Local Ollama runtime
Discover compatible installed models, stream thinking and answers, and keep model inference out of a hosted AI account.
Native macOS app · Local Ollama models · Open source
A local AI agent that can reason over your documents, run supervised terminal work, use native Mac context, and operate a real browser from screenshots.
PrivateAI connects compatible Ollama models to bounded, inspectable macOS capabilities. Model inference and conversation history stay on your Mac; public services are contacted only when a requested Tool needs them.
Discover compatible installed models, stream thinking and answers, and keep model inference out of a hosted AI account.
Use a live WebKit browser, rendered screenshots, frame-bound coordinates, and closed-loop actions for modern public websites.
Read searchable PDFs, Markdown, text, structured data, HTML, and source files with bounded access and resumable analysis.
Run visible zsh commands, builds, tests, scripts, and diagnostics with output limits, timeout, cancellation, and process cleanup.
Use Apple frameworks for device, locale, storage, network, location, calendar, reminders, contacts, and notifications when authorized.
Keep local SwiftData history in a macOS interface with Markdown, syntax highlighting, tables, math rendering, and live activity.
Each guide describes the production capability, evidence boundary, privacy behavior, and current limitations for a concrete class of work.
The repository separates schemas, direct executors, integration checks, and model-driven acceptance evidence. Capability claims are tied to current implementation, not roadmap documents.