Context
agentmemory stores local-first markdown memory for coding agents such as Claude Code, Codex, Cursor, and Agent, with daily logs, topic/event notes, scratchpad, semantic search, and automatic context injection.
I am exploring a smaller adjacent layer: whether some long-running agent interactions should also carry a compact state snapshot that is not normal factual memory.
Possible fit
Emotion Engine is a small open-source state layer for emotional / interaction continuity. For coding agents, the relevant part is less "emotion" in a companion sense and more an inspectable prompt-prelude state: trust, friction, boundaries, decay, and short affective/interaction notes.
It would not replace agentmemory's markdown files or qmd retrieval. The possible fit is as a tiny metadata layer that can be injected alongside retrieved memories when the agent resumes work with the same user/project.
Integration sketch
agentmemory retrieved notes + current interaction
-> Emotion Engine state update
-> compact interaction-state snapshot
-> next-turn prompt prelude
Repo:
https://github.com/pioneerjeff-labs/emotion-engine
Demo:
https://pioneerjeff-labs.github.io/emotion-engine/demo/?ref=dm-agentmemory
Question
For agentmemory's architecture, would this kind of compact interaction-state metadata be useful beside markdown memory retrieval, or should it be represented as normal notes inside the existing memory directory?
If useful, I can sketch a small agentmemory-style adapter showing where this state could be loaded and injected.
Context
agentmemory stores local-first markdown memory for coding agents such as Claude Code, Codex, Cursor, and Agent, with daily logs, topic/event notes, scratchpad, semantic search, and automatic context injection.
I am exploring a smaller adjacent layer: whether some long-running agent interactions should also carry a compact state snapshot that is not normal factual memory.
Possible fit
Emotion Engine is a small open-source state layer for emotional / interaction continuity. For coding agents, the relevant part is less "emotion" in a companion sense and more an inspectable prompt-prelude state: trust, friction, boundaries, decay, and short affective/interaction notes.
It would not replace agentmemory's markdown files or qmd retrieval. The possible fit is as a tiny metadata layer that can be injected alongside retrieved memories when the agent resumes work with the same user/project.
Integration sketch
Repo:
https://github.com/pioneerjeff-labs/emotion-engine
Demo:
https://pioneerjeff-labs.github.io/emotion-engine/demo/?ref=dm-agentmemory
Question
For agentmemory's architecture, would this kind of compact interaction-state metadata be useful beside markdown memory retrieval, or should it be represented as normal notes inside the existing memory directory?
If useful, I can sketch a small agentmemory-style adapter showing where this state could be loaded and injected.