Blog
From the foundation4 team.
Notes on running AI search and retrieval inside your own security perimeter.
All posts
RSS feed
Technical Foundations
One key per classification: what foundation4 encrypts, and what it doesn't
A map of foundation4's encryption for security reviewers. Which data is encrypted, with which key, where each key lives, what stays readable to the database, and which controls stay with the operator.
Platform
Tracing: see exactly which documents your agent retrieved for an answer
When an agent answer looks wrong, the first question is what the model was given. A foundation4 execution trace shows the fragments each placeholder retrieved, with scores, and the exact prompt sent to the model.
Technical Foundations
Scoping every search to the reader with classifications
One foundation4 pipeline can serve many audiences, each seeing only its share. This post explains classifications, inheritance and matching modes, and how scoping runs in the same query as the ranking.
Technical Foundations
Running retrieval on an isolated network: what works offline
A planning checklist for installing foundation4 on a network with no internet access. Which models, splitters and model servers work offline, how they reach the cluster, and what to confirm before you start.
Technical Foundations
API keys for retrieval integrations: one key for each application
A practical guide to the API keys a retrieval integration needs in foundation4. It covers creating a key, granting permissions, classification allow-lists, expiry and rotation, and the errors you will see.
Platform
Where your content goes, and the one condition on zero egress
We follow one document from the API call to a search result and name every place its text exists. A security reviewer can then check the zero-egress promise, and its one condition, instead of taking it on faith.
Platform
A short list to review: what a foundation4 deployment contains
The full bill of materials for a foundation4 installation, written for the person who approves it. Two Helm charts, a short list of components, three things that hold state, and the hardening the docs describe.
Platform
The researcher and the librarian: your AI needs knowledge infrastructure
You can already ask an AI to read a document. An organization also needs a librarian that finds the right passages across all of its records, for each reader.
Platform
The self-managed imperative: why secure and regulated operations need sovereign AI infrastructure
Many organizations want AI search over their own records but cannot send those records to an outside service. This post explains what running the retrieval layer yourself means in foundation4, from where the data and keys live to how content is removed.
Technical Foundations
Why we built on PostgreSQL
Why foundation4 keeps text, vectors, metadata and version history in one PostgreSQL database, how that compares with document databases and dedicated vector stores, and what the choice costs.
See foundation4 inside your perimeter.
Tell us about your environment and we will walk you through how foundation4 fits.