One file. Double-click. LeadGemma finds qualified business leads using a local AI model — no API keys, no cloud, no monthly fees.
v0.0.4 · ~6 MB download · Sets up automatically on first run (~10 GB)
Grab the single file for your operating system. It is about 6 MB.
Double-click or run from a terminal. On first launch it downloads Ollama, the AI model, Node.js, and Chromium automatically.
Type a plain-English brief like "Find luxury hotels in Rome that serve US tourists" and the AI builds a search plan.
Leads are scored, stored locally, and downloadable as a CSV with names, websites, phones, emails, and fit scores.
The first time you run LeadGemma, the launcher automatically downloads and sets up everything it needs. This is a one-time process — subsequent launches take 5–10 seconds.
| Ollama (AI runtime) | ~90 MB |
| Gemma 4 model (gemma4:e4b) | ~9.6 GB |
| Node.js v22 LTS | ~35 MB |
| Chromium (browser engine) | ~200 MB |
Total: ~10 GB on first run. All files are cached — re-downloads happen only when you upgrade.
leadgen-mac-arm64-v0.0.4.bin (Apple Silicon) or leadgen-mac-amd64-v0.0.4.bin (Intel).mv ~/Downloads/leadgen-mac-arm64-v0.0.4.bin ~/leadgen
chmod +x ~/leadgen
~/leadgenhttp://localhost:3000 when ready. Press Enter in the terminal to reopen the browser at any time.leadgen-windows-v0.0.4.exe.http://localhost:3000.leadgen-linux-amd64-v0.0.4.bin.mv ~/Downloads/leadgen-linux-amd64-v0.0.4.bin ~/leadgen
chmod +x ~/leadgen
~/leadgensudo apt install -y libnss3 libatk1.0-0 libatk-bridge2.0-0 libcups2 libxdamage1 libgbm1When you submit a search brief, LeadGemma runs a multi-step pipeline entirely on your machine:
Type your search brief here and press Enter. The AI will ask follow-up questions if your brief is too vague before starting a run.
Only leads scoring at or above this threshold (0–10) are saved and exported. Default: 7. Lower it to get more leads, raise it for stricter quality.
The run stops once this many leads have been found. Default: 50. Set it to 10 for a quick test or 100+ for comprehensive coverage.
Downloads leads from the current run only. Available as soon as the first lead is found — you do not need to wait for the run to finish.
Downloads all leads across every run that meet the current Min Fit Score threshold. Useful for combining multiple searches.
Gracefully stops the current run after the current task finishes. Leads collected so far are saved and downloadable.
A good brief tells the AI what kind of business to find, where, and why they are a good lead.
"Find luxury hotels and boutique resorts in Rome and Florence that cater to international tourists and offer high-end services."
"Find private dental clinics in Milan that serve wealthy patients and may have international clients."
"Find businesses in Italy."
"Find hotels."
If your brief is unclear, the AI will ask you a clarifying question before starting. Answer it in the chat to provide more context.
Each business is scored 0–10 by the AI classifier using your original brief as the scoring criteria:
| Score | Meaning |
|---|---|
| 8–10 | Strong match — business clearly fits the criteria, high-value transactions, likely international clients |
| 6–7 | Moderate match — fits the industry but some criteria uncertain |
| 4–5 | Weak match — right general area but unlikely to be a good lead |
| 0–3 | Poor match — does not fit the criteria |
The default threshold of 7 keeps results tight. If you're getting too few leads, try lowering it to 5 or 6.
LeadGemma is designed to keep your data local. Here is exactly what leaves your machine:
| Service | What is sent | Why |
|---|---|---|
| Google Maps / Google Search | Search queries (e.g. "luxury hotel Rome") | To find businesses |
| Everything else | Nothing — runs on your machine | Website reading, AI planning and classification |
All leads and run history are stored in a local SQLite database at:
~/Library/Application Support/LeadGemma/app/data/leadgen.db%APPDATA%\LeadGemma\app\data\leadgen.db~/.local/share/leadgemma/app/data/leadgen.dbAll versions are available to download. We recommend always using the latest release.
The binaries are unsigned (code-signing certificates cost ~$100–400/year). To bypass Gatekeeper: right-click the file in Finder, choose Open, and click Open in the security dialog. You only need to do this once.
This is Windows SmartScreen warning about unsigned software. Click More info, then Run anyway. You only need to do this once per version.
Press Enter in the terminal window where LeadGemma is running — this reopens the browser. Alternatively navigate to http://localhost:3000 manually.
Another application is using port 3000. Stop it, then relaunch LeadGemma. On macOS/Linux you can find the process with lsof -i :3000.
If Ollama is already installed on your system, LeadGemma uses it automatically. If the gemma4:e4b model is missing, the launcher will pull it (this takes 10–30 minutes on the first run and requires ~9.6 GB of disk space).
Install the required system libraries:
sudo apt install -y libnss3 libatk1.0-0 libatk-bridge2.0-0 libcups2 libxdamage1 libgbm1After the first round of searches, the AI reviews the results and may decide coverage is already sufficient. This is intentional. If you want more leads, increase Max Leads and start a new run with a more specific brief.
Yes. Each search creates a separate run. Use "Download Run CSV" to get results per run, or "Download All Leads CSV" to export everything at once across all runs.
All data stays on your machine. Leads are stored in a local SQLite database — see the Data and Privacy section above for the exact paths per operating system.