From plur
Persistent memory for AI agents using open engram format. Learns from corrections, remembers across sessions, and transfers knowledge across domains via automatic injection and explicit learn/forget commands.
How this skill is triggered — by the user, by Claude, or both
Slash command
/plur:plur-memoryThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Persistent memory for AI agents. Corrections, preferences, and patterns are stored as **engrams** that strengthen with use and decay when irrelevant. The system gets smarter the longer you use it.
Persistent memory for AI agents. Corrections, preferences, and patterns are stored as engrams that strengthen with use and decay when irrelevant. The system gets smarter the longer you use it.
Always. Memory is not a feature you toggle — it's a layer that runs continuously.
The plugin automatically injects relevant engrams into every conversation turn via the pre_llm_call hook. You don't need to call plur_inject manually unless you want full hybrid search (the automatic path uses fast BM25 search).
<plur-memory> blocksplur_learn with a clear statementplur_learn immediately with the correctionplur_feedback with signal "positive"plur_feedback with signal "negative"plur_forget with the engram IDEnd your responses with a learning section when you discover reusable insights:
---
🧠 I learned:
- Insight one (min 10 characters)
- Insight two
The plugin auto-captures these — no manual plur_learn call needed. This is a convenience fallback; calling plur_learn directly is preferred for important learnings.
On first install, PLUR has zero engrams — injection returns empty. This is expected.
Your first 5 sessions are the bootstrap period. Actively learn:
plur_learn for every correction the user makesplur_learn for stated preferences ("always use X", "never do Y")plur_learn for discovered patterns and conventionsAfter ~20 engrams, injection starts returning useful context automatically. To accelerate, install a community pack via plur_packs_install.
Periodically run plur_extract_meta to distill cross-domain principles from your engrams.
The extraction is a multi-turn conversation:
plur_extract_meta — returns analysis prompts with "status": "prompts_ready"plur_meta_submit_analysis with your responses as {"responses": [...]}{"status": "complete"}If you call plur_meta_submit_analysis with no active pipeline, you'll get {"status": "no_active_pipeline"} — call plur_extract_meta first.
Meta-engrams are the highest-value knowledge: principles that transfer across domains.
npx claudepluginhub plur-ai/plurExtracts durable learnings (corrections, conventions, preferences) at session end and saves them as structured engrams for future reuse.
Implements CoALA memory architecture for Claude: stores working, episodic, semantic, procedural memories in filesystem for recall, reflection, learning across sessions.
Persists learnings into a 5-layer memory hierarchy (CLAUDE.md files, memory/MEMORY.md) and consolidates by pruning outdated entries and promoting recurring patterns. Triggers on 'extract learnings', 'remember', 'dream'.