imemory

Every new session forces you to re-explain your dev environment, project context, and personal preferences. imemory solves this in DSH by turning long-term memory into two plain-text files: MEMORY.md for your notes, lessons, and environment context, and USER.md for identity and preferences. Entries are separated by a standalone section sign on its own line, freely editable at any time, with zero database dependency and effortless cross-machine migration.

Three mechanisms drive its core behavior. After each root-session turn, the LLM distills durable information from the conversation and writes it to memory with an importance score from 1 to 10. Before every dialogue step, candidates are recalled by keyword scoring, then semantically re-ranked by the LLM and injected into context. Repeated workflows crystallize into learned-* skills that build your personal playbook over time. A background consolidation pass runs every six hours to merge conflicts, deduplicate, and retire low-value entries. A web management panel lets you inspect and edit memories on the fly.

It suits DSH developers who want an assistant that continuously remembers them and gets smarter with use, especially those juggling multiple projects, switching contexts often, or looking to distill scattered experience into reusable skills. The memory format is compatible with Hermes, so existing users can point to the same directory and pick up where they left off.