What Canopy Is and Isn't
The one-line version
Section titled “The one-line version”Canopy is the self-hosted code intelligence platform that makes your AI coding agent dramatically smarter — no cloud, no credits, one binary.
That sentence has a lot of words chosen carefully. Let’s unpack them.
What Canopy is
Section titled “What Canopy is”A context engine
Section titled “A context engine”Canopy does not include an AI model and does not generate code. It is a context engine — software that transforms a raw code repository into a structured, queryable graph of symbols, dependencies, health findings, and history that an AI agent can interrogate precisely.
The AI agent (Claude Code, Cursor, Codex CLI, etc.) supplies the intelligence. Canopy supplies the facts about your specific codebase. Together they produce dramatically better results than either can alone.
Think of it this way: an AI agent without Canopy is a very smart developer who has never seen your codebase. With Canopy, that same developer has a complete architectural map, annotated with which files are risky to touch, which exports are dead weight, and which modules have been thrashed by recent commits.
Self-hosted
Section titled “Self-hosted”Canopy runs entirely on your machine. The binary indexes your repo into a local SQLite database. The MCP server that answers tool calls runs on your laptop or CI server — not on Ironpine Labs infrastructure, not in any cloud.
Your code never leaves your machine. The only outbound connection Canopy makes is a weekly license health check that sends a hash of your license key, your Canopy version, and your platform. No code. No queries. No file paths. See License and Privacy for the exact payload.
A single binary
Section titled “A single binary”No Docker. No daemon to manage. No service mesh. One static binary that you put
on your PATH. canopy serve . starts the MCP server. canopy index . builds
or updates the index. canopy health runs checks. Everything else follows.
An MCP server with 21 tools
Section titled “An MCP server with 21 tools”Canopy exposes its intelligence through the Model Context Protocol (MCP). Any MCP-compatible AI client (Claude Code, Cursor, Windsurf, Zed, Continue, and others) can connect and use all 21 tools. The tools are organized into three tiers: three workflow composites that cover 80% of use cases, and 18 detailed tools for deep analysis. See Tool Catalog for the full list.
What Canopy is not
Section titled “What Canopy is not”Not an AI assistant
Section titled “Not an AI assistant”Canopy does not write code, explain concepts, or answer questions about programming in general. It answers structural questions about your specific codebase: “Who imports this file?” “What exports are unused?” “Which files have changed most in the last 30 days?” “Is this module reachable from any entry point?”
Not a linter or static analyzer
Section titled “Not a linter or static analyzer”Canopy is not a replacement for ESLint, Clippy, mypy, or similar tools. Those tools enforce code style and language-specific rules. Canopy operates at the architectural level: dependency edges, module reachability, dead code at the module boundary, secrets in source files. The two complement each other.
Not a hosted service
Section titled “Not a hosted service”Canopy does not have a cloud pricing tier. There is no “Canopy Cloud” or “Canopy SaaS.” The product is a binary you download and run. If you need zero network calls (defense, healthcare, classified environments), the Air-Gapped tier compiles out all networking at build time. See License and Privacy.
Not for enterprise (yet)
Section titled “Not for enterprise (yet)”Canopy’s current tiers (Solo, Pro, Team) target individual developers and small
teams of 3–10. There is no SSO, audit logging, or SLA. Enterprise teams with
those requirements should contact [email protected] for
custom contracts.
How Canopy compares
Section titled “How Canopy compares”| Canopy | Augment Code | Sourcegraph Cody | |
|---|---|---|---|
| Where it runs | Your machine | Cloud | Enterprise self-hosted |
| Code leaves your machine? | Never | Yes | Depends on config |
| Pricing model | One binary, per-seat annual | Credit-based | Enterprise only (free/pro discontinued) |
| AI model included | No | Yes | Yes |
| MCP server | Yes (21 tools) | Partial | Partial |
| Air-gapped option | Yes | No | Yes (enterprise) |
The honest tradeoffs
Section titled “The honest tradeoffs”Canopy has real limits worth knowing:
- Five languages only (v1.4.0): TypeScript, JavaScript, Python, Rust, Go. Other languages are indexed for text search but not AST-parsed. More languages are on the roadmap.
- First index takes time: A 50K-file repo takes 5–15 minutes for a full index with search and git history. Incremental updates after the first run are fast.
- Weekly heartbeat exists: Canopy makes one outbound network call per week for license validation. It’s minimal (hash + version + platform), but it exists. If you need zero network calls, the Air-Gapped tier is the answer.
- Not a linter: Canopy finds dead exports and broken imports, but it won’t catch type errors, style violations, or logic bugs. Keep running your existing analysis tools alongside Canopy.