MS Duplex is your AI model and an ACI model, working as one.
The combination is the product: runtime intelligence for your AI and your agents, for one person or the whole enterprise. Together they learn from every result, plan, simulate and decide each step, act within your rules, acquire skills and carry what they learn across sessions, agents and models. What it builds is yours.
A persistent learning runtime for AI agents.
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.
Your AI
What your people and agents work with
The foundation model
Can change: a new version, a new vendor
What MS Duplex has learned
Stays: yours, and still growing
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.
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.
Built on ACI.
MS Duplex is built on ACI, Adaptive Continual Intelligence: a new category of intelligence that learns, plans, simulates, decides and acts at runtime, in tandem with AI models. With ACI, your AI learns on the spot, keeps up with every change and gets faster and cheaper with use, and what it learns is yours.
MS Duplex
The product you useMS Duplex works inside the assistants your team uses and inside the agents your company builds. It starts informed, decides the route, acts within your rules and carries its decisions, its learning and your context forward.
- Inside your assistants
- Inside your own agents
- Within your rules
- In the placement you choose
ACI
The intelligence in MS DuplexIt learns, plans, simulates, decides and acts while the work happens, leaves your model’s weights alone, and everything it learns belongs to you.
- Learns from every result
- Acts within your rules
- Keeps up with every change
- Yours, wherever you work
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.
Continual learning is the holy grail of AI.
MS Duplex is built on ACI, Adaptive Continual Intelligence: MatterSpace’s proprietary algorithms and architecture for intelligence that keeps learning at runtime. It learns from every confirmed result, still knows what it learned before, and lets you change any piece of it: plasticity, stability and editability, all at once.
MS Duplex learns from every confirmed result, and builds on it.
Step 1 of 5: It learns
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
A look-alike
Retrieval Returns what looks similar, right or wrong.
MS Duplex Understands the pattern, and sees the difference.
New words
Retrieval Finds nothing close enough.
MS Duplex Learned the pattern, so it still knows.
A rule changes
Retrieval Serves the old answer until someone edits the store.
MS Duplex Follows the new rule from the next request.
With use
Retrieval Only grows, and every search gets bigger.
MS Duplex Compresses what it learned, and gets faster and cheaper.
What comes next
Retrieval Waits to be asked.
MS Duplex Predicts the next step, and acts on it within your rules.
- RAG Finds passages that look like the question
- Vector databases Find close matches
- Knowledge graphs Find what someone linked
- Caches Return what was asked before
- Memory Recalls what was saved
MS Duplex can use every one of them as a tool, and it learns from what they return.
For many tasks, you no longer need to fine-tune.
Teams fine-tune for their knowledge, their way of working and better results. MS Duplex brings all three from your own work, and your model stays as it is.
- Your terms and knowledge, learned from results your team confirms
- Your format, tone and rules, on every answer
- Better results: it beat tuned fine-tuning in 6 / 6 tests*
- Lower cost: 90.3% fewer model calls*
- A change in your work, live by the next request
- 0 retraining when you change models*
- Any lesson taken back on its own
- No training runs, and no GPU of its own
A fine-tune rewrites the weights, one model and one training run at a time.
Step 1 of 4: A training run
It learns, plans, simulates, decides and acts, at runtime.
An ACI model beside your AI gives it what it lacks while the work happens. Follow one job through everything MS Duplex does.
Step 1 of 5: Learns
A job arrives that needs a skill MS Duplex does not have yet, so it learns it where it is safe to try, and checks it before the job relies on it.
114 capabilities in 17 groups.
From the context a task starts with to an advantage that compounds with every task. Press a group to open it in the list.
- Context and observation
- Rules and behavioural control
- Tool execution
- Learning and personalisation
- Long Horizon: long-running work
- Planning and decision support
- Skill Fabric: reusable skills
- Persistent threads and sharing
- Forgetting and data control
- Status and explanations
- Persistent intelligence
- Personalisation
- Multimodal intelligence
- Confidence and autonomy
- Runtime safety and oversight
- Operational intelligence
- Integration and portability
Faster, cheaper, more accurate.
Of every 100 decisions, how many each approach got right, before and after a rule change.
- −90.3%*Model calls, with accuracy up →
- 0%*Stale answers after a rule change →
- 6 / 6*Tests where MS Duplex beat fine-tuning →
- 3 / 3*Frontier models beaten at predicting system changes →
* Figures from MatterSpace’s internal benchmarks, not independently verified. Read the disclaimer
Every result →Step 1 of 4: The model alone
Called on every decision, the model alone got 59.8% of them right.*
The work it changes.
Each kind of work, on the same Agent Operating System, and what changes once MS Duplex learns it.
Step 1 of 6: Coding
Coding and desktop assistants: Your project’s conventions, applied in every session.
Coding and desktop assistants
Your project’s conventions, applied in every session.
MS Duplex applies your project’s conventions and confirmed fixes in new files and new sessions, inside Claude, Codex, Cursor and Hermes.
See the plugins →Support and operations agents
A policy change takes effect on the next request.
Returns, routing and exceptions are answered in milliseconds, and a changed policy applies on the next request.
See the agent benchmark →Enterprise workflow automation
Agents that know how your systems respond.
MS Duplex learns what each action changes and plans the route. It beat three frontier models at predicting system changes.
See the workflow benchmark →High-volume AI products
90% fewer model calls, with accuracy up from 59.8% to 94.6%.*
MS Duplex answers the decisions it has learned, in milliseconds, and your AI reasons through the rest.
See the cost results →Teams adapting a model
Adapt your model without retraining it.
MS Duplex fits your model to your work without retraining it, and the fit moves with you to the next model.
See the adaptation results →Regulated and sovereign work
Learning you can inspect, correct and keep in your boundary.
Each customer owns what MS Duplex learns for it, with central rules, approvals and an exportable history, in your cloud.
See security →
