how to run gemma locally with openclaw (step-by-step guide)

step 1 — install ollama

download and install ollama: download ollama. then verify it works:

ollama --version

if that prints a version — you're good

step 2 — pull gemma model

you have two options here:

option A (recommended): let openclaw handle it automatically
→ skip this step and go to step 3

option B (manual): install model yourself

ollama pull gemma:26b

or if you want a lighter/faster version:

ollama pull gemma:7b

then test it:

ollama run gemma:26b

if it responds – model is working locally

step 3 – run openclaw with ollama backend

you don’t need to manually install openclaw – it can run directly. run:

npx openclaw@latest --local --agent main -m gemma:26b

what this does:

• installs openclaw (if not installed)
• connects it to ollama
• launches the default agent
• uses your local gemma model as brain

---

how it works (simple explanation)

• ollama runs the model locally
• gemma – the actual LLM
• openclaw – agent layer (tools, workflows, automation)

together: you get a local AI agent system

---

system requirements (realistic, not marketing)

for gemma 26b:

• 32–64 GB RAM recommended
• Apple Silicon (M1/M2/M3) or strong GPU

for gemma 7b:

• works on most laptops (16 GB RAM is enough)

if it crashes – switch to 7b

---

common issues (and fixes)

1. “model not found”

ollama list

→ check exact model name (case matters)

2. openclaw doesn’t connect to ollama

make sure ollama is running:

ollama serve

3. very slow responses

you're running too big model. switch:

-m gemma:7b

---

quick test

once openclaw starts, try:

hello, what model are you using?

if it answers – everything works. congrats!

---

what you can do with this

• run autonomous agents locally
• build workflows without api costs
• test ideas privately
• prototype ai products fast