codegraph-go

Source code knowledge graph for smart code review context — zero dependencies, single binary

---

Why This Exists

AI coding assistants waste tokens reading entire codebases when reviewing changes. codegraph-go builds a structural knowledge graph of your code, then uses that graph to select only the most relevant context for a given diff — giving you smart, token-efficient code review context.

This is a Go reimplementation of code-review-graph (3.7K stars, Python). Our version provides:

• Zero external dependencies — uses Go's go/ast for Go files, regex for others
• Single static binary — no Python, no pip, no tree-sitter C bindings
• Concurrent parsing — goroutine-per-file with configurable parallelism
• Token budget optimization — explicit knapsack-style context selection within a token limit
• DOT graph export — visualize your code's dependency graph with Graphviz

Features

Feature	Description
Multi-language parsing	Go (via go/ast), Python, TypeScript, JavaScript
Code knowledge graph	Functions, classes, imports, call relationships, test coverage
Diff-aware context	Given a git diff, find the most relevant code context
Risk scoring	Security keywords, caller count, test coverage → priority ranking
Token budget	Select most relevant context within a configurable token limit
DOT export	Export graph in Graphviz DOT format for visualization
JSON export	Export stats and subgraphs as JSON
Incremental updates	Remove and re-parse only changed files
Test gap detection	Identify functions without test coverage in changed code

Installation

From Source

go install github.com/JSLEEKR/codegraph-go/cmd/codegraph@latest

Build from Repository

git clone https://github.com/JSLEEKR/codegraph-go.git
cd codegraph-go
go build -o codegraph ./cmd/codegraph/

Quick Start

1. Build the Code Graph

# Parse all source files in the current directory
codegraph build-graph
# Or specify a repository root
codegraph build-graph --repo /path/to/project

Output:

Building code graph for /path/to/project...
Found 42 source files
Build complete in 85ms
Files: 42
Nodes: 287
Edges: 534
Languages: go, python, typescript
Saved to: /path/to/project/.codegraph/graph.json

2. Query the Graph

# Search by name
codegraph query --name "ParseFile"# Filter by kind
codegraph query --kind Function --limit 10
# List all nodes in a file
codegraph query --file "parser/parser.go"# JSON output
codegraph query --name "Server" --format json

Output:

Found 2 results:
[f] ParseFile
Qualified: parser/parser.go::ParseFile
File: parser/parser.go:35-68
Params: (filePath string, source []byte)
Returns: (*ParseResult, error)
Calls: parseGo, parsePython, parseTypeScript
Called by: (none)
[f] parseGo
Qualified: parser/go_parser.go::parseGo
File: parser/go_parser.go:15-120
Params: (filePath string, source []byte)
Returns: ([]graph.Node, []graph.Edge, error)

3. Get Diff Context

# Find relevant context for recent changes
codegraph diff-context --base HEAD~1
# With custom token budget
codegraph diff-context --base main --budget 4000
# JSON output for programmatic use
codegraph diff-context --base HEAD~3 --format json

Output:

Changes in 3 files affecting 5 code entities
Impact radius: 12 nodes
Test gaps: 2 functions without tests
Token budget: 3200 / 4000 used
Functions missing tests:
- validateInput (handler.go:45)
- parseConfig (config.go:12)
Review priorities (by risk score):
## validateInput (Function) [risk: 0.50]
File: handler.go (lines 45-60)
Tokens: 120
​```
func validateInput(input string) error {
if input == "" {
return fmt.Errorf("empty input")
}
// ... sanitize ...
}
​```

4. View Statistics

codegraph stats
codegraph stats --format json

Output:

Graph Statistics:
Total nodes: 287
Total edges: 534
Files: 42
Languages: go, python, typescript
Nodes by kind:
Function 142
File 42
Class 35
Type 28
Test 40
Edges by kind:
CALLS 215
CONTAINS 187
IMPORTS_FROM 89
TESTED_BY 31
INHERITS 12

5. Export Graph

# Export as DOT (for Graphviz)
codegraph export --format dot > graph.dot
dot -Tpng graph.dot -o graph.png
# Export filtered subgraph
codegraph export --format dot --filter "parser.go::ParseFile,parser.go::parseGo"# Export as JSON
codegraph export --format json --output stats.json

Architecture

codegraph-go/
├── graph/ # In-memory knowledge graph with JSON persistence
│ └── graph.go # Node/Edge types, BFS impact analysis, save/load
├── parser/ # Multi-language source code parser
│ ├── parser.go # Language detection, file-level orchestration
│ ├── go_parser.go # Go parser using go/ast (stdlib)
│ └── regex_parser.go # Python/TypeScript parser using regex
├── diff/ # Git diff parsing and node mapping
│ └── diff.go # Unified diff parser, line range overlap detection
├── context/ # Token-budget-aware context selection
│ └── context.go # Risk scoring, greedy knapsack, context formatting
├── export/ # Graph export (DOT, JSON)
│ ├── dot.go # Graphviz DOT format with color-coded node types
│ └── json.go # JSON stats and subgraph export
└── cmd/codegraph/ # CLI entry point
└── main.go # Command router and flag parsing

How It Works

Source Files → Parser (go/ast + regex) → Nodes + Edges
↓
In-Memory Graph
↓
Git Diff → Changed Ranges → Node Mapping → Risk Scoring
↓
Token Budget → Context Selection
↓
Review Priorities

1. Parsing: Go files are parsed with go/ast for perfect fidelity. Python and TypeScript files use regex patterns to extract classes, functions, imports, and calls.

2. Graph Construction: Nodes (File, Class, Function, Type, Test) and Edges (CALLS, IMPORTS_FROM, INHERITS, IMPLEMENTS, CONTAINS, TESTED_BY, DEPENDS_ON) are stored in an in-memory graph with index maps for fast lookup.

3. Diff Analysis: Git diffs are parsed into line ranges. Each range is overlapped against graph nodes to find affected code entities.

4. Risk Scoring: Each affected node gets a risk score (0.0-1.0) based on:

   • Caller count (more callers = higher blast radius)
   • Test coverage (untested = higher risk)
   • Security keywords (auth, password, sql = higher risk)

5. Context Selection: Nodes are sorted by risk and selected greedily within the token budget. Source code is read from disk and included in the output.

Node Types

Kind	Description	DOT Shape
File	Source file	folder
Class	Class/struct definition	box
Function	Function/method	ellipse
Type	Interface/type alias	hexagon
Test	Test function	diamond

Edge Types

Kind	Description	Example
CALLS	Function invocation	main() → helper()
IMPORTS_FROM	Module import	handler.go → fmt
INHERITS	Class/struct embedding	Dog → Animal
IMPLEMENTS	Interface implementation	Server → Handler
CONTAINS	Structural containment	file → function
TESTED_BY	Test coverage link	Parse() → TestParse()
DEPENDS_ON	Dependency relationship	module → package

Language Support

Language	Parser	Functions	Classes	Imports	Calls	Tests
Go	go/ast (stdlib)	Yes	Yes (structs)	Yes	Yes	Yes
Python	Regex	Yes	Yes	Yes	Yes	Yes
TypeScript	Regex	Yes	Yes	Yes	Yes	Yes
JavaScript	Regex	Yes	Yes	Yes	Yes	Yes

Go Parser Details (go/ast)

The Go parser provides first-class support via the standard library:

• Functions and methods with receiver types
• Struct and interface declarations
• Struct embedding (→ INHERITS edges)
• Import resolution
• Call extraction including package-qualified calls
• Test function detection (Test*, Benchmark*, Example*)
• Full parameter and return type extraction

Python/TypeScript Parser Details (Regex)

The regex parsers extract:

• Class definitions with inheritance
• Function/method definitions with parameters
• Import statements
• Function calls (direct and method calls)
• Arrow functions (TypeScript)
• Interface and type alias declarations (TypeScript)
• Test file and test function detection

Configuration

Ignored Directories

The following directories are automatically skipped during parsing:

.git, node_modules, vendor, __pycache__, .codegraph,
.cache, dist, build, .next, .nuxt, target, bin,
.idea, .vscode, .vs

Graph Storage

The graph is stored in .codegraph/graph.json in the repository root. This file can be committed to the repository for faster subsequent builds, or added to .gitignore.

Comparison with code-review-graph (Python)

Feature	code-review-graph (Python)	codegraph-go
Dependencies	tree-sitter, networkx, mcp, watchdog	Zero (stdlib only)
Languages	25+ (via tree-sitter)	4 (Go, Python, TS, JS)
Distribution	pip install + tree-sitter compilation	Single binary
Parsing	tree-sitter (C bindings)	go/ast + regex
Storage	SQLite	In-memory + JSON
Concurrency	Single-threaded	Goroutine-per-file
Token Budget	Implicit (bounded BFS)	Explicit knapsack
Graph Export	HTML visualization	DOT + JSON
MCP Server	Yes (24+ tools)	No (CLI only)
File Watcher	Yes (watchdog)	No

When to Use Which

• codegraph-go: When you want a fast, zero-dependency CLI tool for code review context, especially for Go projects
• code-review-graph: When you need MCP integration, 25+ language support, or VS Code extension

Development

# Run tests
go test ./...
# Run tests with verbose output
go test ./... -v
# Build binary
go build -o codegraph ./cmd/codegraph/
# Run vet
go vet ./...

License

MIT License — see LICENSE for details.

Credits

Inspired by code-review-graph by tirth8205.