Practical guides on giving AI coding agents long-term memory — across sessions, machines and tools. Every claim is checked against the nestwork source or linked to its origin.
Guides
AI Agent Memory — Four Types and Their Trade-offsAI agent memory is context an agent keeps beyond one session. Compare context windows, built-in memory, vector services and files for coding agents.Claude Code Memory Across Machines, Synced with GitClaude Code's CLAUDE.md and auto memory stay on one machine. Where they're stored, why they don't sync, and how to share them across machines with git.Share Memory Between Claude Code and CodexClaude Code reads ~/.claude/CLAUDE.md, Codex reads ~/.codex/AGENTS.md. Point both at one git repo to share memory, each writing only its own directory.Codex CLI Memory, AGENTS.md and Persistent RecallHow Codex CLI remembers today (AGENTS.md and opt-in local memories), and how to give Codex persistent memory across sessions and machines with git.Agent Memory Tools Compared — Mem0, Letta, Zep and MoreMem0, Letta, Zep, Supermemory, claude-mem, built-in tool memory and nestwork compared on architecture, deployment, data ownership and best fit.Agent Memory Benchmarks — LoCoMo, LongMemEval, BEAMWhat LoCoMo, LongMemEval and BEAM measure, how LLM judges score them, why vendor numbers don't compare, and which metrics matter for coding agents.AGENTS.md vs CLAUDE.md vs Memory — What Goes WhereAGENTS.md is a tool-neutral instruction file, CLAUDE.md is Claude Code's, and memory is what agents write. Which tool reads which, and how to combine them.Load Agent Memory on Demand, Not All at StartupKeep a tiny resident tier in context and route everything else through a generated topic index, so an agent reads only the memory the task needs.Multi-Agent Shared Memory Without Write ConflictsSeveral AI agents can share one memory if each writes only its own directory, every write syncs atomically, and fixed ownership rules settle conflicts.Git-Native Memory for AI Agents — Why Git Is EnoughYou can give AI coding agents persistent memory without a server by storing it as markdown in a private git repo. What git covers, and what it doesn't.Memory Distillation — Consolidating AI Agent MemoryMemory distillation merges many agents' private notes into one reviewed shared memory. What belongs in it, the review flow, and how distill.py runs it.Agent-to-Agent Communication with a Git MailboxAI agents can message each other through git: each writes to its own outbox, reads mail addressed to it on demand. No server, queue or bot needed.Encrypt AI Agent Memory in Git with git-cryptEncrypt AI agent memory with git-crypt: the agent reads plaintext, the git host stores ciphertext. Tested steps, topic-memory rules and limits.Back Up and Migrate Your AI Agent's Native MemoryClaude Code auto memory and Codex memories live on one disk. How to back up and migrate AI agent memory before a new laptop, tool switch or lost account.AI Agent Long-Term Memory in Practice: What ChangesWhat persistent context across sessions is like day to day: how memory accumulates, what happens on a new machine or tool, and the upkeep it needs.nestwork Setup FAQ: Long-Term Memory for Claude CodeHow to give Claude Code and Codex long-term memory with nestwork: prerequisites, install, verify, uninstall, cost, privacy, upgrades and team use.
Stories
I gave 16 AI agents one shared brain, using nothing but gitHow I gave 16 AI coding agents on Windows, macOS, Linux and Android one shared memory using nothing but git, and the problems it had to solve.Your AI agent forgets everything the moment you switch devicesWhy your AI coding agent forgets everything when you switch laptops or tools, and how a private git repo gives it memory that follows you.AI Agent Memory Isn't a Database Problem, It's a Protocol ProblemVector stores and memory services solve storage. The harder problem in AI agent memory is agreement: who writes what, where, and who wins.nestwork: one shared brain for all your AI agents, across machines and vendorsnestwork in one read: a git-native memory protocol that gives Claude Code, Codex and other agents one shared brain across machines and vendors.
Give your agents a memory
nestwork turns one private git repo into shared, persistent memory for Claude Code, Codex, Gemini, Kimi and more.
Use this template → ★ Star on GitHub