Back home

Second

x+web 0.93
Visit source

Intro

Second is the product name First Person Labs PBC introduced for self-sovereign infrastructure that builds and serves a portable user model across a person’s devices and agents. The official blog post “A Model of One’s Own” is subtitled “Introducing Second by First Person Labs PBC,” and says: “We’re excited to introduce Second, a self-sovereign infrastructure to autonomously build and serve your portable user model across your devices and agents.” Related sites are 2.computer and fpl.company.

The company separates this from a chat assistant. Second starts as a streaming-state, portable context vault that builds itself, with background agents learning continuously from raw computer activity at the OS level — memories of recent projects, people, writing samples, or what happened minutes ago. Over time it is meant to become a fuller model of goals, priorities, style, preferences, commitments, and intent, served over local or cloud APIs or MCPs under the user’s control: “the API of you.”

On 17 November 2026, Andreessen Horowitz’s Joel de la Garza and Martin Casado published an investment announcement covering the company and founders Ken Liu and Erik Chi. This entry takes the first-party blog as the product definition and treats the a16z post as contemporaneous public signal, not as a specification.

Advertisement

Explain

The official post treats privacy and sovereignty as invariants, not add-ons. It compares a model of you to a home: it holds personal things, and you decide who comes in, what they see, and when they leave. Concrete claims include: Second runs on the user’s device by default and lives in a folder that can be inspected, analyzed, edited, backed up, or even moved to a USB stick; intelligence can come from an existing Claude or ChatGPT subscription, a bring-your-own API key, local inference, or private inference tokens bought from the company; First Person Labs does not see the user’s data, and data does not pass through FPL, unless the user chooses hosted private inference or Second Cloud; the company says it does not train on, log, or sell that data unless asked to, and that it supports zero data retention (ZDR).

Those are the company’s architecture commitments. Whether data actually stays off FPL’s path depends on whether the user chooses its hosted options. The post also argues that whoever holds user models will steer what software does, which is why it wants control in the hands of the people those models describe, including versus FPL itself.

The name is easy to collide with. Second here is this First Person Labs infrastructure, not the generic “second brain” of note-taking apps, and not other unrelated projects called Second. Citations should include the company name or a 2.computer link.

Usage

The first-party write-up is A Model of One’s Own; company and product entry points also include 2.computer and fpl.company. In the official design, Second keeps building context in the background, so the next question — inside Second or in another agent — does not require re-explaining who you are or what you just did. The same context can be served, under the user’s control, to other software over local or cloud APIs or MCPs.

For inference, the post lists an existing subscription, a bring-your-own key, local inference, or buying private inference or Second Cloud from FPL. Only the last of those sends data through the company; that is the boundary the blog itself draws, and the one to check when assessing privacy.

Background on the company and founders is in a16z’s 17 November announcement. Install steps, system support, and availability should be taken from the product pages at the time of use.