤖 Multi-Agent Architecture

Agent Table

🛡️ Guardrails

CI Safety Jobs

📁 Repo Structure

⚙️ Configuration

🚀 Quick Commands

📝 Roadmap

🤝 Contributing

📄 License

Ecosystem Role

Embed Badge