Ollama Integration Setup Guide

April 25, 2026 · View on GitHub

📖 Latest Version: Always refer to the OLLAMA_SETUP.md on GitHub for the most up-to-date instructions.

ThinkReview now supports local AI code reviews using Ollama! This gives you complete privacy and offline capabilities.

Why Use Ollama?

  • 🔒 Privacy: Your code never leaves your machine
  • 💰 No API Costs: Free local inference
  • ⚡ Speed: Can be faster with the right hardware
  • 🌐 Offline: Works without internet connection
  • 🎯 Custom Models: Use specialized code review models

Prerequisites

  1. Install Ollama: Download from ollama.ai/download
  2. Sufficient RAM: At least 8GB recommended (16GB+ for larger models)
  3. Disk Space: 5-20GB depending on model size

Quick Start

1. Install Ollama

macOS/Linux:

curl -fsSL https://ollama.ai/install.sh | sh

Windows: Download the installer from ollama.ai/download

2. Pull a Code Review Model

We recommend Gemma 4 — Google DeepMind’s multimodal family with strong coding and reasoning, native function calling, and a 128K context window on the default and edge variants (larger variants offer 256K). For ThinkReview, start with the default tag or pick a size that fits your RAM:

  • gemma4 (latest, ~9.6GB) — default choice for strong overall quality
  • gemma4:e4b / gemma4:e2b — smaller “effective” models for laptops and limited VRAM
  • gemma4:26b / gemma4:31b — workstation-class local models (larger download)
  • gemma4:31b-cloud — runs via Ollama’s cloud when you prefer not to host the full 31B locally

Sampling (optional): For Gemma 4, Ollama’s docs suggest temperature=1.0, top_p=0.95, and top_k=64 as a standardized setup; you can set these in the extension’s Ollama settings.

Also worth trying: Codestral, gpt-oss, and coder-focused models below.

Tested / common alternatives:

  • gemma4 — Recommended — best default for ThinkReview with Ollama
  • codestral — Strong code model (Mistral, 32k context)
  • qwen2.5-coder:30b / qwen3-coder:30b — Widely used for code
  • qwen2.5:8b — Lighter option
# Recommended (best results with Ollama + ThinkReview)
ollama pull gemma4

# Smaller or larger Gemma 4 variants (pick one that fits your machine)
ollama pull gemma4:e4b
ollama pull gemma4:e2b
ollama pull gemma4:26b
ollama pull gemma4:31b

# Other strong options
ollama pull codestral
ollama pull qwen2.5-coder:30b
ollama pull gpt-oss:20b            # OpenAI open-weight, 128K context
ollama pull codellama              # Fast, smaller footprint
ollama pull qwen2.5-coder:7b
ollama pull deepseek-coder:6.7b

3. Start Ollama Server with CORS Enabled

Important: Ollama runs on your computer while the extension runs in the browser. The steps below restart Ollama with settings that let extensions connect (including CORS), which fixes many Ollama-related errors.

Where to run these commands: use a real shell on your computer — Terminal (macOS), your terminal app (Linux), Windows PowerShell, or Command Prompt. Do not paste them into the browser’s developer console; the extension talks to Ollama from your machine, and Ollama must be started with the right environment on that same machine.

Firefox browser (Mozilla): in the commands below, replace chrome-extension://* with moz-extension://* in OLLAMA_ORIGINS.

Note: The killall / OLLAMA_ORIGINS="..." ollama serve style commands below are for macOS and Linux. Windows uses different commands (taskkill, set / $env:) — see Option A — Windows below.

Option A — Quick start (one-time), macOS & Linux:

# Step 1: Stop Ollama
killall ollama 2>/dev/null || true; killall Ollama 2>/dev/null || true; sleep 2

# Step 2: Start Ollama with CORS enabled for browser extensions
OLLAMA_ORIGINS="chrome-extension://*" ollama serve

Option A — Quick start (one-time), Windows:

Stop the Ollama app if it is running (system tray Quit, or use the commands below), then start the server from a terminal with OLLAMA_ORIGINS set.

PowerShell (recommended on Windows):

# Step 1: Stop Ollama
Stop-Process -Name ollama -ErrorAction SilentlyContinue; Start-Sleep -Seconds 2

# Step 2: Start Ollama with CORS enabled for browser extensions
$env:OLLAMA_ORIGINS='chrome-extension://*'; ollama serve

Command Prompt:

REM Step 1: Stop Ollama
taskkill /IM ollama.exe /F 2>nul & timeout /t 2 /nobreak >nul

REM Step 2: Start Ollama with CORS enabled for browser extensions
set OLLAMA_ORIGINS=chrome-extension://* && ollama serve

If taskkill reports the process was not found, Ollama may already be stopped — continue with step 2.

Option B — Make it permanent (recommended):

Set OLLAMA_ORIGINS permanently so you don't need to set it every time you run ollama serve:

macOS & Linux:

  1. Open your shell configuration file:

    # For Zsh
nano ~/.zshrc

# For Bash
nano ~/.bashrc

  2. Add this line at the end:

    export OLLAMA_ORIGINS="chrome-extension://*"

  3. Save and reload:

    source ~/.zshrc  # or source ~/.bashrc

  4. Now you can simply run ollama serve without setting the environment variable each time.

Windows (PowerShell):

  1. Open PowerShell and run:

    [System.Environment]::SetEnvironmentVariable('OLLAMA_ORIGINS', 'chrome-extension://*', [System.EnvironmentVariableTarget]::User)

  2. Restart PowerShell or your terminal, then run ollama serve

Windows (Command Prompt):

  1. Open Command Prompt as Administrator and run:

    setx OLLAMA_ORIGINS "chrome-extension://*"

  2. Restart Command Prompt, then run ollama serve

Note: After setting it permanently, you can simply run ollama serve and it will automatically use the CORS settings. Keep the terminal running while using the extension.

4. Configure Extension

  1. Open the ThinkReview extension popup
  2. Scroll to "AI Provider" section
  3. Select "🖥️ Local Ollama"
  4. Verify URL is http://localhost:11434 (default)
  5. Select your installed model from the dropdown
  6. Click "Test Connection" to verify
  7. Click "Save Settings"

Usage

Once configured, ThinkReview will automatically use Ollama for all code reviews:

  1. Navigate to a GitLab merge request or Azure DevOps pull request
  2. Click the ThinkReview button
  3. Your code will be analyzed locally using Ollama
  4. Review results appear in the integrated panel

Tested Models

ModelSizeStatus
gemma4 (default / e2b / e4b / 26b / 31b)~7GB–20GB+✅ Recommended — best default with Ollama
codestral~13GB✅ Strong code reviews
gpt-oss (20b / 120b)14GB / 65GB✅ Strong reasoning, 128K context
qwen2.5-coder:30b / qwen3-coder:30b~20GB✅ Common coder picks
qwen2.5:8b~5GB✅ Lighter option

Other models may work but are not officially tested.

Troubleshooting

Connection Failed (403 Forbidden)

CORS / allowed origins are not set correctly. Use Option A above to restart Ollama with OLLAMA_ORIGINS (pick the block for macOS/Linux or Windows).

Connection Failed (Cannot Connect)

  1. Verify Ollama is running: ollama list
  2. Check URL: http://localhost:11434
  3. Restart with CORS (Option A above)

Model Not Found

  1. List models: ollama list
  2. Pull the model: ollama pull gemma4 (or the tag you use, e.g. gemma4:e4b)
  3. Refresh in extension settings (🔄 button)

Advanced

Custom Ollama URL

Update URL in extension settings (e.g., http://192.168.1.100:11434)

Performance Tuning

export OLLAMA_NUM_GPU=1      # GPU layers
export OLLAMA_NUM_THREAD=8   # CPU threads
ollama serve

Support