PrivyAgent
Get started
A local-first workspace for macOS

Your projects. Your models.
Your own AI workspace.

Bring your files, knowledge, and AI together. Let agents take the next step, with your work on your Mac and you in control.

Explore PrivyAgent Discover the features

Built for macOS 14 and later.

YOUR WORK, WITH CONTEXTILLUSTRATIVE WORKSPACE

Research projectBuilt-in · On-device
Review my project notes and draft a concise summary.

I’ll find the relevant sources, then prepare a draft for your review.

Searched project knowledge
Read workspace notes
Review before writing
+ summary.mdWaiting for your approval
On-device models Project-based context Approval before changes
A place for the whole task

From a conversation
to work that moves forward.

The context of your project. The tools to act. A clear view of every step.

A workspace for every project

Keep files, conversations, and knowledge together. Each project has its own workspace and context, so your work stays organized.

Project isolation · File import · Conversation history

An agent that works with you

Ask, reason, and work on files in one conversation. Follow each tool call in a live timeline, stop a run, and resume interrupted work.

Streaming replies · Tool timeline · Recovery

Answers grounded in your knowledge

Import text and text-based PDFs. Search with keywords and optional embeddings, then follow citations back to the source.

Source citations · Search · Project memory

Turn repeat work into workflows

Start with a template or import your own. Run independent steps in parallel and resume unfinished steps from checkpoints.

Templates · Checkpoints · Schedules

The right agent for the job

Use General, Researcher, or Editor profiles. Choose a model for each role and follow handoffs in the timeline, with tool limits enforced.

Specialist profiles · Model choice · Tool boundaries

Connect tools on your terms

Add Streamable HTTP MCP servers and enable them per project. Review the destination and arguments before approving each call.

MCP tools · Per-project access · Signed extensions

Local by default. Open by choice.

Your model.
Your way of working.

Run a model directly inside PrivyAgent, connect a local model service, or choose a remote provider. Change your setup without leaving your workspace.

Find your starting point

Built right in

Download and run Qwen3.5 4B or 9B with the built-in llama.cpp engine.

Your local services

Connect Ollama, LocalEngine, or LM Studio.

A provider of your choice

Use OpenAI or an OpenAI-compatible endpoint with your own credentials.

Built-in inference stays on your Mac. Remote providers receive the context sent to them.

Useful autonomy. Clear boundaries.

Let your agent work.
Keep the final say.

Local-first is more than where a model runs. It means knowing what the agent can access, what it wants to change, and what you choose to remember.

  • See the change before it happens

    File writes show a diff and require approval. Deletion is off by default. Scheduled tasks never approve actions for you.

  • Keep context within the project

    Workspace tools stay within the project directory. Knowledge and long-term memory are isolated per project.

  • Decide what gets remembered

    Automatic conversation memory is off by default. Agent memory saves need approval, and you can inspect or delete saved memories.

  • Keep credentials in Keychain

    Provider and MCP credentials stay in macOS Keychain. Project exports exclude credentials.

A few things to know

A clearer starting point.

Which platforms does PrivyAgent support?

PrivyAgent currently supports macOS only, with macOS 14 or later required. There are no Windows, Linux, iPhone, or iPad editions at this time.

Can I work entirely offline?

After downloading a built-in model, on-device inference can run offline. Remote providers, model downloads, web search, and external MCP services require a network connection.

What can I add to the knowledge base?

Import UTF-8 text files and PDFs with a text layer. Scanned PDFs need OCR first. Keyword search works locally; semantic search requires a configured embedding model.

Do scheduled workflows run when the app is closed?

No. Scheduled workflows run while PrivyAgent is running. You can choose to catch up on a missed run when the app starts. Interrupted workflows can be resumed manually.

Does local-first mean nothing ever leaves my Mac?

Built-in model inference runs on your Mac. If you choose remote providers or external tools, the context and arguments needed for those requests are sent to the selected service.

Your next project starts here

Make room for a different
way to work.

One Mac. Your context. An agent ready for the next step.

Get to know PrivyAgent