Concepts

This section explains the ideas behind Canopy. Read these if you want to understand what’s happening under the hood — why the product works the way it does, what its limits are, and how to reason about it.

Pages in this section

What Canopy Is and Isn’t — positioning: a local code-intelligence engine with one answer tool, the benchmark evidence, and what Canopy deliberately is not.
How Canopy Works — the pipeline: one local index (AST symbols, import graph, full-text, embeddings, optional summaries), one routed answer call, one server-side recipe.
AST Parsing — why Canopy indexes ASTs instead of grepping text, the 55+ languages and formats it extracts symbols and imports from, and the limits of static analysis.
MCP Protocol — what the Model Context Protocol is, how Canopy serves it over stdio, and why the server exposes exactly one tool.
License and Privacy — exactly what Canopy sends over the network (one weekly heartbeat), what it doesn’t (your code, queries, file paths), and the air-gapped option.
Machine Fingerprints and Bindings — how licenses bind to machines, what gets hashed, and how stale bindings are evicted.
One Tool: canopy_answer — the whole MCP surface: the fourteen question shapes, the deterministic router (360/360 on its labelled corpus), and the honest no-match behavior.
Phrasing Picks the Recipe — how to ask so the router locks on: name the entity type, backtick paths and symbols, with good/weak phrasing pairs for every shape.

If you’re in a hurry

Wondering if Canopy will send your code anywhere? Read License and Privacy first.
Evaluating Canopy for your team? Read What Canopy Is and Isn’t for honest positioning with the benchmark numbers.
Want to know what you can ask? Read One Tool: canopy_answer — the shape list is a one-page cheat sheet.
Getting weak answers? Read Phrasing Picks the Recipe and sharpen the question.