MS Duplex — The Agent Operating System powered by Runtime Intelligence.
Rent the model. Own the intelligence.
Give your agents a goal. MS Duplex plans the work, weighs the options, decides within your rules or asks a person first, and acts through your tools, checking every model call and tool action before it runs. Long work keeps its course through pauses and change, every result is checked against the goal, and what MS Duplex learns is yours, in every session, agent and model.
* Figures from MatterSpace’s internal benchmarks, not independently verified. Read the disclaimer
- Rented intelligenceOwned intelligence
Your AI is rented by the call. What MS Duplex learns is yours. So are the decisions.
- Fixed before deploymentLearning, planning, simulating, deciding and acting while you work
It learns, plans, simulates, decides and acts at runtime, and carries the work forward.
- One model on its ownTwo intelligences, one system
Your AI brings general intelligence. An ACI model learns fast, plans, simulates, decides and acts. One system: MS Duplex.
Everything a goal needs, working as one.
Planning, simulation, problem solving, decisions, your rules and approvals, the work itself, checks, context, long-running work, recovery and learning are not separate tools to wire together. MS Duplex coordinates them around your goal. Pick an area to see what it gives the goal, or let the example play through every one.
Example goalGet a delayed order to the customer this week, within your delivery and refund rules.
Before anything runs
While it runs
Over time
Context
What it gives the goal
Every step starts from the facts, the history and the decisions that matter.
In the example
It brings in the order, its history, your delivery rules and the courier’s latest update.
Where it acts in the journey
- The goal
- The work
- A change
Agents do the work.You set the rules.
MS Duplex puts your agents to work in the real world, for you and your whole enterprise. You set what matters: the goals, the rules and the actions that need your OK. MS Duplex handles the routine inside those rules and holds every agent to them.
Decide once. Every agent follows, every session, everywhere it runs.
- 01
Checked before it runs
Every model call and tool action runs only after MS Duplex checks it against your rules.
- 02
Your OK where it counts
You choose the actions that need a person’s approval, and MS Duplex asks first.
- 03
Decisions that stick
What your team approved carries into every session, every agent and every hand-off.
- 04
Learning on your terms
Nothing MS Duplex learns goes live until you say so.
What MS Duplex does for your AI.
24 ways it makes your AI better. Watch them play, or press any one.
Every conversation becomes intelligence your company owns.
The answers, fixes and decisions your AI works out stay with your company, and every task makes the next one better.
See it at work →Step 1 of 4: Confirmed
Confirm a result, and MS Duplex learns it on the spot.
How MS Duplex works as the Agent Operating System.
MS Duplex is the Agent Operating System powered by Runtime Intelligence: the layer every agent runs on, whichever model it uses. Agents now work for hours or days, across sessions, apps and models, but each one still builds its own memory, rules and tool connections, and loses them when the model or the app changes. MS Duplex gives them one layer for all of it, so every agent keeps its context, rules and tools whichever model or app it uses.
Step 1 of 9: One layer
01 · One layer
MS Duplex — The Agent Operating System powered by Runtime Intelligence.
Every agent runs on it, whichever model it uses. Rent the model. Own the intelligence.
What MS Duplex gives every agent, whichever model it uses.
- Context
Context that outlives the session
The goal, the decisions and the unfinished work carry into the next session, agent or model.
- Rules
Rules before every action
Every model call and tool action is checked against your rules before it runs, and the actions you choose wait for a person’s OK.
- Long work
Long work that carries on
Work that spans days, sessions and agents stays one job, with every step recorded and a checked finish.
- Your stack
Your stack, one toolset
Your retrieval, caches, knowledge graphs, APIs and models become one toolset for every agent, and permission stays with your controls.
- Scopes
Each scope kept apart
What a person, a team and the organisation learn stays in its own layer, and each customer stays separate.
- History
One history
Every step leaves a receipt, in one history you can check and export.
- Any model
Models you can change
Change the model, the client or the vendor, and what your agents learned stays.
- Powered by Runtime Intelligence
Every goal, coordinated
It plans, simulates and decides each step within your rules, acts through your tools, checks the result and learns from what was verified, so every agent on it gets the work done and gets better with use.
How it works.
Start with one request. MS Duplex sees it before your model and before any tool, and does one of three things: it answers what it has already learned, brings in your model for what is new, or declines what breaks your rules. Every result comes back to MS Duplex, so the next request like it can be answered without a model call.
- 01
It sees every request first
Before your model or any tool, MS Duplex checks the request against what it has learned and against your rules.
- 02
It decides who answers
It answers what it is sure of, brings in your model, briefed, for what is new, or declines what breaks your rules.
- 03
It learns from the result
Each result comes back to it, and what your team confirms serves the next request like it, without a model call.
- 04
It carries it forward
What it learned carries into every session, agent and model you allow, and every step is on record.
Underneath is ACI, Adaptive Continual Intelligence: a new category of intelligence that works in tandem with your model.
Step 1 of 6: What ACI is
Plasticity. Stability. Editability.
ACI is the result of years of research into neural nets and their trilemma: plasticity, stability and editability. Learn fast, keep what it knows, or let you change what it learned: a neural net gets two at best. ACI holds all three.
Plasticity: a neural net that learns fast overwrites what it knew. Researchers call it catastrophic forgetting.
Step 1 of 5: Plasticity
A persistent learning runtime for AI agents.
Today’s AI is fragmented: the model, retrieval and application code each hold a piece of the intelligence, and none learns from the work. MS Duplex is the infrastructure layer that binds them into one system that learns, plans, simulates, decides and acts. The centre of gravity moves to the whole system: the model can change, and MS Duplex carries the rest forward.
Step 1 of 4: The model
The model reasons about the task, and forgets it when the task is done.
MS Duplex makes AI agents remember, learn, plan, simulate, decide, improve, obey and continue, across sessions and models.
It combines parts of a vector database, a memory plugin, a prompt manager, an MCP server, a workflow engine, a policy gateway and a fine-tuning system into one runtime that learns.
For business
AI that improves through use, and keeps what it learned across agents and model vendors.
For research
Continual adaptation, uncertainty, memory, planning, reversible learning and evidence from every medium, in one system working behind real agents.
For the field
Intelligence no longer has to live entirely inside the model. What is learned can become a separate layer: governed, portable and yours.
How MS Duplex is different.
A harness gives your AI tools, prompts and steps. MS Duplex pairs it with an ACI model, a second intelligence that learns, plans, simulates, decides and acts at runtime, and works with the assistants and agents you already use.
On a new task, MS Duplex starts with what your team has already confirmed.
Step 1 of 6: A new task
Intelligence, not retrieval.
RAG, vector databases, knowledge graphs, caches and memory store what happened and find what looks like it. MS Duplex learns: it understands the pattern behind each result, compresses it into what generalises, predicts what comes next, and acts on it.
When a request only looks familiar, retrieval returns the nearest match. MS Duplex understands the pattern, and sees the difference.
Step 1 of 5: A look-alike
- 01
It understands
The pattern behind a result: why it worked, and where it applies.
- 02
It compresses
What it learned into what generalises, so it gets faster and leaner with use.
- 03
It predicts
What comes next, from everything your work has taught it.
- 04
It acts
On what it learned, at runtime, within your rules.
Intelligence, engineered for everyday work.
Start with one person and grow to the whole enterprise: it is the same MS Duplex all the way.
For you, MS Duplex learns how you work, in the assistant you already use.
In Claude, Codex, Cursor or Hermes. What it learns is yours, and it goes where you go.
Step 1 of 5: For you
In Claude, Codex, Cursor or Hermes. What it learns is yours, and it goes where you go.
What it learns is yours, wherever you work.
Confirm a fix in Cursor, and MS Duplex learns it.
Step 1 of 4: A fix in Cursor
It learns, plans, simulates, decides and acts, at runtime.
Pretrain MS Duplex, and it starts informed.
Step 1 of 5: Pretrain it
Two intelligences, one system.
Your AI answers, and learns nothing from the work.
Step 1 of 4: Work arrives
More accurate, with far less model work.
3,000 decisions, with five rule changes along the way: MS Duplex kept improving, and called your AI less and less.
See the results →- Accurate over 3,000 decisions
- 94.6%*
- The model alone 59.8% · with a cache 77.3%
Accurate over 3,000 decisions The model with MS Duplex 94.6% The model with a cache 77.3% The model alone 59.8% - Model calls, with accuracy up
- −90.3%*
- Against calling the model every time
Model calls in 3,000 decisions The model with MS Duplex 277 The model alone 2,864 - A decision MS Duplex has learned
- ≈97 ms*
- A model call: ≈2,231 ms
A decision MS Duplex has learned A decision MS Duplex has learned ≈97 ms A model call ≈2,231 ms
* Figures from MatterSpace’s internal benchmarks, not independently verified. Read the disclaimer
Continual learning is the holy grail of AI.
Every model stops learning when its training ends. Learning from every result without losing what it knew is the prize the field has chased for decades. MS Duplex is built on ACI, Adaptive Continual Intelligence: MatterSpace’s proprietary algorithms and architecture for intelligence that keeps learning at runtime.
Your team fixes a wrong answer on Monday.
Step 1 of 5: A fix
Own it as a person. Own it as a company.
One MS Duplex, two ways in: inside the assistants your team uses, and inside the agents your company builds, through MCP + hooks.
In your assistants
For you and your team- Works
- Inside Claude, Codex, Cursor and Hermes
- Reaches you through
- MCP
- Builds
- Your personal intelligence
Your assistant sees the input first and brings MS Duplex in at each step.
In your own agents
For your company- Works
- Inside every agent it builds
- Reaches you through
- MCP + hooks
- Builds
- Your company intelligence
Your agents stay in charge, and the hooks bring MS Duplex in at every step, before each action runs.
You start with a pilot on your own work, with the measures agreed before it begins.Every integration →