𝗞𝗠𝗠 𝘃𝟬.𝟬.𝟮: 𝗕𝗨𝗜𝗟𝗗𝗜𝗡𝗚 𝗔𝗜 𝗔𝗚𝗘𝗡𝗧 𝗞𝗡𝗢𝗪𝗟𝗘𝗗𝗚𝗘 𝗣𝗜𝗣𝗘𝗟𝗜𝗡𝗘𝗦
Large context windows do not solve everything.
If your AI Agent starts every chat from zero, you have a problem. It forgets yesterday's research, last week's PDFs, and last month's competitor analysis.
Most people try to fix this with RAG and vector databases. They fail because they put low-quality data into the system. The problem is not how you search. The problem is how you collect data.
Knowledge-and-Memory-Management (KMM) v0.0.2 fixes the intake process. It provides a framework to manage over 40 collection tools.
KMM organizes tools into four categories:
• Web: 6 engines for extraction, including tools to bypass Cloudflare and Chrome DevTools automation. • Video: 8 tools for metadata, subtitles, and ASR. Includes yt-dlp and Whisper for transcription. • Content: 10+ sources like RSS, blogs, and news aggregators. • Documents: Full support for PDF, PPT, and Word using high-precision OCR.
This is not just a list of scripts. It is a five-step pipeline: Collect → Analyze → Note → Graph → Cloud.
A key feature is the AugmentedSearch logic. It prevents your Agent from lying to you.
The system searches your local notes first. If the local match score meets your threshold, it returns the result. If the score is too low, it triggers a web search. It marks web results clearly as "source: web." This stops the Agent from pretending web results are your own private knowledge.
The NoteGenerator follows a clear path:
- Collect raw material via collectors.
- Use an LLM to structure arguments, data, and timelines.
- Write to local Markdown notes.
- Map data into a knowledge graph.
- Sync to cloud storage like OneDrive via rclone.
You can even turn a PDF into a usable skill. The refine_pdf tool converts a document into a structured note and an AI skill automatically.
If your Agent keeps forgetting what you told it, do not buy a new database. Build a better collection pipeline. KMM provides the orchestration framework to do exactly that.
Optional learning community: https://t.me/GyaanSetuAi