I built an MCP server that gives AI assistants persistent memory across sessions. No more re-explaining your projects every time you start a new chat.

  Elara Core runs locally, stores everything in ChromaDB vector databases on your machine, and exposes 34 tools through
  the Model Context Protocol (MCP).

  What makes it different from basic RAG/memory:

  - Emotional state tracking — mood decays naturally between sessions using exponential decay toward a learned
  temperament baseline
  - Correction system — save mistakes, they get surfaced semantically before you repeat them
  - Dream mode — weekly/monthly pattern discovery across sessions (inspired by sleep consolidation in neuroscience)
  - Reasoning trails — track hypothesis chains when debugging, including what was abandoned and what triggered the
  breakthrough
  - Episodic memory — sessions are typed (work/drift/mixed) with mood sampled throughout
  - Session handoff — structured carry-forward with overdue detection for items postponed 3+ times

  Technical details:

  - 19K+ lines of Python, 91 tests
  - 7 ChromaDB collections (cosine similarity)
  - Pydantic schemas, atomic writes, mixin composition for large modules
  - Zero cloud dependencies — everything in ~/.elara/
  - BSL-1.1 license (free for personal use, converts to Apache 2.0 after 4 years)

  Built it because I wanted my AI to feel like the same entity across sessions, not a stranger every time. One week of
  intense development, 70+ sessions.

  pip install elara-core

  Docs: https://elara.navigatorbuilds.com
  GitHub: https://github.com/aivelikivodja-bot/elara-core
  Architecture: https://elara.navigatorbuilds.com/architecture.html