OwnSafeAI Explore the flow
OwnSafeAI setup and Today planning workspace

Local-first desktop assistant

Your plans.
Your model.
Your machine.

OwnSafeAI turns scattered thoughts into reviewable tasks and realistic daily plans while keeping personal content on your computer.

See how it works

Desktop alpha · Apple Silicon first · Manual mode always available

Local inferenceSmall models run on hardware you already own.
Review before actionAI proposes. You edit and confirm.
Useful without AITasks and planning remain accessible in manual mode.

01 / Product walkthrough

From a loose thought to a plan you can trust.

OwnSafeAI is not another empty chat window. It is a focused workspace where model output becomes an editable proposal before it touches your saved tasks.

Step 1

Start with the machine in front of you.

OwnSafeAI detects local hardware and presents compatible model profiles with download and memory estimates. Choose a local model, connect an existing Ollama installation, or continue without AI.

  • Hardware-aware recommendations
  • Verified model downloads
  • No silent cloud fallback
Model selection beside the OwnSafeAI Today workspace
Setup and Today. Design reference; the desktop alpha implements this core flow.

Step 2

Capture naturally. Save deliberately.

Type a reminder or task in plain language. The assistant extracts the useful structure, then shows a familiar form with uncertainty called out before anything is saved.

  • Tasks, reminders, and notes
  • Ambiguity stays visible
  • Canonical records remain editable
OwnSafeAI Inbox and reminder review dialog
Inbox and proposal review. The model never gets the final say.

Step 3

Shape a day, not a generated essay.

Build a daily plan from due work, priorities, and available time. Reorder, pin, defer, or adjust durations before accepting it. The explanation stays separate from your real task data.

  • Reviewable daily plan
  • Visible tradeoffs and deferred work
  • Model and memory controls nearby
OwnSafeAI daily plan and local model controls
Plan review and model management. Quiet controls for repeated, practical work.

02 / The loop

A short path from intent to follow-through.

  1. 01Capture

    Write what is on your mind.

  2. 02Interpret

    Local inference proposes structure.

  3. 03Review

    Correct dates, details, and ambiguity.

  4. 04Plan

    Fit important work into real time.

  5. 05Follow through

    Complete, defer, or reschedule.

03 / Trust boundary

Private by architecture, not by a toggle buried in settings.

Available in the alpha

  • Local SQLite workspace data
  • Built-in llama.cpp runtime on Apple Silicon
  • Optional loopback-only Ollama models
  • Streaming responses, Stop, and idle unload
  • Manual task and project workflows

Permissioned work ahead

  • Web research with per-request approval and citations
  • Selected-folder analysis with read-only grants
  • Dependable native reminder scheduling
  • Windows and Nvidia runtime packaging
  • Additional focused local agents

Today, the local chat does not browse the web, inspect files, run shell commands, or silently perform external actions.

In active development

The foundation is running. The safety boundaries come next.

The current desktop alpha includes hardware detection, verified model installation, local chat through built-in GGUF models or Ollama, persisted tasks and projects, cancellable streaming, and model idle unload. Public downloads are not open yet.