Everything below is real, running code, not a roadmap. Some of it is a working prototype. None of it is a finished product. We are naming that distinction on purpose, because a lot of what gets called AI capability right now is a slide, not a system.
The underlying thesis is one bet, two layers. First, an organisation’s memory is scattered across documents, email, calendar, Slack and meetings, and needs to be extracted into something structured before anything can reason over it. Second, once that memory exists, it can be reasoned over, to surface what a team already knows but has not written down, where accountability actually sits, and where risk is quietly concentrated.
Each of these is real, running code you can inspect yourself, not a case study written after the fact.
An agent that reads documents, email, calendar entries, Slack threads and meeting transcripts, and extracts what is actually happening into seven structured memory types, instead of one flat summary. This is the raw material layer: before anything can reason about an organisation, its scattered communication has to become something structured enough to reason over.
Status: working prototype. Public, MIT licensed, with a CLI and a Python API.
View the repoMemory extraction only matters if it feeds something. This repo runs a full synthetic organisation, generated communications across email, Slack, CRM, HR records, calendar and meeting transcripts, through a three-stage pipeline: parse the raw data, extract memory from it, then synthesize that memory into personas and early insight.
Status: early-stage R&D. All data is synthetic; the insight-synthesis stage is a working prototype, not a finished product.
View the repoA customer-owned checkpoint between an organisation’s people and any AI provider they use. It classifies and redacts sensitive data before it leaves an organisation’s own infrastructure, applies policy on what is allowed to go out, and returns signed, tamper-evident audit receipts, without storing the underlying content.
Status: what its own documentation calls a security-oriented MVP, not a certified product. Public, Apache 2.0.
View the repoThe same structured-extraction approach, one layer down: from an organisation’s memory to an individual’s professional one. Given a person’s own documented history, it extracts structured insight about career trajectory, skills and working patterns, building a persistent professional profile instead of starting every conversation from zero.
Status: early-stage R&D. Public; extraction and profile detection are working, later stages are still roadmap.
View the repo