
Anneal Memory
This implements episode storage with compression and citation-based validation to help agents build continuity without accumulating noise. You get four tools: record observations and decisions as typed episodes, recall by keyword or type, prepare_wrap to fetch uncompressed episodes for LLM summarization, and save_continuity to validate and store the compressed output with hash-chained audit trails. The validation layer checks for cited evidence and flags patterns without episode support. Reach for this when you need persistent agent memory that degrades gracefully instead of drifting into hallucinated patterns. Ships as a Python library first with CLI and MCP transports over the same core. Works with LangGraph, CrewAI, OpenAI Agents, or any framework that can call record and recall hooks around agent loops.
- GitHub stars
- 8
- tools
- 17
- transport
- STDIO
- last updated
- Apr 1, 2026
Tools
17 toolsPublic tool metadata for what this MCP can expose to an agent.
recordRecord something important that happens, such as a decision, observation, tension, question, outcome, or context change.
Record something important that happens, such as a decision, observation, tension, question, outcome, or context change.
No parameter schema in public metadata yet.
recallRecall prior memory before making decisions that might have prior context, querying by time, type, keyword, or ID.
Recall prior memory before making decisions that might have prior context, querying by time, type, keyword, or ID.
No parameter schema in public metadata yet.
prepare_wrapAt session end, returns episodes, current continuity, and compression instructions.
At session end, returns episodes, current continuity, and compression instructions.
No parameter schema in public metadata yet.
save_continuityAfter compressing, validates the structure and citations of continuity and saves it.
After compressing, validates the structure and citations of continuity and saves it.
No parameter schema in public metadata yet.
delete_episodeRemoves content that should not exist, such as PII or sensitive data.
Removes content that should not exist, such as PII or sensitive data.
No parameter schema in public metadata yet.
statusChecks memory health, including episode counts, wrap history, continuity size, and association formation state.
Checks memory health, including episode counts, wrap history, continuity size, and association formation state.
No parameter schema in public metadata yet.
wrap_cancelAbandons a wrap that is open and will not be finished.
Abandons a wrap that is open and will not be finished.
No parameter schema in public metadata yet.
crystal_recallRecalls crystallized (graduated, long-term) patterns relevant to a query.
Recalls crystallized (graduated, long-term) patterns relevant to a query.
No parameter schema in public metadata yet.
crystal_indexProvides the always-on name and one-clause menu of the crystallized store.
Provides the always-on name and one-clause menu of the crystallized store.
No parameter schema in public metadata yet.
spore_addAdds an open loop to the prospective (spore) layer.
Adds an open loop to the prospective (spore) layer.
No parameter schema in public metadata yet.
spore_listLists spores in the prospective (spore) layer.
Lists spores in the prospective (spore) layer.
No parameter schema in public metadata yet.
spore_surfaceSurfaces spores from the prospective (spore) layer.
Surfaces spores from the prospective (spore) layer.
No parameter schema in public metadata yet.
spore_getGets a single spore from the prospective (spore) layer.
Gets a single spore from the prospective (spore) layer.
No parameter schema in public metadata yet.
spore_updateUpdates a spore in the prospective (spore) layer.
Updates a spore in the prospective (spore) layer.
No parameter schema in public metadata yet.
spore_touchTouches a spore in the prospective (spore) layer.
Touches a spore in the prospective (spore) layer.
No parameter schema in public metadata yet.
spore_descendDescends a spore in the prospective (spore) layer.
Descends a spore in the prospective (spore) layer.
No parameter schema in public metadata yet.
spore_ascendAscends a spore in the prospective (spore) layer.
Ascends a spore in the prospective (spore) layer.
No parameter schema in public metadata yet.
Sponsors
Readme
anneal-memory logo
anneal-memory
Living memory for AI agents. Episodes compress into identity.
Memory without grounding is amplification infrastructure.
Persistent user memory profiles increase agent sycophancy 16–45% across models (Gemini 2.5 Pro at 45%, others lower). Production deployments accumulate 97.8% junk entries within weeks. Clinical research documents memory scaffolding delusions across sessions. The failure mode here isn't memory. It's memory with nothing checking what gets kept.
anneal-memory puts a check between what an agent writes down and what it comes to believe. That idea isn't rare any more. MemTxn refuses a fact update whose values don't appear in the source it cites, and Agent Zero Memory answers only from evidence its reader actually opened.
Details
- Repository
- github.com/phillipclapham/anneal-memory
- Category
- AI & LLM Tools
- Package
anneal-memory- Transport
- STDIO
- Registry
- MCP registry, active