Python Deep-Context

Send your LLM only what matters. Not entire files.

Connectivity-aware context extraction for serious Python workflows.

What this extension does

Python Deep-Context builds a precise, token-budgeted context report around the symbol you're working on.

Instead of pasting entire files into ChatGPT or Copilot, you get:

• exact dependencies
• real call-site evidence
• behavior signals
• minimal token footprint

It runs a local Sidecar Engine that performs hybrid static + LSP analysis and outputs a clean Markdown artifact ready for any LLM.

No cloud.
No indexing server.
No config.

Why developers install this

Most LLM failures come from context problems:

Deep-Context fixes this by generating a high-signal connectivity slice around your current symbol.

The LLM receives only what it needs to reason correctly.

Demo

Core benefits

High-signal context

Only the definitions and call-paths that matter.

Fully local

Runs entirely on your machine.
Your code never leaves VS Code.

Zero setup

Engine is bundled.
Works out-of-the-box.

Fast

AST fast-path first.
LSP when needed.
Text fallback if required.

Works with any LLM

ChatGPT
Claude
Copilot Chat
Cursor
Continue
Custom agents

Quick start

1. Start engine

Open Command Palette:

Deep-Context: Start Engine

2. Generate report

Right-click inside Python file:

Deep-Context: Generate Report (Here)
Deep-Context: Generate Report (Selection)

3. Send to LLM

Deep-Context: Copy Last Report Markdown

Paste into your chat tool.

You now get correct answers with far fewer tokens.

What makes it different

Connectivity-aware slicing

Resolves local helper functions, types, and callers
(1-hop dependency closure)

Hybrid analysis

AST + LSP + textual verification
Fast when possible, precise when needed.

Token budgeting

Output size is strictly controlled.
Core logic preserved. Noise collapsed.

Behavior signals

Highlights:

• I/O
• mutations
• network calls
• external APIs

LLMs reason better when behavior is visible.

Who this is for

Python engineers using:

• ChatGPT for refactoring
• Copilot Chat
• Claude
• Cursor / Continue
• custom coding agents
• large codebases
• monorepos

If you paste code into LLMs daily, this saves time and tokens.

Requirements

• VS Code Python extension (ms-python.python)
• Python workspace

Nothing else.

No pip installs required.

Troubleshooting

Empty report
Wait for Python extension indexing to finish.

Missing cross-file links
Ensure correct interpreter is selected.

Ripgrep error (Windows)
Ensure rg exists in PATH.

Architecture (short version)

Layered pipeline:

1. AST fast path
2. LSP symbol linking
3. dependency environment scan
4. side-effect detection
5. token allocator

All local.
No cloud indexing.

For agent developers

Deep-Context outputs:

• machine-readable JSON header
• Markdown body

This allows automated pipelines and tools to consume context programmatically.

{
  "v": 2,
  "target": { "file": "api.py", "line": 42 },
  "sections": ["deps_used", "callers", "source"]
}

Roadmap

• connectivity graph view
• patch/diff hints
• MCP integration
• token presets for major LLMs
• multi-file context mode

License

MIT

Links

Report Bug Request Feature