ifactorial
Agentic specification platform for enterprise modernisation
Turns codebases and documents into reviewable, citation-grounded specifications. Ingest sources into temporal buckets (Present / Legacy / Future / Policy), ask an agent that retrieves and cites evidence, draft specs in a collaborative editor with inline agent tags, chain multi-step Runbooks, grade sections in Spec Review against policy, and export a Publish Pack (spec, citations, review). Built as a monorepo (web / API / worker) for programmes where the specification has to be signed before code. Under the hood it is an agent harness: specialised agents (`source-retrieval`, `spec-draft`, `citation-gap`, `change-impact`, `question`, `general`) share a common dispatch path. Hub retrieval runs a fixed five-stage pipeline—query rewrite/HyDE, dual vector + knowledge-graph retrieval with grading and ranking, then guarded cited prose—while editor `@`-tags load an eager context envelope (doc, anchor, thread, sources, RAG/KG hits) plus a project-scoped read-only tool surface so specialists can go deeper without leaving the comment thread. Runbooks chain those capabilities into gated multi-step workflows; Spec Review batches the same harness across tabular sections against Policy. Ingestion clusters and knowledge-graph extraction feed the retrieval substrate so answers stay grounded in project evidence.


