Teixido-Code-Manifold (TIG) v3.4

Stop sending your entire codebase to Claude. It's causing structural hallucinations and burning your usage limit.

TIG is a Teixido Inhibition Gate for codebases. It intercepts LLM context requests, analyzes your Abstract Syntax Tree, and applies Topological Analytical Homeostasis (TAH) to compress your project down to its "Teixido Skeleton" — the 3-5% of functions that hold the actual structural load of your application.

The result: 80-95% fewer tokens sent to Claude per session, which directly stretches your Claude Code usage limit by the same factor.

---

Why TIG Beats Basic Codebase Indexers

Standard tools (like codeindex) map everything with regex and give you a "blast score" — still drowning the AI in noise.

TIG applies Weighted Topological Strain to mathematically isolate the Domination Roots:

strain(f) = log1p((in_degree × 0.6 + normalized_complexity × 0.4) × 100)

Then applies Adaptive TIG Gate (Topological Inhibitory Gating) using dynamic thresholding:

threshold = mean(strain) + ε × std_dev(strain)

Only nodes above this threshold are fed to Claude. Everything else is structural noise.

Metric	codeindex	TIG
Parsing	Regex	Python AST + robust regex
Algorithm	Raw BFS count	Weighted topological strain
Context compression	❌ No	✅ 80-95%
MCP server	✅	✅
PyPI	❌	✅
Tests	❌	✅ CI on every push
Languages	12 (regex only)	13+ (AST for Python)

---

Installation

pip install teixido-code-manifold

Or from source:

git clone https://github.com/johnvteixido/TIG.git
cd TIG
pip install -e ".[dev]"

---

Usage

CLI

# Analyze a single file
teixido-extract --file src/main.py

# Analyze a whole project
teixido-extract --dir ./my-project

# Output raw JSON
teixido-extract --file app.py --json

# Adjust TIG sensitivity (higher = fewer roots)
teixido-extract --dir . --epsilon 1.5

Python API

from core.manifold import TeixidoEngine

# Analyze source code string
result = TeixidoEngine.analyze(source_code, language="python")
print(result["domination_roots"])       # ['main', 'process_data', 'connect_sql']
print(result["compression_ratio"])      # 87.5 (%)
print(result["token_savings"])          # {'token_savings_pct': 91.2, ...}

# Analyze a whole directory
result = TeixidoEngine.analyze_directory("./my-project")
print(result["files_analyzed"])         # 47
print(result["domination_roots"])       # top load-bearing functions across all files

FastAPI (Vercel)

teixido-serve   # starts on http://localhost:8000
# or deploy to Vercel: link this repo, vercel.json handles routing

Endpoints:

• GET / — Health check
• GET /api/info — Capabilities + supported languages
• POST /api/extract — Analyze source code string
• POST /api/compress — Analyze a local directory

---

🤖 Claude Code MCP Integration

This is the killer feature. Add TIG as a native MCP server so Claude Code can call it directly during every session — automatically compressing context before Claude reads your code.

Step 1: Install TIG

pip install teixido-code-manifold

Step 2: Add to Claude Code MCP config

Edit ~/.claude/claude_desktop_config.json (or equivalent):

{
  "mcpServers": {
    "teixido": {
      "command": "teixido-mcp",
      "args": []
    }
  }
}

Step 3: Use in Claude Code

Once configured, Claude Code has access to three tools:

Tool	What it does
extract_manifold	Analyze a source code string → Domination Roots
compress_codebase	Analyze your whole project → architectural skeleton
get_token_savings	Show exactly how many tokens + % you're saving
store_memory (Pro)	Save a durable fact/decision to project-scoped long-term memory
search_memory (Pro)	Recall stored facts across sessions (survives context compaction)
delete_memory (Pro)	Prune outdated memories by id

Example Claude Code prompt:

"Before we start, run compress_codebase on ./src and tell me the Domination Roots."

Claude will call TIG, get back the 3-5 structurally critical functions, and work only from those — using a fraction of your hourly token budget.

---

Token Savings & Usage Limit Impact

Source Size	Without TIG	With TIG	Savings
500-line file	~2,000 tokens	~80 tokens	96%
2,000-line file	~8,000 tokens	~120 tokens	98.5%
50-file project	~80,000 tokens	~1,000 tokens	98.75%

If Claude Code limits you to, say, 50,000 tokens/hour — with TIG, that same budget covers the equivalent of 1,000,000+ tokens of raw codebase, effectively making the hourly limit invisible for normal development workflows.

---

Supported Languages

Language	Parsing Method
Python	✅ stdlib ast (full AST, exact line numbers)
JavaScript	Robust multi-pattern regex
TypeScript	Robust multi-pattern regex + /// references
Go	Multi-line import block regex
Rust	use, mod, extern crate patterns
Java	import statement regex
Kotlin	import statement regex
Swift	import statement regex
C / C++	#include directive regex
PHP	require, use, namespace regex
Ruby	require, require_relative regex
C#	using directive regex

---

Architecture

teixido-code-manifold/
├── core/
│   ├── languages.py      # Multi-language parser (AST + regex)
│   ├── manifold.py       # Weighted Topological Strain engine
│   └── mcp_server.py     # MCP stdio server (Claude Code integration)
├── api/
│   └── index.py          # FastAPI serverless entry point (Vercel)
├── cli.py                # Console entry points
├── tests/                # pytest suite
├── pyproject.toml        # PyPI packaging config
└── vercel.json           # Vercel deployment config

---

License

Proprietary Software - All Rights Reserved.

Architect & Creator: John V. Teixido © 2026 tigcode.app Unauthorized copying, reverse-engineering, distribution, or commercial use of this file or any part of the TIG codebase (including the client engine and the server nodes) is strictly prohibited.