Adaptive Summarization System + Memory Pipeline Integration#110

Open

basnijholt wants to merge 37 commits into

mainfrom

Commits on Apr 22, 2026

update docs/architecture/memory.md
basnijholt
committed
Turn off ChromaDB telemetry
basnijholt
committed
feat(memory): add output validation with ModelRetry for reconciliation
basnijholt
committed
feat(memory): add self-model features to AI journal POC
basnijholt
committed
docs: add detailed comparison between AI journal POC and aijournal
basnijholt
committed
feat(memory): add adaptive summarization with hierarchical storage
basnijholt
committed
refactor(summarizer): improve code quality and add Letta-style features
basnijholt
committed
refactor(summarizer): replace class with functional API
basnijholt
committed
refactor(summarizer): make internal modules private and simplify public API
basnijholt
committed
refactor(memory): wire AdaptiveSummarizer into memory pipeline
basnijholt
committed
docs: add summarizer spec and update memory docs for hierarchical summaries
basnijholt
committed
Add example script
basnijholt
committed
refactor(summarizer): YAGNI cleanup and fix prior_context bug
basnijholt
committed
feat(cli): add summarize command for adaptive hierarchical summarization
basnijholt
committed
refactor(memory): remove dead parent_group field and bundle metadata args
basnijholt
committed
perf: lazy imports for pydantic_ai, sounddevice, and numpy
basnijholt
committed
refactor: reduce duplication in memory store and summarizer
basnijholt
committed
refactor: simplify docstrings and remove unused upsert_hierarchical_summary
basnijholt
committed
fix(summarizer): strip special tokens from LLM output
basnijholt
committed
docs: correct Mem0 attribution in summarizer documentation
basnijholt
committed
fix(memory): summarize raw conversation turns, not extracted facts
basnijholt
committed
docs: clarify research foundations vs original design in summarizer
basnijholt
committed
refactor(summarizer): simplify to NONE/BRIEF/MAP_REDUCE levels
basnijholt
committed
refactor(summarizer): consolidate shared code to reduce duplication
basnijholt
committed
refactor(summarizer): remove redundant _summarize_text and safety guard
basnijholt
committed
refactor(summarizer): remove redundant exception re-wrapping
basnijholt
committed
refactor(summarizer): remove defensive guards for impossible conditions
basnijholt
committed
feat(scripts): add summarizer comparison script with needle-in-haystack test
basnijholt
committed
docs(summarizer): update architecture doc to reflect current implementation
basnijholt
committed
docs: update memory.md for 3-level summarizer
basnijholt
committed
refactor(summarizer): rename STANDARD_SUMMARY_PROMPT to GENERAL_SUMMARY_PROMPT
basnijholt
committed
docs: clarify prompt comments to avoid confusion with level names
basnijholt
committed
Chunk memories
basnijholt
committed
refactor(summarizer): remove dead code and reorganize models
basnijholt
committed
refactor(summarizer): simplify API with target_tokens/target_ratio parameters
basnijholt
committed
chore(summarizer): remove dead code
basnijholt
committed
fix(summarizer): persist final summaries as L3 entries
basnijholt
committed