Features & comparison

On this page

Why greplm?

Every other code-search tool is built for a human scrolling a terminal. greplm is built for the agent loop: small, structured, token-compact results over a hot local index — the kind a program can hammer thousands of times a session without drowning in file text.

	greplm	ripgrep	ctags / LSP
Content search (literal/regex/word)	✅	✅	❌
Symbol definitions (14 languages)	✅	❌	✅
Find references	✅	⚠️ text-only	✅
Call graph: callers / callees	✅	❌	⚠️ LSP only
Blast radius (transitive impact)	✅	❌	❌
Typed go-to-definition	✅	❌	✅ LSP
Structural / AST search	✅	❌	❌
Git time-travel (blame/history/changed)	✅	❌	❌
Task context packs (budgeted)	✅	❌	❌
Persistent incremental index	✅	❌ (scans each run)	⚠️ regenerate
Warm daemon (sub-ms queries)	✅	❌	❌
Token-compact output for agents	✅	❌	❌
Optional offline semantic search	✅	❌	❌
MCP server + ready-made agent files	✅	❌	❌
Fully offline / no network	✅	✅	✅

Reach for ripgrep when you are scanning a terminal. Reach for greplm when an agent needs a queryable index it can hammer thousands of times without re-scanning the tree.

Features

• Fast content search — literal, regex, whole-word, with language and path filters.
• Symbol lookup — find definitions by exact, prefix, substring, or fuzzy match across Rust, Python, JavaScript, TypeScript/TSX, Go, Java, C, C++, C#, Ruby, PHP, Swift, and Dart (Flutter).
• References — locate every occurrence of an identifier, definitions first.
• Call graph — callers / callees / impact (blast radius) built from a structural reference index, plus xref for resolved references (definitions, calls, imports).
• Typed go-to-definition — def <file> <line> <col> resolves the identifier under the cursor using scope, usage context, and imports, flagging the unambiguous target.
• Structural (AST) search — match a tree-sitter query or a $NAME meta-variable pattern, trigram-prefiltered so it stays fast.
• Context packs — pack "<task>" --budget N assembles exactly the code relevant to a task, ranked by lexical relevance and call-graph centrality, packed to a token budget.
• Git time-travel — blame, symbol history, and changed <rev> annotated with symbols.
• File outlines & snippets — read a file’s structure or an exact slice with context.
• Incremental indexing — only re-indexes what changed; an optional watcher and warm daemon keep queries sub-millisecond.
• Offline & private — no network calls, nothing leaves your machine.