ParseCTX

Your codebase is millions of tokens. Your LLM isn’t. ParseCTX decides what actually matters.

ParseCTX is a deterministic codebase snapshot and intelligent context engine.
It can flatten your entire project, track changes, search everything instantly, and assemble relevance-ranked context windows for LLMs.

Core Capabilities

Snapshot Layer (Deterministic Foundation)

• build — Flatten entire codebase into a numbered snapshot
• change — Detect diffs using Myers algorithm
• search — Full-text search across all files
• stat — File stats, breakdowns, token estimates

Context Intelligence Layer

• pack — Build token-budgeted, relevance-ranked LLM context
• inspect — Analyze dependency graph and file importance
• init — Configure project defaults

Why This Exists

| Problem | Typical Approach | ParseCTX | |--------|----------------|----------| | Codebase too large | Paste random files | Packs only relevant context | | LLM confusion | Flat dumps | Structured, ranked context | | Finding dependencies | Manual grep | Import graph traversal | | Token waste | Include everything | Budget-aware packing | | Tracking changes | Git diff noise | Clean, line-level output |

Quick Start

npx parsectx build

or

npm install -g parsectx

Commands

parsectx build

Creates a deterministic snapshot:

• Flattens all files into parsectx/build.txt
• Adds line numbers
• Stores SHA-256 hashes in meta.json
• Uses atomic writes

parsectx change

Detects changes since last snapshot:

• New files
• Modified files
• Deleted files
• Line-level diff (Myers algorithm)

Output:

parsectx/change.txt

parsectx search <term>

Search entire codebase:

parsectx search "TODO"

Options:

• --case, -c — Case-sensitive
• --max=N — Limit results

parsectx stat

Displays snapshot analytics:

• File counts
• Output sizes
• Directory breakdown
• Top file types

Intelligent Context Engine

parsectx pack

Builds a token-budgeted context window for LLMs.

parsectx pack --budget 100000 --focus src/index.js --depth 2

Process

1. Parses imports (JS, TS, Python, Go, Rust, Java)

2. Builds a dependency graph

3. Scores files using:

   • PageRank-style centrality
   • Recency
   • Focus proximity
   • Size penalty

4. Packs highest-value files first

5. Summarizes large low-priority files

6. Stops at token budget

Output Formats

• XML (Claude)
• Markdown (ChatGPT)
• JSON (pipelines)

parsectx inspect

Analyze codebase structure:

parsectx inspect --top 20
parsectx inspect --graph src/index.js

Outputs:

• Most important files
• Dependency graph
• Structural insights

parsectx init

parsectx init

Creates:

.parsectx/config.json

How It Works

your-project/
├── src/
├── package.json
└── parsectx/
    ├── build.txt
    ├── change.txt
    ├── pack.md / pack.xml
    └── meta.json

Snapshot Engine

• Recursive scan
• Binary detection (null-byte heuristic)
• Deterministic ordering
• Atomic writes

Context Engine

• Dependency graph construction
• Centrality scoring
• Token-aware packing
• Relevance-first selection

Default Exclusions

• node_modules
• .git
• parsectx
• .parsectx
• venv

Binary files are skipped automatically.

Use Cases

• LLM context generation (Claude, GPT, Gemini)
• Full-repo search without tooling overhead
• CI snapshot artifacts
• Deterministic project archiving
• Understanding large codebases
• Fast iteration without Git diff noise

Requirements

• Node.js >= 18
• No external dependencies

Philosophy

ParseCTX follows a simple principle:

Context is only useful if it is relevant and constrained.

License

MIT © Sagnick Das