01
Retain what matters
Save reusable knowledge while skipping temporary, duplicate, or low-value details.
Predictive memory infrastructure for AI
AI learns continuously. Its memory should too.
MemorySafe gives AI a decision layer for what to retain, protect, replay, and forget—before useful knowledge disappears or low-value memory takes over.
Free during beta · macOS, Windows & LinuxModel teams · tell us what you need
02
Keep control visible
Review, protect, and delete memories instead of treating storage as a black box.
03
Govern within limits
Estimate vulnerability, respect bounded memory, and record why each decision was made.
Choose your path
Two ways to use MemorySafe.
Start with the free local beta for AI assistants, or tell us how your model decides which training examples to keep.
A
For people using AI assistants · free local betaSee and control what your assistant saves.
If your assistant saves a project preference, you can review, protect, or delete it on your device. The private beta opens in waves for Codex and Claude workflows on macOS, Windows, and Linux.
Join the free beta list →CL
For teams training models · let's talkWhich examples can your model afford to forget?
As data changes, a limited memory of past training examples can lose rare cases. MemorySafe for Continual Learning helps teams decide which samples stay available for replay. Tell us where this is a problem for you.
Tell us about your model →How MemorySafe works
Make memory a decision.
Memory governance means choosing what to keep or forget and being able to see why. MemorySafe makes those choices before limited capacity is allocated.
01
Assess
Consider value, estimated vulnerability, redundancy, and available capacity.
02
Apply policy
Decide what should be retained, protected, replayed, merged, or forgotten.
03
Record the reason
Keep governance actions visible and auditable instead of silently changing memory.
52-second founder overview
English audio · embedded captions
Research setting · possible applications
Where forgetting
has consequences.
Our continual-learning evaluation uses medical images. Financial models, edge devices, and robotics are possible applications to explore with partners.
Medical AI
Rare clinical cases can be overwritten as models absorb new data.
Prioritize vulnerable, clinically important samples.
Model-risk & retraining
Changing data can make previously reliable financial models drift.
Preserve important evidence across retraining cycles.
Edge AI
Strict storage and compute limits make every retained sample matter.
Allocate bounded memory intentionally.
Robotics
New environments can interfere with previously learned skills.
Protect fragile capabilities during adaptation.Product evidence
See the controls. Examine the research.
The Agent dashboard shows reviewable memory controls. A separate held-out evaluation tests the continual-learning policy.
Memory decisions are visible.
The dashboard shows selection, protected memories, duplicate prevention, and review and deletion controls.
A directional result, with limits.
The pneumonia evaluation was numerically ahead of reservoir on the primary measure, but the difference was not statistically significant.
Read the continual-learning evaluation and methodology
In the held-out pneumonia run, combined AUPRC was 0.702 ± 0.041 for MemorySafe versus 0.687 for reservoir across paired seeds 52–71. The primary comparison was not statistically significant (paired t p = 0.11; Wilcoxon p = 0.15). Rare-class recall was 0.763 versus 0.730 for reservoir.
All policies used the same 500-sample buffer, backbone, schedule, and task order. Only sample selection varied. These results describe the continual-learning kit, not the local Agent beta.
Partners & programs
Member of NVIDIA Inception, Google for Startups Cloud, and AWS Activate. Supported by Startup en Résidence — Desjardins.
awsACTIVATE MEMBER
Benefit & ROI simulator
What could MemorySafe change for your system?
Explore how governed memory can affect time, replay volume, and operating value using your own assumptions.
Simple team example · change any number
Preferences, project details, or recurring instructions.
At 50%, 10 minutes of repeated explanation becomes about 5. Adjust the scenario to match your workflow.
Annual value scenario
Your inputs
Current repetition time / year
—
Time potentially returned / year
—
Illustrative annual value
—
Want to calculate ROI too?
Pricing is not final. ROI appears only when you provide your own estimate.
ROI using your estimate
Illustrative value minus estimated cost
Net annual value
—
Illustrative ROI
—
Simple payback
—
How this works
Team time today = people × weekly minutes × 52. Potential time returned applies the selected scenario. Illustrative value multiplies those returned hours by the hourly value.
Uses your inputs for planning. Actual results depend on workflow, setup, model costs, and answer quality.
Technical planning example · keep any value you do not know
This changes only the use-case wording—not the calculation.
A replay sample is one saved training example reviewed again later; 500 is an example.
Use dataset or storage information; 256 KB is an example.
Check training or deployment logs; 200 is an example.
At 30%, 500 becomes about 350. Adjust this assumption to explore the replay strategy that fits your system.
Annual replay scenario
Your inputs
Modeled replay data not processed / year
—
Modeled replay runtime avoided / year
—
Illustrative gross annual value
—
ROI using your estimate
Gross value minus estimated investment
Net annual value
—
Illustrative ROI
—
Simple payback
—
How this scenario is calculated
Samples not replayed = current buffer − modeled buffer. Annual data applies sample size, runs per day, and 365 days. Runtime assumes duration falls in direct proportion to buffer reduction.
Use this planning model to compare scenarios. Deployment results depend on workload, buffer policy, and model quality.
Local control
Your memory policy stays inspectable.
The governed-memory database is stored locally. MemorySafe is designed to make stored knowledge and governance actions reviewable rather than invisible.
L
Local database
Governed memory records are stored on the user’s device.
C
User controls
Review, protect, and delete memories from the product interface.
A
Auditable reasons
Retain and evict actions can carry a visible reason.
Connected AI services may process selected context under their own terms. “Local” describes the MemorySafe governed-memory database, not every connected service.
Founder & CEO
Carla Centeno
Neuroscience-informed founder building memory infrastructure for AI, inspired by human memory and grounded in continual-learning research.
LinkedIn →Leadership & advisory
Built with operational and startup experience.
Advisors bring financial discipline, company-building strategy, technical perspective, and go-to-market experience.
Strategic Advisor
Ankit Mishra, MBA
Startup operator, AI strategist, and venture-capital professional with 13+ years of experience.
LinkedIn →
Advisor, GTM
Grégoire Gervais-Vachon
Bilingual commercial leader connecting financial-services needs with governed AI-memory solutions.
LinkedIn →
Advisor, operations & finance
Larissa Centeno
Finance leader bringing cost discipline, risk governance, and structured growth strategy.
LinkedIn →
Advisor, project controls & delivery
Grégoire Dubé
Project-controls professional bringing planning, cost discipline, reporting, and delivery governance from complex infrastructure projects.
LinkedIn →
Technical Advisor
Alexandre Rosenfeld
Software architect bringing experience in agentic AI, software engineering, and technical leadership.
LinkedIn →Answers
Frequently Asked Questions
What memory governance means, what the local beta includes, and how model teams can talk with us.
What does memory governance mean?
It means setting rules for what an AI system keeps, being able to inspect the decision, and changing it when needed. In the local Agent beta, you can review, protect, and delete saved memories. For model teams, the focus is which past training examples stay available as learning continues.
Is the local MemorySafe beta free?
Yes. The local Agent version is free during beta. Continual-learning pilots are a separate conversation with model teams.
Which systems and AI tools is it for?
The private beta is opening in waves for macOS, Windows, and Linux, with Codex and Claude workflows. Joining the list requests an invitation; availability may vary by platform and integration.
What stays on my device?
The governed-memory database is stored locally. Connected AI services may process selected context under their own terms; “local” does not mean every connected service runs on your device.
Can I control what MemorySafe remembers?
The product interface is designed to let you review, protect, and delete stored memories and inspect governance decisions.
Can we explore a model-memory use case?
Yes. If you train a model on changing data, tell us which past examples are hardest to keep and why they matter. We can then decide together whether a pilot would answer that question. This is separate from the free Agent beta.
Choose your path
Start with the free local beta.
Using Codex or Claude? Join the Agent beta for macOS, Windows, and Linux. Training a model on changing data? Tell us what it needs to keep and why.