Mitosis Labs is a critical player in the agentic memory layer of the AI stack. As AI agents move from simple chatbots to persistent assistants, the bottleneck becomes "state"—how an agent remembers past interactions and preferences across different sessions and tools. Mitosis provides this state through Cortex, a model-agnostic memory layer that prevents vendor lock-in. This is particularly relevant for the AI agent ecosystem because it allows specialized agents from different developers to tap into a single, user-controlled source of truth without each agent needing to rebuild the user's context from scratch.

By championing a "third-party memory" architecture, Mitosis enables a modular agent ecosystem where users can swap specialized agents in and out of their workflow. Their work on scoped permissions and private inference directly addresses the trust barrier that currently prevents many businesses from deploying autonomous agents. They are active in the infrastructure and privacy layers of the stack, pushing for standards that prioritize data sovereignty and interoperability between models.

The struggle for user memory

The current trajectory of artificial intelligence is defined by a massive participation gap. While the industry fixates on the capabilities of the latest large language models, Mitosis Labs points to a different statistic: 83% of people have never used AI. For the 15% who have, the experience is often transactional and extractive. Users hand over their data to a handful of dominant platforms in exchange for temporary utility, building a digital memory they do not actually own. Mitosis Labs, a San Francisco-based research lab, is attempting to redraw this boundary by building a human-AI interface focused on data sovereignty.

Founded by Alex Morris and Prakshal Jain, the company is built on the thesis that memory should outlive the platform wars. Morris previously scaled a security and cryptography company to $100 million in value, while Jain was a Meta engineer whose work on the Quest headset reached billions. Their combined background in high-stakes security and mass-market user experience informs the company’s core product: Cortex.

Cortex and the neutral layer

Cortex is a private memory layer that functions as a source of truth for AI agents. The central problem Mitosis identifies is data absorption. When a user connects their email, documents, and calendar to a primary AI platform, that platform learns from the data and eventually competes with the user's own business or interests. Mitosis builds a "neutral layer" instead. Cortex is third-party by design; it works with any model but belongs to none of them. This allows a user to switch from an OpenAI model to a local Llama instance without losing the context of their past decisions, relationships, or business logic.

Technically, Mitosis employs private inference, allowing users to run their memory against frontier models when they need power, or against local models when they require absolute privacy. This creates a scoped permission system where users decide exactly what context leaves their controlled environment for any given query. This is not just a privacy feature; it is an architectural commitment to portability. If the model you love today is leapfrogged tomorrow, your memory compounds independently of the underlying compute provider.

Interface and accessibility

While Cortex provides the back-end infrastructure for data ownership, Mitosis also develops front-end interfaces to lower the barrier to entry for AI. Yappy is their voice-first product, designed to turn normal conversation into the primary way users interact with models. This targets the segment of the population that finds traditional prompt-based interfaces unintuitive. By removing menus and complex prompting, the company hopes to close the adoption gap while ensuring that the first-time users' data is protected from the start.

The lab operates with a small, family-style team in San Francisco, building in the open with a community of operators. Their partnerships with projects like XMTP for encrypted messaging and Supabase for private data architecture further signal their intent to build a stack that values ownership over platform dominance. For Mitosis, the most important property line of the next decade is the one drawn between a user's mind and the corporate models that attempt to absorb it.