Claude Code config
Pi Mem
Plain-Markdown persistent memory for AI coding agents. Long-term facts, daily logs, scratchpad, and semantic search — works with pi, Claude Code, and any LLM agent.
Install
git clone https://github.com/jo-inc/pi-mem.gitFrom the README
pi-mem surprisingly useful daily memory for the pi coding agent Read the blog post Inspired by OpenClaw's approach to agent memory. Layout Memory files live under ~/.pi/agent/memory/ (override with PI_MEMORY_DIR): Identity and behavioral files (e.g. SOUL.md, AGENTS.md, HEARTBEAT.md) can also live in the memory directory and be injected into context via PI_CONTEXT_FILES. Tools Context Injection The default system mode keeps MEMORY.md in a cached system-prompt section, condensed to about 3K tokens (a longer file is hard-cut at a line boundary with an ellipsis and a pointer to memory_read(target='long_term')), followed by the recall directive. Per turn it adds one trailing message, only when there is something to say: a retrieval block from other memory files, similar past conversations, and stored facts (capped near 3K tokens; whole items dropped, never fragments, with a memory_search pointer when anything was cut), plus today's and yesterday's daily logs with the latest entries kept (2,000-char cap, memory_read(target='daily') pointer). The long prompt prefix stays cacheable; only the small trailing message varies. Other modes: - bootstrap — a per-turn recall block including MEMORY.md itself, injected as a user message every turn (no cached section) - full — inject context files, MEMORY.md, today's and yesterday's logs, and catchup indexes every turn - off — inject nothing SCRATCHPAD.md, files in notes/, and older daily logs are always available on demand through the memory tools. Session Ancestry (opt-in) Pi records session lineage but offers no easy way to read it: a forked or cloned session stores a parentSession path in its JSONL header, and entries within one session link via parentId. An agent only gets PI_SESSION_FILE at runtime, so it has no simple…
Details
- Repository
- jo-inc/pi-mem
- Type
- Config & memory
- Language
- TypeScript
- License
- MIT
- Last updated
- 2 November 2026
- Topics
- agent-memoryai-agentcoding-agentcontext-managementllmmarkdownmemorypi-coding-agenttypescript
FAQ
What is Pi Mem?
Plain-Markdown persistent memory for AI coding agents. Long-term facts, daily logs, scratchpad, and semantic search — works with pi, Claude Code, and any LLM agent. It is a config for Claude Code, from jo-inc/pi-mem on GitHub.
How do I install Pi Mem?
Clone jo-inc/pi-mem from GitHub and follow the setup steps in its README.
Is Pi Mem popular?
jo-inc/pi-mem has 79 GitHub stars and 12 forks. It was last updated on 2 November 2026.
More config & memory
Contextual Memory Virtualisation
83IngressionLabs/contextual-memory-virtualisation
Contextual Memory Virtualisation (CMV): git-like snapshots, branching and trimming for AI coding agent context. Demonstrated with Claude Code.
AI Config Sync Manager
47slash9494/ai-config-sync-manager
Two-way sync of Claude Code and Codex instructions, agents, hooks and permissions.
Dot Claude
83PaulRBerg/dot-claude
PRB's Claude Code config
Claude Workflow
75sbusso/claude-workflow
Claude Code configuration for SDLC guidance
Clausona
57larcane97/clausona
Switch between multiple Claude Code and OpenAI Codex CLI accounts with one command. Skills, hooks, plugins and settings stay shared, and conversation history can be too. clausona can also pick the account with quota left for each run, and show the skills, MCP servers and hooks every account loads.
Claude Config
74timholy/claude_config
Configuration files for claude code
Browse the most popular Claude Code mods, all config & memory, or the skills and MCP server directories.