Shared temporal memory

A little context. A longer memory.

Give products and agents memory they can trace, question, and revisit in time.

Starting with Skynet. Access by pilot enrollment.

Illustrative timelineSynthetic example

Same moment.
A different view.

A person’s cityBostonSource-backed candidate · Example only
What had been recorded by…

At time 70, viewed with what was known at 150: Boston. The correction had not been recorded.

Meet the two clocks
Context, with a past.

What applied then.
What was known then.

See how it works

Keep the whole story

Useful memory.
Room for a second thought.

Sources, boundaries, and time give an agent something better to work with.

01Remember what changed. And when.
Ask what applied at a moment with as_of, and what had been recorded by a moment with known_at. A correction learned today can change what you now know about last month without rewriting what you knew last month.
02Context with something behind it.
Keep the exact claim, its sources and its trust status together. New observations are candidates. Verification requires a separate authorized attestation; a stored memory is not automatically true.
03A shared memory. Clear boundaries.
Workspaces and explicit namespace grants determine who can read, contribute, verify, correct and forget. Agents receive a bounded context, with source references and visible retrieval limitations.
04Built for useful recall.
The temporal ledger and private evidence are authoritative. Hybrid search is designed as a rebuildable projection. High-volume events, telemetry and product working state stay in their own stores.

For products & agents

One memory.
A few ways in.

Use the workspace to inspect it. Use REST or MCP to bring bounded context into your own product.

Explore the docs
Two clocks, one questionIllustrative query
as_of:   70
known_at: 150

→ Boston

as_of:   70
known_at: 200

→ New York
Understand the result