01
Memory is the missing primitive.
Bigger context windows aren't the answer. Agents need durable, structured memory: searchable, governed, and shared across sessions and teammates. Not a cache. A first-class layer of the stack.
Your next user isn't human.
We're building the shared memory layer for AI agents, so every interaction compounds into knowledge the next agent, and the next team, can use.
The thesis— Thesis
AI writes nearly half our code. We're still building tools for humans.
The feedback loop is broken, and that's the opportunity.
Generate. Review. Fix. Regenerate. The cycle repeats because the systems we hand to agents don't remember what they learned the last time around. Every session starts from zero. Every fix gets re-discovered. Every team pays the same tax twice.
We think the next generation of developer tools will treat machines as first-class users, not as awkward guests in software written for humans. APIs, documentation, codebases, runbooks: all of it should speak fluently to both.
memco is the layer that closes the loop. When agents can remember, every interaction makes the next one smarter. Generate-fix-repeat becomes generate-learn-compound.
— What we believe
Three convictions guide the work.
02
Learning compounds, or it dissipates.
Lessons learned in one session should make the next one cheaper, faster, and better. Without a memory layer, every discovery evaporates the moment a task ends. With one, knowledge accrues for the agent, the team, and the company.
03
Private by default.
Institutional knowledge is the asset. Memory has to live inside your boundary, governed like any other production data store. We design for that constraint first, not as an afterthought.
— Founded 2025
Built by operators who've shipped memory systems before.
Scott Taylor
Co-Founder & CEO
Former Global Head of AI Products at AIG Investments ($320B AUM), building machine-learning trading systems through the pre-GPT era. Previously CEO at miDrive (EdTech, scaled to 300k+ downloads), CPO at Livedrive, and NED / Founding Director at Simba Sleep. A builder who's been shipping production AI at serious scale for the better part of a decade.
Valentin Tablan
Co-Founder & CTO
Former Lead Scientist for Amazon Alexa, with 20+ years at the cutting edge of natural-language and knowledge-based AI. Chief AI Officer at Ieso Digital Health, where he created Velora — the world's first clinically validated generative AI therapy agent, with outcomes on par with human-delivered care.
Thomas Hall
Co-Founder & GTM
Fifteen years selling technical products into serious engineering orgs, most recently at the AI/ML infrastructure layer. Led sales at Pachyderm (acquired by HPE), Qwak (acquired by JFrog), and Ambassador (acquired by Gravitee), with prior leadership seats at Simplismart and BackerKit. Partners with the design partners and early enterprise customers shaping how Spark gets adopted in the wild.
Kristoffer Bernhem
Co-Founder & Principal AI Engineer
Previously at Legora and Sana, building AI systems at two of Europe's standout AI companies. Owns the retrieval, ranking, and reflection layers at the core of Spark — the architecture that decides what the system remembers and how it gets surfaced.
Begin the conversation