Concepts

Veska is a local code-intelligence daemon. It parses your repo into a graph, embeds that graph semantically, and exposes both through MCP so your editor and your AI agent reason from the same model of the codebase. The data lives in ~/.veska/ on your machine - no upstream, no shared service, no multi-tenant tier.

These pages give you the mental model an operator needs. They are deliberately lighter than the detailed architecture summary at docs/ARCHITECTURE.md; refer to that when you want the binding details.

• The code graph - how your repo becomes nodes and edges.
• Promotion & staging - the two clocks: structural recall on save, semantic recall on commit.
• Semantic search & embeddings - the in-process embedder and the lexical fallback.
• Finding duplicate & similar code - exact, structural (renamed), and near tiers, repo-wide and cross-repo.
• Daemon topology - one binary, three personalities, and what the daemon owns.

The whole product, in three boxes

One daemon (veska-daemon) owns one SQLite file under ~/.veska/. Two thin clients talk to it: the veska CLI and the veska-mcp editor shim. The default embedder runs in-process, so the default config makes no outbound connection at all.