Mantic.sh

Summary

Mantic is a context-aware code search engine that prioritizes relevance over raw speed. After testing across 5 repositories (cal.com, next.js, tensorflow, supabase, chromium), it demonstrates superior result quality compared to grep/ripgrep, despite some trade-offs in speed for very large codebases.

Overall Assessment: 4/5 - Excellent for AI agents, good for developers, needs speed optimization for 100K+ file repos.

What's New in v1.0.25

Enterprise-Grade Context Infrastructure

• Semantic Reranking (Hybrid Intelligence): Combines heuristic speed with neural understanding. Uses local embeddings (transformers.js) to find "conceptually relevant" code even without exact keyword matches.
  • Usage: mantic "verify user" --semantic
• Code Intelligence: Deep understanding of your codebase structure using Tree-sitter.
  • Go to Definition: mantic goto UserService returns the exact line number across your entire monorepo.
  • Find References: mantic references handleLogin finds every usage, respecting .gitignore.
• Learned Context (Team Memories): Mantic now remembers which files solved previous queries. These patterns are saved locally (.mantic/search-patterns.json) and can be committed to git to share knowledge across your team.
• Python Support: Now includes first-class support for Python imports in the dependency graph.
• Security & Stability:
  • Regex DoS protection for user inputs.
  • Command injection mitigations for VS Code extension.
  • Safe fallback for non-git directories (scans allow-listed extensions).

Performance Update: v1.0.25 is ~2x faster than previous versions, scanning Chromium (481K files) in <2 seconds.

Tested on 481K files (Chromium) with 100% multi-repo accuracy.

See the CHANGELOG for detailed release notes.

Table of Contents

• About the Project
• Proprietary vs Mantic
• Performance Benchmarks
• Accuracy & Relevance
• Feature Comparison
• Use Case Recommendations
• Installation
  • CLI Installation
  • MCP Server Installation
• Usage
• Agent Rules
• How It Works
• License

About the Project

Mantic is an infrastructure layer designed to remove unnecessary context retrieval overhead for AI agents. It infers intent from file structure and metadata rather than brute-force reading content, enabling retrieval speeds faster than human reaction time.

Key Benefits

• Speed: Retrieval is consistently under 500ms for most repos, under 4s for massive monorepos (Chromium).
• Efficiency: Reduces token usage by up to 63% by filtering irrelevant files before reading.
• Privacy: Runs entirely locally with zero data egress.

Proprietary vs Mantic (Cost Analysis)

For a team of 100 developers performing 100 searches per day (approx. 3M searches/year):

Tool	Annual Cost (Est.)	Per-Search Cost	Privacy
Mantic	$0	$0	Local-First
Vector Embeddings (DIY)	$1,680 - $10,950*	$0.0005 - $0.003	Cloud
SaaS Alternatives	$46,800+	$0.015+	Cloud

Note: Mantic costs are zero. Vector/SaaS costs are estimates based on standard managed infrastructure (e.g. Pinecone/Weaviate managed pods + compute) or per-seat Enterprise licensing (e.g. GitHub Copilot Enterprise).

Performance Benchmarks

Speed Comparison (Real-world queries)

Repository	Files	Query	Mantic v1.0.25	ripgrep	fzf	Verdict
cal.com	9.7K	"stripe payment"	0.288s	0.121s	0.534s	Fast
next.js	25K	"router server"	0.440s	0.034s	0.049s	Fast
tensorflow	35K	"gpu"	0.550s	0.022s	N/A	Fast
chromium	481K	"ScriptController"	1.961s	0.380s	0.336s	<2s (Massive)

Speed Verdict:

• Caching works well for most repos (4-17% improvement on second run).
• Large repos (Chromium) show modest but consistent improvements.
• Mantic is slower than ripgrep/fzf on raw speed but prioritizes ranking.

Accuracy & Relevance Analysis

Major Strengths

1. Exact Path Matching

• Query: "router server" in next.js
• Mantic: Found packages/next/src/server/lib/router-server.ts (Score: 220)
• ripgrep: Found files mentioning "router" and "server" separately (many false positives)
• Verdict: Mantic found the exact file that matches the intent.

2. CamelCase Detection

• Query: "ScriptController" in chromium
• Mantic: Found script_controller.h, script_controller.cc (Score: 200)
• ripgrep: Requires manual regex script.*controller
• Verdict: Mantic's CamelCase detection is production-ready.

3. Directory Boosting for Acronyms

• Query: "gpu" in tensorflow
• Mantic: Prioritized files in tensorflow/lite/delegates/gpu/
• Verdict: Mantic correctly prioritizes structural relevance.

4. Path Sequence Matching

• Query: "blink renderer core dom" in chromium
• Mantic: Found third_party/blink/renderer/core/dom/README.md
• Verdict: Mantic matches multi-term path queries perfectly.

Feature Comparison Matrix

Feature	Mantic	ripgrep	ag	fzf
Text Search Speed	2-10x slower	Fastest	Slow (large repos)	Very Fast
Relevance Ranking	Excellent	None	None	Basic
Path Structure Awareness	Perfect	None	None	Partial
CamelCase Detection	Yes	No	No	No
Exact Filename Matching	Yes	No	No	Yes
Multi-Word Queries	Semantic	Regex needed	Regex needed	AND logic
Go to Definition	Yes (Cross-Repo)	No	No	No
Find References	Yes	No	No	No
Impact Analysis	Yes	No	No	No
Zero-Query Mode	Yes	No	No	No

Use Case Recommendations

Best For

1. AI Agents (Context-aware search with metadata)
2. Finding Files by Intent ("Where is payment code?")
3. Understanding Code Structure (Path sequence queries)
4. Code Reviews (Impact analysis shows blast radius)

Not Ideal For

1. Quick Text Searches ("Find all TODOs" -> Use ripgrep)
2. Very Large Repos (100K+) (Speed tradeoff: 4s vs 0.3s)
3. Exact String Matching (Use ripgrep with -F)
4. Interactive File Browsing (Use fzf)

Installation

CLI Installation

Quick Start (no installation required):

npx mantic.sh@latest "your search query"

New Commands:

# Semantic Search (Neural Reranking)
npx mantic.sh@latest "verify user identity" --semantic

# Go to Definition
npx mantic.sh@latest goto "UserService"

# Find References
npx mantic.sh@latest references "handleLogin"

From Source:

git clone https://github.com/marcoaapfortes/Mantic.sh.git
cd Mantic.sh
npm install
npm run build
npm link

MCP Server Installation

Mantic works as an MCP (Model Context Protocol) server for Claude Desktop, Cursor, VS Code, and other MCP-compatible tools.

One-Click Install:

• Install in Cursor
• Install in VS Code

Manual Configuration (for Claude Desktop or other MCP clients):

Add this to your MCP settings file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "mantic": {
      "command": "npx",
      "args": ["-y", "mantic.sh@latest", "server"]
    }
  }
}

Usage

Basic Search

Find files matching your intent:

mantic "stripe payment integration"

Returns JSON with ranked files, confidence scores, and token estimates.

Advanced Usage

Zero-Query Mode (proactive context):

mantic ""
# Shows modified files, related dependencies, impact analysis

Context Carryover (session mode):

mantic "auth logic" --session my-task
# Previously viewed files get +150 boost

Impact Analysis:

mantic "payment processing" --impact
# Shows blast radius and dependents

### Session Management (CLI)

You can manage sessions directly from the terminal to persist context across multiple runs:

```bash
# Start a named session with an intent
mantic session start my-feature --intent "implement payment webhook"

# List all active sessions
mantic session list

# Get detailed info (viewed files, query history)
mantic session info <sessionId>

# End a session
mantic session end <sessionId>

### CLI Options

```bash
mantic <query> [options]

Options:
  --code          Only search code files (.ts, .js, etc)
  --config        Only search config files
  --test          Only search test files
  --json          Output as JSON (default, includes metadata)
  --files         Output as newline-separated file paths
  --markdown      Pretty terminal output
  --impact        Include dependency analysis and blast radius
  --session <id>  Use session for context carryover
  --path <dir>    Restrict search to specific directory
  --include-generated  Include generated files (.lock, dist/, etc)
  --quiet, -q     Minimal output mode
  --semantic      Enable neural reranking (slower, but "smarter")

Code Intelligence Commands:

mantic goto <symbol>        # Find definition of a symbol
mantic references <symbol>  # Find all usages of a symbol

MCP Tools

When using Mantic through MCP (Claude Desktop, Cursor):

• search_files - Primary search (supports semantic: true for neural reranking)
• get_definition - Go to definition of a symbol
• find_references - Find usages of a symbol
• get_context - Zero-query mode for proactive context
• session_start/end - Manage coding sessions
• session_record_view - Track viewed files
• session_list/info - View session history
• analyze_intent - Understand query intent

Agent Rules (Auto-Pilot)

Want Cursor or Claude to use Mantic automatically?

1. Copy the Agent Rules.
2. Paste them into your AI tool's system prompt or "Rules for AI" section.
3. The Agent will now automatically use mantic to find context before writing code.

How It Works

Architecture Overview

User Query
    ↓
Intent Analyzer (categorizes: UI/backend/auth/etc)
    ↓
Brain Scorer (ranks files using metadata)
    ↓
File Classifier (filters by type: code/config/test)
    ↓
Impact Analyzer (calculates blast radius)
    ↓
Output (JSON/Files/Markdown/MCP)

Core Algorithm (v1.0.21)

1. Intent Recognition: Analyzes query to determine code category (e.g., "auth", "ui")
2. File Enumeration: Uses git ls-files for tracked files (significantly faster than traversals)
3. Normalization & Matching:
   • CamelCase detection: "ScriptController" -> "script controller" for matching
   • Word-boundary matching: "script" won't match "javascript"
   • Path sequence matching: Multi-term queries match consecutive path components
   • Directory boosting: Single-term queries prioritize files in matching directories
4. Structural Scoring: Ranks files based on:
   • Exact filename match: +10,000 points for perfect matches
   • Path relevance: packages/features/payments indicates high signal
   • Filename matching: stripe.service.ts > stripe.txt
   • Business logic awareness: .service.ts boosted over .test.ts
   • Boilerplate penalties: index.ts or page.tsx ranked lower
5. Progressive Disclosure: Calculates metadata (size, tokens, confidence, timestamps)
6. Context Carryover: Applies +150 boost to session-viewed files
7. Learning: Caches successful patterns for future queries

Configuration

Mantic works out of the box with zero configuration for most projects.

Environment Variables

MANTIC_MAX_FILES=5000         # Maximum files to scan
MANTIC_TIMEOUT=30000          # Search timeout in ms (default: 30000)
MANTIC_IGNORE_PATTERNS=...    # Custom glob patterns to ignore
MANTIC_FUNCTION_SCAN_LIMIT=30 # Top files to scan for function names (default: dynamic, max 50)

License

Mantic.sh is released under the MIT License.

You can use, copy, modify, merge, publish, distribute, sublicense, and sell copies of the software as long as the copyright notice and license text are included.

See LICENSE for the full MIT terms.

Note: earlier releases published under AGPL-3.0 remain available under the terms they were originally shipped with. The repository is MIT-licensed going forward.