[User Query + Files]
        │
        ▼
┌──────────────────────────────────────┐
│     MCP Server: fittok    │
│                                      │
│  ┌──────────┐                        │
│  │ Graphify │ → parse code into      │
│  └────┬─────┘   knowledge graph      │
│       │                               │
│  ┌────▼─────┐                        │
│  │  slurp   │ → select relevant      │
│  └────┬─────┘   subgraph (budget)    │
│       │                               │
│  ┌────▼──────┐                       │
│  │LLMLingua │ → compress to target   │
│  └──────────┘   token count          │
└──────────────────────────────────────┘
        │
        ▼
[Compressed Context] → Send to LLM

# From source
pip install -e .

# With all extras
pip install -e ".[dev,gpu,ui]"

export FITTOK_MODEL="microsoft/phi-2"  # GPU recommended
export FITTOK_DEVICE=auto  # auto | cuda | cpu
python -m fittok

from fittok.llmlingua_wrapper import compress_context
result = compress_context(text, "question", target_tokens=500, model="microsoft/phi-2")

{
  "mcpServers": {
    "fittok": {
      "command": "python",
      "args": ["-m", "fittok"]
    }
  }
}

python -m fittok

from fittok.graphify import parse_codebase, save_graph
from fittok.slurp import query_graph
from fittok.llmlingua_wrapper import compress_context

# Step 1: Parse codebase
graph = parse_codebase("/path/to/code")
save_graph(graph, "graph.json")

# Step 2: Query for relevant subgraph
markdown, count, tokens = query_graph(graph, "How does auth work?", token_budget=4000)

# Step 3: Compress
result = compress_context(markdown, "How does auth work?", target_tokens=500)
print(result["compressed"])

from fittok.server import optimize_context_tool

result = optimize_context_tool(
    codebase_path="/path/to/code",
    query="How does authentication work?",
    token_budget=500,
)
print(result["optimized_context"])

from fittok.server import optimize_context_batch

result = optimize_context_batch(
    codebase_path="/path/to/code",
    queries=["How does auth work?", "What is the entry point?"],
    token_budget=500,
)
for r in result["results"]:
    print(f"Q: {r['query']}\nA: {r['optimized_context']}\n")

from fittok.pii_scrubber import scrub_text

result = scrub_text("Contact admin@company.com with key AKIAIOSFODNN7EXAMPLE")
print(result["scrubbed"])
# "Contact [REDACTED_EMAIL] with key [REDACTED_AWS_ACCESS_KEY]"

pip install -e ".[dev]"
pytest tests/ -v

fittok/
├── pyproject.toml
├── src/fittok/
│   ├── __init__.py
│   ├── server.py              # MCP server (FastMCP)
│   ├── graphify.py            # Code → knowledge graph
│   ├── slurp.py               # Graph query engine
│   ├── llmlingua_wrapper.py   # Compression wrapper (CPU + GPU)
│   ├── models.py              # Pydantic data models
│   ├── tokens.py              # Shared token counting
│   ├── cache.py               # 3-level persistent cache
│   ├── diff.py                # Graph diffing
│   ├── pii_scrubber.py        # PII detection & redaction
│   ├── watcher.py             # File watcher for incremental updates
│   └── ui.py                  # Web visualization (Gradio + pyvis)
├── tests/
│   ├── test_graphify.py
│   ├── test_slurp.py
│   ├── test_llmlingua.py
│   ├── test_server.py
│   ├── test_server_v2.py
│   ├── test_cache.py
│   ├── test_diff.py
│   └── test_pii_scrubber.py
├── examples/
│   └── usage.py
├── README.md
└── LICENSE