Agentic research
Type a question instead of a URL. Piri Reis decomposes it into focused sub-queries, fans out across the web, reads the actual pages (not just snippets), and writes a clean, sourced report.
Powered by Guidance Intelligence
Piri Reis is an agentic browser. Ask a question, it plans the research, reads the sources, and writes a cited report — while you stay in control of where the work happens: cloud, your own machine, or both.
Three things ordinary browsers don't do.
Type a question instead of a URL. Piri Reis decomposes it into focused sub-queries, fans out across the web, reads the actual pages (not just snippets), and writes a clean, sourced report.
JavaScript-heavy sites, paywalls, news articles — Piri Reis renders them in a hidden Chromium view and extracts the meaningful content with Mozilla Readability. No more "preview text" answers.
Every deep-research report builds a radial concept map of how ideas connect across the sources. Click any node to research it. The map is extracted from the actual report, not made up.
The same five steps that a careful human researcher would take — just automated.
The planner breaks your question into 5–8 focused sub-queries covering the angles that matter.
Brave Search runs every sub-query in parallel and the top results are de-duplicated, capped per domain, and prioritized for quality.
Each page is fetched and read in a real browser engine, with a fallback to a hidden Chromium view for JS-rendered sites.
The model writes a structured, multi-section report from the full evidence — citing every claim, never inventing sources.
A radial concept graph is extracted from the report and the sources, ready to explore further with one click.
Three honest engine modes. Switch them per query, no restart.
Fast, best quality
Synthesis on your machine, planning on cloud
Fully offline. Nothing leaves the machine.
Cloud is the default and the floor. Local modes are an opt-in layer that arrives in the background and switches instantly when you want it. Honest performance estimates on every model card, based on your actual hardware.
Piri Reis detects your RAM, GPU, and disk on launch, then ranks the local models that actually fit your machine — not a hardcoded menu. A 4 GB-VRAM laptop sees one list, a workstation sees another. Speeds are shown for your hardware, never an optimistic cloud-vendor benchmark.
Models discovered, ranked, and installed from a curated registry (Qwen, Llama, Mistral, Phi, Gemma). Pause, stop, resume, uninstall — all of it auditable, all of it local. Cloud queries keep working at full speed during a download.
Strong Turkish and English support at every tier. Sub-queries, synthesis, and source reading work across languages — no separate "translate first" step.
Open beta. Windows installer first; macOS and Linux follow.
Cloud mode needs an Anthropic API key and a Brave Search key — both set inside the app, encrypted on disk. Local mode just needs disk space.