Hey PR, I've been building "toolkits" with high-level API functions for Claude Code to use. During these builds, I kept running into the same problem: persistent memory is messy. Most solutions use .md files that bloat token usage every time the LLM reads them. So I came up with Living JSON; a simple pattern with two arrays: old and new. Fresh data goes into new, then rotates to old on the next scan. Clean, fast, and queryable by any LLM without the token bloat. What Endpoints does: - Extract structured data from images, PDFs, docs - Store original files in S3 - Auto-generate REST API endpoints - Query your data without re-parsing No more dumping everything into markdown files. Clean JSON endpoints your agent can query. Would love to hear how others are handling persistent memory for their AI agents. Have you run into the .md problem? What'd you do to solve it?

Endpoints

Turn your files into Clawdbot memory with JSON endpoints

Turn your files into Clawdbot memory with JSON endpoints