flock capital

Flock Capital

Flock Capital is an independent lab for AI alignment and safety, based in Seattle. We study how models behave, how safeguards fail, and how to tell an aligned system from an agreeable one. We also build software: the tooling our research runs on, and products of our own.

Our products

reflet reflet.sh A reverse-engineering focused code editor: a code editor and a disassembler in one app. invite-onlymacOS & Linux causerie causerie.sh A salon of AI regulars, each with one job. They work while you’re away and ask before they act. websigns in with flock id cantrip github.com/flockcap/cantrip Speak plainly, get the incantation: a shell companion that drafts the command through the coding agent you already use. No API key. open sourcemacOS, Linux & Windows

Our research

Dictionary Learning as Unsupervised Steering-Vector Generation Ontologizers vs. sparse autoencoders on SONAR sentence embeddings: a discrete classifier code as an unsupervised source of steering vectors. working notesOctober 2026with Wakaru

The questions we work on:

We also run a small consulting practice: model evaluations, adversarial testing, software engineering, and cybersecurity. Engagements are scoped, written up properly, and pay for the research. Start an engagement.

The lab is self-funded and supported by research grants. If you work on similar problems and want to compare notes, we’d like to hear from you.

For inquiries, reach us at contact@flock.holdings or use the contact form.

Write to the lab

Or email contact@flock.holdings. Inquiries are read by the people who would do the work.

Submissions are transmitted privately to the lab and are not stored by this website. See privacy.