Announcing TokenCap: The Local-First Context Layer for AI Coding Agents

Today, we are officially launching TokenCap: a zero-network-egress repository intelligence platform designed for developers building and pairing with AI coding agents.

What is TokenCap?

TokenCap bridges the gap between raw source code and LLM context windows. Instead of blindly dumping full files or relying on fuzzy text search, TokenCap parses your codebase into an AST symbol graph, clusters modules, and budgets context deterministically.

Core Verified Benchmarks and Capabilities

1. 12.2x Token Reduction: Compresses typical 150k-token repository payloads down to 12.5k tokens without dropping critical call signatures.
2. Even-Span Sampling: Proportional sampling across file heads, bodies, and exports that eliminates line-300 cutoffs.
3. Spatial Impact Analysis: tokencap ask impact <file>:<symbol> calculates caller blast radius and risk before an agent modifies code.
4. The Four Sovereignty Laws: Zero runtime network egress, zero API keys, no compilation build steps, and under 4 dependencies.
5. Multi-Surface Availability:
   • CLI: npx tokencap make
   • VS Code Extension: Auto-capture and spatial menus
   • Local Bridge & Companion: Direct loopback injection into ChatGPT, Claude, and Gemini
   • Agent Skill: Native discovery for OpenCode, Claude Code, Cursor, and Antigravity

Quick Start

Install globally or run directly with npx:

# Generate your repository intelligence snapshot
npx tokencap make

# Start the local loopback bridge for browser companions
npx tokencap serve --daemon

# Query the blast radius of a function
npx tokencap ask impact src/auth.js:verifyToken

Explore the documentation, read the architecture specs, and install the extensions at tokencap.vansharora.app