AI Cortex

The unified company knowledge base for an agentic SDLC. AI Cortex turns everything connected to a software project — code repositories, work items, wikis, and uploaded documents — into a single, queryable knowledge graph, and answers natural-language questions about it with cited, cross-repository context.

It ingests broadly: Azure DevOps, GitHub, GitLab, Bitbucket, Gitea and generic Git for code; ADO wikis and work items, Confluence, Notion, SharePoint, Google Drive, public doc URLs, and uploads for knowledge — including OCR and vision, so a diagram inside a scanned PDF becomes searchable rather than an attachment nobody opens. Ingestion builds a typed graph of repositories, files, classes, functions, API endpoints, and documents, with the relationships between them.

The differentiator is cross-repo retrieval by graph traversal: when service A calls service B, that link is a real edge discovered at ingestion time. So a question like "why is the getUser API failing?" reaches the calling service, the endpoint, the wiki page, and the related work item by walking typed edges — not by hoping a vector search surfaces them. Repositories are ingested once for the whole organization and attached to as many projects as need them; syncs after the first are incremental.

AI Cortex is a substrate, not a single product. Each project workspace is also a hard security boundary, enforced at every layer rather than by convention, so an agent answering inside one project is structurally incapable of reading another's code. On top of that isolated, graph-grounded retrieval layer runs a family of five agents covering the whole lifecycle; each brings only a prompt, an edge policy, and a surface, while the graph, retrieval, and isolation are shared. It self-hosts on the organization's own infrastructure — single VM, Docker Swarm, or air-gapped — and stays model-agnostic behind any OpenAI-compatible endpoint, switchable live. An MCP server exposes the same graph to editors like Cursor and Claude Desktop, so a developer can run impact analysis without leaving the IDE.