BEFORE · START EMPTY
Blank task
0 relevant memories · empty by design
“What must I know before changing the billing migration?”
Memory for the next task
The next task starts with the right past.
An open-source memory layer for AI agents and engineering teams.
Save one decision. In a later task, Vault finds what matters and prepares bounded context the next task can use.
REAL VAULT ENGINE · bounded fixture writes · real recall/context · browser-scoped private demo · no AI model · no production data
LIVE PROOF / MEMORY LOOPwaiting for a story
AFTER · USE WHAT FITS
Relevant context
No relevant memories yet
Run the story to watch the next task get a head start.
Save → retrieve → prepare contextPress P for presentation mode
Open sourceReal Vault enginePrivate demo sessionNo production data
One guided story
Watch memory change the next task.
Start empty, stage three bounded illustrative fixtures, retrieve only what fits, and leave with a short summary the next task can actually use.
01 · START
Blank slate
Nothing saved yet
02 · SAVE
Save what matters
0 bounded fixtures staged in this session
03 · FIND
What Vault found
Ask what you are about to do
04 · USE
Ready-to-use summary
Your short summary will appear here
Every run starts with a clean, private demo session: START begins empty, SAVE stages three bounded labeled fixtures through Vault's trusted seed path, FIND retrieves only what fits, and USE prepares the real bounded artifact.
Step 1 · save a memory
Save something worth remembering.
The conference story uses three bounded labeled fixtures through Vault's trusted seed path. FIND and USE call the real Vault MCP service; manual unapproved writes stay fail-closed.
Manual writes require admission evidence
Saved in this session · bounded fixtures0
Step 2 · use it
Start with the right information.
The payoff is not “search succeeded.” It is the relevant past arriving before the next task repeats the work.
Memories that fit
Run the demo to see the right memory appear here.
Ready-to-use summary
Run the demo to see a short summary you can use to start the task.
The observable change
The next task gets a head start.
Before, this task had nothing saved. Now it starts with only the memories that fit the question.
0 found0 staged0 used0 chars ready
Inspect this run
Not ready
0
fixtures staged
0
memories found
0
summaries prepared
0
times marked useful
Trust without theater
The demo shows the memory loop — not a made-up business result.
This page uses the real Vault engine with browser-scoped illustrative inputs. The guided story writes only a bounded, labeled fixture set through Vault's explicit trusted seed path; recall and context remain real Vault operations. It does not call an AI model, use production data, or claim provider savings.
What is real here?
Real: Vault's bounded trusted seed, recall, context, and feedback responses from the hosted demo service.
Seeded: the example query and three memory texts supplied by this browser story.
Not demonstrated: agent action, customer outcome, provider-billed savings, or a production audit receipt.
How teams can measure value later
This page demonstrates the memory loop—not a promise about customer savings.
ONLY IN A REAL DEPLOYMENT · Not generated by this demo.
In a real deployment, Perseus Ledger can record provider usage and the prompt Vault replaced. That comparison belongs to a separate evidence path with its coverage and method shown.
About this demolive · private demo
where it runsHosted demo
what powers itReal Vault engine
your dataPrivate to this browser
extra proofOptional
You are using the real Vault engine in a private demo session. Nothing here touches production data, and the demo does not call an AI model.