Codex
Codex is an internal knowledge surface for engineering organisations. It treats code, decisions, and incidents as a single retrievable substrate — observed continuously, not indexed once.
01 — What it is
A retrieval and reasoning surface for engineers and engineering leaders. Codex holds the full living archive of an organisation's technical work — repositories, design documents, incident reviews, architectural decisions — and answers questions against it with citations.
It does not replace the act of writing things down. It makes that act compound.
02 — How it differs
Most internal AI tooling indexes documents once, then quietly drifts out of date. Codex observes the underlying systems continuously — pull requests, runbooks, postmortems — and treats freshness as a property of the index itself.
Answers carry citation chains. Engineers can ask why a decision was made and arrive at the original document, not a paraphrase of it.
03 — Where it runs
Deployed inside three engineering organisations. Codex respects existing access controls — no document is ever surfaced to a person who could not have read it directly.
Engineered to be inspected.
Retrieval
Hybrid (semantic + lexical + structured)
Index freshness
Sub-minute
Deployment
Self-hosted
Access model
Existing IDP-aware