Science & research
Founded on
neuroscience.
Keepsake started with two researchers who'd been studying memory failure for years. They thought the problem was solvable, so they built a solution. The company came after, because they believed right to remember your own life needed to exist outside a research setting.
The science of
forgetting.
Your brain doesn't store memories the way your computer stores files. When you remember something, you're rebuilding it by seeing smaller fragments, filling in gaps, and adjusting the emotional tone to fit how you feel now. In neuroscience, it is well established that every time you recall an event, you're also changing it slightly. "Memory consolidation rewrites as it saves."
Memories then drift, and conversation shift. The sequence of events gets rearranged. Something that felt minor at the time can become formative years later, or the other way around, all depending on what you've needed that memory to do for you. This is how our brains have worked for hundreds of thousands of years.
Keepsake doesn't reconstruct, opting instead to record. What you access is what happened, exactly as your senses captured it at the time. It's not rebuilt from pieces. The difference between reconstruction and recovery is why Keepsake exists.
Capture. Index. Experience.
Keepsake Three is a 40nm subdermal (under the skin) device, roughly the footprint of an ant, implanted outpatient under local anaesthetic in a thirty-minute procedure. It passively records sensory input at 300 data points per second across seven channels: ambient audio, visual context (via peripheral retinal sampling), proprioceptive state, cardiovascular markers, ambient environmental data, and two internal channels reserved for emerging indexing modes under active research.
Captured data is encrypted on-device before transmission, before it ever leaves your brain. Your Vault receives only ciphertext, which it stores and indexes against a continuous timeline. Indexing using a technique called homomorphic encryption, that allow keyword and contextual search without decrypting the underlying content, meaning Keepsake can return accurate search results for queries like "October 2019, the café, the blackberry menu" without ever reading the data itself.
Decryption occurs locally, on your device, using keys that live on your implant. We do not hold these keys, because we cannot hold them. This is an architectural constraint, being a design requirement from the first prototype, and was not a policy decision made later.
How we work.
Capture without intervention
The implant is passive. It does not require activation, annotation, or any conscious act of recording. Experience is captured as it is lived, not as it is curated after the fact.
Index without interpretation
The Vault architecture does not assign meaning to what it captures. Using information about time, location, acoustic signature, visual context, and physiological state, Keepsake can index without editorializing about what those details signify.
Preserve without transmission
Zero-knowledge encryption means that the archive never leaves your control. Keepsake's servers hold only ciphertext; anything that can be read is entirely private.
Return without reconstruction
What you see in your Vault is recovered, not rebuilt. The distinction from biological memory is precise and consequential in that your record does not update, does not revise, and does not forget.
We started with patients.
Mnemix Research Laboratory founded at MIT. Early NIH funding for prosthetic memory research in Alzheimer's populations.
Series A funding. First subdermal prototype implanted in a non-human primate model. Continuous recording validated at 300 data points per second.
First human clinical trial at the University of Pittsburgh Medical Center. Twelve participants, early-stage Alzheimer's disease. IRB-approved, fully voluntary.
Results published in Nature Neuroscience. Continuous experiential indexing associated with measurable reduction in clinical distress markers in six of twelve participants.
Neural indexing architecture independently replicated at Stanford, UCL, and the Karolinska Institute. Long-term implant biocompatibility confirmed at five-year follow-up.
Zero-knowledge encryption architecture published in Science. Peer-reviewed key derivation model for biometric neural recording systems.
Twelve-year longitudinal study published in the New England Journal of Medicine. Long-term outcomes in memory disorder patients: no adverse neurological events in the implant cohort.
Keepsake Three receives FDA Class II clearance. 40nm process node. Outpatient surgical procedure. Six-month passive battery life.
Meta-analysis published in Psychological Science. Accuracy of autobiographical recall relative to continuous neural recording across 14,000 participants. Mean divergence: 34%.
Peer-reviewed research.
Solberg, M., & Veith, T. (2007). Continuous experiential indexing as a prosthetic memory aid in early-stage Alzheimer's disease.
Nature Neuroscience, 10(4), 441–449.
Veith, T., Nakamura, Y., & Solberg, M. (2011). Biocompatibility and long-term stability of subdermal neural recording arrays: A five-year cohort study.
Journal of Neural Engineering, 8(6), 066003.
Anand, P., & Solberg, M. (2014). Zero-knowledge architectures for biometric neural recording systems.
Science, 344(6188), 1123–1127.
Chen, R., Mbeki, S., Solberg, M., et al. (2018). Long-term outcomes of continuous experiential recording in memory disorder patients: A twelve-year longitudinal study.
New England Journal of Medicine, 378(14), 1289–1301.
Anand, P., Osei, D., & Reyes, C. (2023). Accuracy of autobiographical memory relative to continuous neural recording: A pre-registered meta-analysis across fourteen cohorts.
Psychological Science, 34(2), 195–211.
A complete list of peer-reviewed publications by Keepsake researchers is available in the annual transparency report. Correspondence regarding research collaboration should be directed to research@keepsake.com.
Experience the archive.
Walk through what the Vault looks like in practice.