Knowledge Bases for AI Agents: Recall, Validate, and Write Back
Use ChatCrystal MCP tools and task memory to build scoped, revisable agent memory, separating real tool capabilities from automation promises.
Notes on frontend engineering and AI coding: problems, implementation choices and lessons learned.
Use ChatCrystal MCP tools and task memory to build scoped, revisable agent memory, separating real tool capabilities from automation promises.
Compare keyword search, vectors, and knowledge graphs. Trace one-hop candidate expansion in ChatCrystal and identify boundaries relationships cannot prove.
Trace storage, embeddings, summaries, and agent recall in ChatCrystal to identify which data boundaries local-first architecture actually controls.
Freeze and release the AudioContext timeline as streamed frames arrive, so frame availability drives playback while audio and video remain synchronized.
Prevent WebSocket, EventEmitter, and DOM listeners from resubscribing on every callback change, using a ref proxy or React 19.2 useEffectEvent.
How to collapse concurrent 401 responses into one token refresh, retry queued requests safely, and settle every waiting Promise when refresh fails.
Explore ChatCrystal, developer tools and daily learning notes on my GitHub. Questions about the implementation are welcome.