Source

Personal memory, for your AI.

Your context.
Still yours.

Your preferences. Your projects. The decisions that got you here. Source keeps them in a source-linked memory you can share with the AI tools you approve.

Free during beta · Connect through MCP

Memory, with a source.Illustrative · fictional data
How you work Your context

Keep the first draft short.

Start with three sentences. Add the supporting detail only when it helps.

For the first pass, give me three sentences. I’ll ask for the detail if I need it.

A conversation you chose to keep

The original context stays attached.
Example recall by an approved AI

Here’s a three-sentence first draft, with the detail saved for the next pass.

Writing preferences

Explore the memory and inspect its source.

This example uses fictional memories. It does not access your data or connect an AI.

The cognitive offloading vision · Product direction

From remembering you
to helping you act.

Source is the personal memory foundation for UNBLOCK’s cognitive offloading vision. The CUA we are building uses that context to anticipate tasks, work with your AI and approved apps, and learn from your corrections. Explore the complete loop →

Less repetition. More continuity.

The conversation changes.
Your context carries forward.

01

Bring the history that matters.

Conversations you own become useful context about your preferences, projects, and decisions. A starting point for the next task.

02

Keep the source attached.

A memory should come with its evidence. Inspect the original context, correct the record, and understand what an AI recalled.

03

Decide who gets the context.

Approve access for the AI tools you choose. Keep a boundary around personal context and revoke access when it is no longer needed.

Memory belongs with the person.

Change your AI.
Keep your context.

Your understanding of the world should have a home beyond any one conversation or model.

Your approved context
PreferencesProjectsDecisions
SourceYour personal memory
The AI you choose

A closer look at memory quality

Evidence matters
here, too.

UNBLOCK reports 93% answer accuracy on LongMemEval v1, a benchmark for remembering and reasoning across conversations.

The public reproducibility package is not attached yet. Read the benchmark note for the scope and limitations.

Read the benchmark note

Start with the context you choose.

Bring yourself
to the next conversation.

Connect Source

Desktop downloads are currently unavailable. Use the MCP connection guide to get started.