llmgrep

v1.2.0 (shipped 2026-02-01)

A command-line search tool for code indexed by Magellan. Queries SQLite graph databases to find symbols, references, and call relationships. Outputs human-readable or structured JSON for programmatic use.

Works with Magellan

llmgrep requires a Magellan database. It does not build or modify databases — Magellan handles indexing and freshness.

Install Magellan:

cargo install magellan
# or: git clone https://github.com/oldnordic/magellan && cd magellan && cargo build --release

Build a database:

magellan watch --root /path/to/repo --db /path/to/repo.db

Crates:

• magellan v1.8.0 - Code indexing and database creation
• sqlitegraph v1.2.7 - Graph database persistence
• llmgrep - This tool (query only)

What it does

• Searches symbols, references, and calls from Magellan's code database
• Emits deterministic, schema-aligned JSON output for LLM consumption
• Supports regex, intelligent relevance ranking, and optional context/snippets
• --sort-by {relevance|position|fan-in|fan-out|complexity} flag for LLM-optimized or performance modes

Features

• Search modes: symbols (definitions), references, calls, auto (all modes)
• Regex support: Pattern matching with --regex
• Intelligent ranking: Relevance mode (exact > prefix > contains scoring)
• Fast position mode: SQL-only sorting, skips in-memory scoring
• Metrics-based: Sort by fan-in, fan-out, or cyclomatic complexity
• JSON output: Schema-aligned for LLM consumption
• Context/snippets: Optional code context and snippets in results
• FQN filtering: Filter by fully-qualified name patterns
• Language filtering: Filter by programming language (rust, python, javascript, typescript, c, cpp, java, go)
• Kind filtering: Filter by symbol kind (Function, Struct, Method, Class, Enum, Module, etc.)
• Security: ReDoS prevention, resource bounds, path traversal blocking
• Error codes: LLM-E### format with remediation hints
• AST filtering: Filter by AST node kind with overlap matching (--ast-kind)
• AST context: Include enriched AST context with depth and parent info (--with-ast-context)
• Depth filtering: Find symbols by AST nesting depth (--min-depth, --max-depth)

v1.2.0 - AST Filtering (2026-02-01)

AST-aware search features:

• AST node kind filtering: --ast-kind flag to filter by AST node type
  • Supports shorthands: loops, conditionals, functions, declarations, unsafe, types, macros, mods, traits, impls
  • Language-aware expansion: shorthands expand to language-specific node kinds
  • Uses overlap matching for robust results with real Magellan databases
• AST context: --with-ast-context flag adds ast_context to results
  • depth: Nesting depth from AST root
  • parent_kind: Kind of parent AST node
  • children_count_by_kind: Count of direct children grouped by kind
  • decision_points: Number of decision point children
• Decision depth filtering: --min-depth and --max-depth flags
  • Depth counts only decision points: if/match/loop/for/while expressions
  • Root-level code has depth 0

v1.1.1 - Bugfix Release

Critical fix: Metrics JOIN condition - metrics (fan_in, fan_out, cyclomatic_complexity) are now correctly returned in search results. Previous versions compared SHA hash strings to integer row IDs, causing never-matching JOINs.

v1.1.0 - Magellan Integration

Magellan 1.8.0 Integration:

• Safe UTF-8 content extraction (no panics on emoji/CJK/accented characters)
• Chunk-based snippet retrieval (eliminates file I/O when chunks available)
• Metrics-based filtering: --min-complexity, --max-complexity, --min-fan-in, --min-fan-out
• Metrics-based sorting: --sort-by fan-in|fan-out|complexity
• SymbolId lookups via --symbol-id for unambiguous reference
• FQN filtering: --fqn (pattern) and --exact-fqn (exact match)
• Language filtering via --language flag
• Enhanced JSON fields: symbol_id, canonical_fqn, display_fqn, metrics, content_hash

Install

cargo build --release
cp target/release/llmgrep ~/.local/bin/llmgrep

Or install from crates.io:

cargo install llmgrep

Quick start

Query your Magellan database with llmgrep:

# Basic symbol search (relevance-ranked, LLM-friendly)
llmgrep --db /tmp/repo.db search --query "parse"

# Position-only sorting (faster, no scoring)
llmgrep --db /tmp/repo.db search --query "parse" --sort-by position

# Regex search with JSON output
llmgrep --db /tmp/repo.db search --query "^Token" --regex --output json

# Reference search
llmgrep --db /tmp/repo.db search --query "MyType" --mode references

v1.1 Examples

SymbolId lookup (unambiguous)

llmgrep --db code.db search --symbol-id abc123def456789abc123def456789ab

Filter by complexity

llmgrep --db code.db search --query "parse" --min-complexity 10

Filter by language and kind

llmgrep --db code.db search --query "Test" --language rust --kind Function

Sort by fan-in (find hotspots)

llmgrep --db code.db search --query ".*" --sort-by fan-in --limit 20

FQN pattern matching

llmgrep --db code.db search --query "helper" --fqn "%module::%"

v1.2.0 Examples

AST kind filtering

# Find all loops using shorthand
llmgrep --db code.db search --query ".*" --ast-kind loops

# Find conditionals in Python
llmgrep --db code.db search --query ".*" --ast-kind conditionals --language python

# Find specific node kind
llmgrep --db code.db search --query "parse" --ast-kind call_expression

Find deeply nested code (code smell detection)

# Find code nested 3+ levels deep
llmgrep --db code.db search --query ".*" --min-depth 3 --with-ast-context

# Find functions at decision depth 0-1 (simple, testable)
llmgrep --db code.db search --query ".*" --kind Function --max-depth 1

View AST context in results

# Search with AST context to see depth and kind
llmgrep --db code.db search --query "process" --with-ast-context

AST Filtering (v1.2)

Filter search results by AST node kind for structural code queries.

Shorthand Groups

Common patterns have memorable shorthands:

# Find all loops (for, while, loop)
llmgrep --db code.db search --query ".*" --ast-kind loops

# Find conditionals (if, match)
llmgrep --db code.db search --query ".*" --ast-kind conditionals

# Find functions and closures
llmgrep --db code.db search --query "process" --ast-kind functions

# Find all declarations (struct, enum, let, const, static)
llmgrep --db code.db search --query "Config" --ast-kind declarations

# Find unsafe blocks
llmgrep --db code.db search --query ".*" --ast-kind unsafe

Multi-Language Support

Node kinds work across indexed languages:

# Python functions
llmgrep --db code.db search --query "test" --ast-kind functions --language python

# TypeScript classes
llmgrep --db code.db search --query "Service" --ast-kind declarations --language typescript

# JavaScript loops
llmgrep --db code.db search --query ".*" --ast-kind loops --language javascript

Combine Shorthands and Specific Kinds

# Combine multiple shorthands
llmgrep --db code.db search --query ".*" --ast-kind loops,conditionals

# Mix shorthands with specific kinds
llmgrep --db code.db search --query "process" --ast-kind loops,function_item

# Find both loops and closures
llmgrep --db code.db search --query ".*" --ast-kind loops,closure_expression

Available Shorthands

See MANUAL.md for complete node kind reference per language.

Compatibility

• Magellan: v1.8.0 or later (for metrics/chunks support)
• sqlitegraph: v1.2.7 (via crates.io)
• Database schema: Magellan 1.x (graph_entities, graph_edges, symbol_metrics, code_chunks)
• Languages: Rust, C, C++, Java, JavaScript, TypeScript, Python (via Magellan parsers)

Documentation

• llmgrep search --help — Built-in usage examples
• MANUAL.md — Full manual with all options
• CHANGELOG.md — Version history
• Magellan repo — Code indexing tool

License

GPL-3.0-only. See LICENSE.md.