Evermind Ai Agent Memory

Evermind

**Cross-session memory recovery for AI agents. Stop your assistant from forgetting — without re-reading everything every session.**

   

Every new session feels like "day one at work"? Your agent doesn't know who you are, what the rules are, or where work left off — so you re-explain everything, and it still drops things.

Evermind hands your agent a **shift handover** before it starts working:

• Always reads the core layer — identity, rules, user profile, latest todos, latest work log (nothing important silently dropped)
• Checks a change index (auto-generated, hash-based) before re-reading secondary files — unchanged files cost ~0
• Defers everything else until it is actually needed

Measured on our own production system: recovery drops from **~44K to ~12K tokens** (~70% less). When the host already injects identity memory (e.g. Hermes), the duplicated read is skipped automatically — **~55-75% cumulative savings** (measured 2026-09-04).

Quick start

# 1. get the skill
git clone https://github.com/ccy123abcd/evermind-ai-agent-memory.git
cp -r evermind ~/.claude/skills/          # or your agent's skills dir

# 2. configure
cd evermind && cp config.example.yaml config.yaml
#   edit config.yaml: memory_root + must_read (L3) + tracked_files (L2)

# 3. generate the change index (schedule it daily)
python scripts/memory_index.py --config config.yaml

# 4. self-test (3 scenarios: first run / unchanged / changed)
python scripts/memory_index.py --config config.yaml --demo

Then, at the start of each new session, in