FundingRounds, leads & terms
Andreessen Horowitz backs First Person Labs, maker of on-device personal AI memory app Second
Andreessen Horowitz has invested in First Person Labs. The startup's product, Second, keeps a continuously updated model of a user on their own device and shares that context with the AI agents they choose. The round reportedly totals about $10M.
From the a16z article of November 17, 2026 and cited sources: First Person Labs came out of stealth with an investment led by a16z's infrastructure team. [1] [2] [6]
SOURCES[1] Investing in First Person Labs (a16z.news) a16z.news[2] a16z backs First Person Labs' $10M bet on portable AI… dealroom.co[6] Investing in First Person Labs (a16z.com) a16z.com
AI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- First Person Labs came out of stealth with an investment led by a16z's infrastructure team. [1] [2] [6]
- Original source
- a16z article, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- November 17, 2026 article published
- Covered
- Nov 18 edition
- Evidence
- 12 sources across 10 sites
- 1Investing in First Person Labs (a16z.news)a16z.news
- 2a16z backs First Person Labs' $10M bet on portable AI…dealroom.co
- 3Second2.computer
- 4EP04: The Privacy Layer of Personal Intelligence…augmented-mind.github.io
- 5Ethereum Foundation launches zkAPI (The Block)theblock.co
- 6Investing in First Person Labs (a16z.com)a16z.com
RisingHow it’s measured
Substack · 1 like · 1 comment
Reach blends public engagement (views, likes, comments, shares, stars) from the platforms where the story or its original source lives: YouTube, Hacker News, GitHub, Substack and similar. A video and the article written from its transcript share one score, taken from the video. Platforms with no data are left out rather than counted as zero, and the age of a story never lowers its score.

First Person Labs came out of stealth with an investment led by a16z's infrastructure team. The Information, as cited by Dealroom, puts the round at about $10M. The a16z announcement gives no terms, no other investors are named and the valuation has not been disclosed. [1] [2] [6]
Second builds a profile of a user's preferences, relationships, goals and intent from sources such as email and messages. The profile is stored in a local folder that the user can inspect, edit, back up and move. Agents reach it through local and cloud APIs and the Model Context Protocol (MCP). Users can choose an existing Claude or ChatGPT subscription, their own API key, a local model or private inference. The product is currently waitlist-only. [3] [7]
The founders come from privacy and anonymity research rather than developer tooling. Ken Liu is a Stanford AI PhD student, and Chi is a University of Michigan security PhD student. Their Open Anonymity Project co-built zkAPI with the Ethereum Foundation, which lets people pay for AI APIs without revealing their identity. The company is a public benefit corporation. [4] [5] [8] [9]
Second enters a crowded field. ChatGPT and Claude already have memory features, but each one stays inside its own product. Developer-focused memory layers have raised similar sums: Mem0 raised $24M and Letta raised a $10M seed. Second's pitch is a profile that the user owns and can take from one AI service to another. [2] [10] [11] [12]
So far the evidence is a funding announcement and design descriptions, with no benchmarks, code or audit. If the approach works, portable user-owned context would reduce memory lock-in at each app.
Read the full assessment
Agent builders would then need to plan for access that users grant and can revoke.
Executive brief
Andreessen Horowitz is backing a startup whose main product idea is that AI memory about you should sit on your own computer, not with the AI provider. First Person Labs is building Second, a "model of you" that runs on your device by default and can feed context to whichever agents you choose (a16z). Dealroom, citing The Information, reports about $10M led by a16z's infrastructure team. a16z did not disclose a round size. Valuation is undisclosed. Second has only a waitlist (Second).
What changed and event timeline
Founder sets out the privacy thesis
Ken Liu, a Stanford PhD student, described "unlinkable inference," blind signatures and secure inference proxies as a privacy layer for personal AI ().
zkAPI ships with the Ethereum Foundation
The founders' Open Anonymity Project co-built zkAPI, which uses zero-knowledge proofs so people can pay for AI APIs without revealing their identity ().
About $10M reported
Dealroom, citing The Information, reports roughly $10M led by a16z's infrastructure team as the company comes out of stealth. No other investors are named ().
Capabilities and access
Read the full section
- Product: Second. No version number and no benchmarks have been published.
- Status: waitlist only, with no pricing or supported platforms listed (Second).
- Function: builds a continuously updated profile of you from sources like emails and messages: your preferences, relationships, goals and intent (Second).
- AI model choice, per the company: use an existing Claude or ChatGPT subscription, bring your own API key, run a model locally, or buy private inference tokens (manifesto).
- Legal form: First Person Labs is a public benefit corporation (PBC) (fpl.company).
Technical analysis for researchers and developers
Everything here comes from company materials; there is no paper or code yet.
Read the full section
Everything here comes from company materials; there is no paper or code yet.
- Storage: the profile lives in a folder on your computer that you can inspect, edit, back up or move. The company calls it a "streaming-state, portable context vault" (manifesto).
- Agent access: local and cloud APIs, plus MCP (Model Context Protocol) compatibility. The company calls this "the API of you" (manifesto).
- Hosted option: an optional zero-data-retention mode with confidential computing, and open-source parts planned for self-hosting (manifesto).
- Not yet documented: the file format, how permissions work, how syncing works, and any evaluation of how well it retrieves the right memories.
Claims and evidence
Read the full section
- Funding: about $10M led by a16z. Reported by The Information via Dealroom. The company has not announced it, no filing confirms it, and a16z's post gives no figure.
- Valuation: undisclosed.
- On-device by default, inspectable, portable: stated by both the investor (a16z) and the company (manifesto). No independent check exists yet.
- Founder backgrounds: confirmed by profiles from outside the company. Liu is a Stanford PhD student in AI; Chi is a University of Michigan PhD student in security, privacy and anti-censorship (Augmented Mind).
- Corroboration: no independent evaluation of Second has been found.
Context and prior work
- Built-in assistant memory: ChatGPT and Claude already offer memory features, but each stays inside its own product (Dealroom; StartupHub).
- Developer memory layers: these startups have raised similar sums. Mem0 raised $24M (TechCrunch) and Letta raised a $10M seed (HPCwire).
- What sets Second apart: a profile the user owns and controls, built by a team that came out of anonymity research rather than developer tooling.
Read the full section
- Built-in assistant memory: ChatGPT and Claude already offer memory features, but each stays inside its own product (Dealroom; StartupHub).
- Developer memory layers: these startups have raised similar sums. Mem0 raised $24M (TechCrunch) and Letta raised a $10M seed (HPCwire).
- What sets Second apart: a profile the user owns and controls, built by a team that came out of anonymity research rather than developer tooling.
Limitations, safety and contested findings
- Sensitivity: a16z itself calls a model of a person's relationships, habits and intentions "extraordinarily sensitive" (a16z).
- Open engineering problems: StartupHub lists permission management, reliable syncing and user oversight as unsolved hurdles (StartupHub).
- Market risks: Dealroom points to a crowded market and the need to earn trust with personal data (Dealroom).
Read the full section
- Sensitivity: a16z itself calls a model of a person's relationships, habits and intentions "extraordinarily sensitive" (a16z).
- Open engineering problems: StartupHub lists permission management, reliable syncing and user oversight as unsolved hurdles (StartupHub).
- Market risks: Dealroom points to a crowded market and the need to earn trust with personal data (Dealroom).
- No external testing: no security audit or third-party evaluation has been published.
- Cloud use: "on-device by default" still lets your context reach cloud models if you choose a cloud AI provider (manifesto).
Business and practitioner implications
- For agent builders: MCP support suggests Second could act as a context source that any agent can query.
- For platforms: a portable profile weakens the lock-in that comes from each app holding its own memory of the user.
- For enterprises: a profile that personal agents read from email and messages raises data-governance questions before anyone uses it at work.
Read the full section
- For agent builders: MCP support suggests Second could act as a context source that any agent can query. Teams should plan for user-granted, revocable access instead of building and keeping their own user profiles.
- For platforms: a portable profile weakens the lock-in that comes from each app holding its own memory of the user. That makes Second a strategic bet, not just a feature.
- For enterprises: a profile that personal agents read from email and messages raises data-governance questions before anyone uses it at work.
- For now: with only a waitlist and no benchmarks, Second is something to track, not to adopt.