This release is focused on one thing: making TokenCap scale to the kinds of codebases people are actually working with.
When I first built TokenCap, the goal was simple—stop AI coding assistants from wasting tokens rediscovering the same project context on every prompt.
As more people started using it on larger repositories and monorepos, new bottlenecks became obvious. v1.9.0 is all about removing them.
What's new
Built for scale
- Bounded parallel graph processing using Node.js worker threads
- Checksum-protected atomic sharded cache
- ~55s cold indexing on a deterministic 100,000-file codebase
- ~44s warm rebuilds
- ~57ms scoped incremental updates
Smarter monorepo support
- Cross-package dependency mapping
- Package-scoped context generation with
--package - Automatic discovery of package-local AI onboarding files and rules
Safe Markdown compression
The new tokencap compress command reduces snapshot size by removing markdown noise while keeping every fenced code block byte-for-byte identical.
No broken examples. No mangled code. Just fewer tokens.
Infrastructure-aware context
TokenCap now understands Docker Compose services and local Terraform modules, allowing infrastructure to become part of the project graph your AI can reason about.
Still local-first
No telemetry.
No analytics.
No network calls.
Your code never leaves your machine.
If you're using Cursor, Claude Code, Copilot, Gemini CLI, or any other AI coding workflow, better context usually beats longer prompts.
Install:
```bash id="buwr3q"
npm install -g tokencap
Generate your first context capsule:
```bash id="h1zkxv"
tokencap make --full
Website
https://tokencap.vansharora.app
Documentation
https://tokencap.vansharora.app/docs
GitHub
https://github.com/vansharora7/tokencap
npm
https://www.npmjs.com/package/tokencap
VS Code Marketplace
https://marketplace.visualstudio.com/items?itemName=VanshArora.tokencap
Open VSX
https://open-vsx.org/extension/VanshArora/tokencap
As always, feedback, issues, and feature requests are welcome.
Top comments (0)