Knowledge-and-Memory-Management v0.0.2: Portable Agent Memory Release

Version 0.0.2 of Knowledge-and-Memory-Management introduces three changes aimed at simplifying agent development. First, a unified Collector interface now handles web pages, video subtitles/transcripts, and PDF/markdown articles through a single fetch-parse-chunk-embed pipeline, storing vectors with source provenance for traceability. Second, memory is split into short-term (working context) and long-term (vector-indexed) tiers, with an automatic consolidation step that merges short-term entries into long-term storage once they hit a relevance threshold, using cosine similarity and timestamp-aware scoring to avoid duplication.

The third and most developer-facing change is full portability: all file paths and database locations now resolve against a single $AGENT_HOME environment variable, eliminating hardcoded paths that previously differed across dev, test, and production. This aligns the system with twelve-factor app principles, and a bundled migration script rewrites existing databases to the new layout while preserving embeddings and metadata.

For engineers building context-aware agents, the practical benefits are fewer moving parts—one import instead of separate scrapers and parsers, no custom dedup/decay logic thanks to automatic consolidation, and no need for symlinks or volume mounts to keep paths consistent across CI and containerized deployments. Known limitations include the lack of native handling for JS-rendered dynamic pages and reliance on external subtitle or transcript sources for video, while long-term memory compression is slated for v0.2.0.