Skip to content
Advancing Bits to Master Atoms

Research memory across sessions.
Every session starts from the current state of the work.

MS Duplex carries the goal, the decisions, the approved methods and the progress into every new session. Your agent picks up where the team left off, with every source on record.

Remember and govern · one of the fourteen capabilities of MS Duplex

An open graphite case filed with rows of thin satin-silver cards, one cobalt-blue divider card raised above the rest.An open graphite case filed with rows of thin satin-silver cards, one cobalt-blue divider card raised above the rest.

Your team gets

  • The AI already knows the project
  • Long projects that keep their thread
  • Context focused on the current task

What research memory means for your team.

Research memory means the AI your team uses remembers the work, across every session. The goal, the decisions made so far, the methods your team approved and the progress on each step carry into the next session as a compact picture of where things stand.

That picture is kept current. When a decision is made or a new result replaces an older one, the memory moves on with it, and the newest confirmed value is the one your agent uses. The history of what changed stays on record, so nothing is lost when something is updated.

The memory travels with the work. A new session, a colleague’s authorized agent or a different model starts from the same picture, so the project keeps its thread through pauses, hand-offs and model changes, without replaying a transcript or asking the same questions again.

Why research needs memory that carries across sessions.

Research projects last longer than any single session. An AI that forgets between sessions leaves the team to carry the thread by hand.

A research project builds up over many sessions: decisions made for a reason, methods approved after review, results that changed the plan. When each session starts from zero, scientists and engineers spend the opening of every session re-explaining the project, and the agent asks questions the team answered long ago.

Pasted transcripts and long notes files carry the words of past sessions and leave the agent to work out which parts still hold. A result that was replaced can come back as if it were current, and the decisions that matter today get lost among everything else that was said.

  • Projects outlast sessions

    Work that spans many sessions needs the goal and the progress in view every time someone picks it up.

  • Re-explaining costs expert time

    Every session spent restating the project is time your scientists and engineers could have spent on the research.

  • Facts move on

    A newer result replaces an older one, and the next session has to use the new one.

  • Focus matters

    A session works best from the state of the work today, with older detail kept on record and out of the way.

Three graphite trays in a row, each holding the same arrangement of silver tokens, with a cobalt-blue token in the same place in every tray.Three graphite trays in a row, each holding the same arrangement of silver tokens, with a cobalt-blue token in the same place in every tray.

How research memory works in your team’s day.

Your team works across as many sessions as the project needs. MS Duplex keeps the thread between them.

  1. 1

    The work moves on

    In each session your agent makes decisions, follows approved methods and records progress, with the tools your team already uses.

  2. 2

    The state is kept

    MS Duplex keeps the goal, the decisions, the approved methods and the progress, each with its source.

  3. 3

    A new session begins

    The next session, on any agent you authorize and any model, starts from a compact picture of where the work stands.

  4. 4

    The thread continues

    Your agent picks up from that picture, without replaying a transcript or asking the same questions again.

An example, in any field where work runs over many sessions.

A team’s agent plans a series of runs toward a design goal. Different people pick the work up on different days, and it keeps its thread.

  1. 1

    The first session

    A researcher sets the goal, agrees on an approach with the agent and approves the method for the first runs. MS Duplex keeps all three, each with its source.

  2. 2

    A changed result

    Later a run comes back that replaces an earlier estimate. The memory moves on to the new value, and the history shows when and why it changed.

  3. 3

    A hand-off

    A colleague opens a new session on a different model. The agent already knows the goal, the approved method, the runs completed and the new value, and asks nothing the team has already answered.

  4. 4

    Back after a pause

    The first researcher returns after a break. The session opens on the current state of the work, focused on the next decision, with every earlier step on record.

Research memory carries the other capabilities from session to session.

What the rest of the layer learns and plans is what each new session starts from.

  • Long-horizon planning

    The plan, its checkpoints and its progress carry into every session, so a long campaign stays one job.

  • Continual learning

    What the team confirms is learned at once, so the next session starts with it already in place.

  • Freedom to switch models

    The memory lives in MS Duplex, so a new model or agent starts from the same state of the work.

  • Personal, team and organization layers

    Each researcher, each team and the organization has its own layer, and your rules decide which memory each session can reach.

What your team gets.

Every session starts where the work stands, with every source in view.

stale answers after facts changed: MS Duplex beside standard retrieval, in MatterSpace’s controlled tests
0% vs 57%
  • Nothing to re-explain

    A new session already knows the goal, the decisions and the progress, so nobody has to restate the project to the AI.

  • Long projects that keep their thread

    Work that spans many sessions, people and agents stays one piece of work.

  • Focused context

    Each session starts from a compact picture of what matters now, with older detail kept on record.

  • Current facts

    When a confirmed result replaces an older one, the newest applies and the history of the change is kept.

  • Smooth hand-offs

    A colleague’s authorized agent picks up from the same state of the work as yours.

  • Every source on record

    Each part of the memory shows where it came from, so anyone can check what the next step relies on.

Common questions.

What does MS Duplex carry into a new session?

The goal, the decisions made so far, the approved methods and the progress, as a compact picture of the work. Every part of it keeps its source, so your team can see where each item came from.

How is research memory different from a memory file or retrieval?

A memory file holds what someone wrote down, and retrieval finds passages that look similar. MS Duplex keeps confirmed outcomes, methods and decisions with their evidence, checks whether the conditions match before applying them and keeps them current as facts change.

What happens when a fact changes?

The newest confirmed value applies from the next request, and the history of the change is kept. Your agent moves on with the work and stops quoting the old value.

Can a colleague or another model pick up the work?

Yes. The agents you authorize start from the same state of the work, on whichever model they use. Your rules decide which layer of memory each person and agent can reach.

Make the next run benefit from the last.

Bring us the research your AI should get better at. We will plan a pilot on your own workflows, with your experts as the judges.