ContextSieve

Control EXACTLY what your agent knows when the context window is about to overflow!

ContextSieve offers multiple tools for transparent context window management on Gemini chats, offering a pin-based UI/UX system per prompt/response that anyone can understand. Define a context limit, and you'll get timely alerts and options to optimize the context stack as it fills up. Try it out now!

Key Features

• Live Token Telemetry: Track token breakdowns across the system prompt, user inputs and assistant outputs.
• Near-Capacity Monitoring: Sends alerts before a user-defined context limit.
• AI Context Summarizer: Compress old turns into executive summaries.
• Sliding Window Pruning: Retain recent chats and preserve pinned facts.
• Active Context Toggling: Omit verbose response blocks individually from the context.
• Inline Editing: Modify context blocks directly inside the active window.

Just one round of summarization + trimming filler text, and ContextSieve saved 79.9% of the context window (see below)!

Quickstart

1. Clone repo:

   git clone [https://github.com/zh-yng/ContextSieve.git](https://github.com/zh-yng/ContextSieve.git)
   cd ContextSieve

2. Install deps:

   npm install

3. Create .env or .env.local and grab an API Key from Google AI Studio:

   GEMINI_API_KEY="your_api_key_here"

4. Run app:

   npm run dev

5. Open browser:

   http://localhost:3000