Focal Harvest Performance Benchmarks
These figures are measured on typical hardware during standard research sweeps. They reflect Focal Harvest's design priority: a low-footprint pipeline that runs comfortably on low-end student laptops without a GPU.
Benchmark Table
| Metric | Manual Browser Sweep | Local LLM (Ollama / Llama3-8B) | Focal Harvest |
|---|---|---|---|
| Install Size | ~150 MB (Chrome) | ~4.7 GB (model weights) | < 5.2 MB (zero-dependency core) |
| Cold Startup | ~1.5s | ~12.5s (model load) | < 0.15s (instant TUI boot) |
| Scrape & Clean Speed | ~45s per page | ~1.2s per page (raw HTML) | < 0.25s per page (parallel readability) |
| Context Density | High noise (ads, nav) | Medium (all HTML tags) | High density (top sentences, cleaned) |
| RAM Usage | ~1.2 GB | ~6.5 GB + dedicated GPU | < 85 MB (runs on low-end hardware) |
What These Numbers Mean in Practice
Install size: Focal Harvest's core dependencies are pure Python. The < 5.2 MB figure covers everything needed for a fully functional offline research sweep. The optional curl_cffi package adds a small C-compiled binary but remains well under 20 MB total.
Cold startup: The TUI boots in under 150ms because there is no model loading step. AI synthesis only happens after scraping completes, and only if API keys are configured.
Scrape speed: The < 0.25s per page figure reflects parallel thread execution across multiple URLs simultaneously. Sequential scraping of the same pages takes considerably longer — the parallel crawler is the primary speed driver.
Context density: Manual browser research includes significant noise: navigation menus, cookie banners, ad containers, and tracking scripts. Focal Harvest's hybrid parser strips all of this before passing content to the synthesis stage, which means the AI sees higher signal per token.
RAM usage: The < 85 MB ceiling holds under normal research sweeps. Very large parallel scrapes with many concurrent threads will push this higher temporarily, but Focal Harvest releases memory after each sweep completes.
Hardware Requirements
| Component | Minimum |
|---|---|
| Python | 3.12+ |
| RAM | 256 MB available |
| Disk | 50 MB free (plus report cache) |
| GPU | Not required |
| Internet | Required for live sweeps; cache works offline |
Focal Harvest has been tested on low-spec student laptops running Windows, macOS, and Linux. If you encounter memory pressure on very constrained hardware, reduce max_results in your search configuration to limit the number of concurrent scrape targets.
Where to Go From Here
The benchmarks reflect what the default configuration delivers. See Configuration → for how to tune the pipeline — including cache settings, retry behavior, and the optional curl_cffi bypass that can affect scrape throughput on protected targets.