Agent-Memory is a clever local-first memory engine for coding agents — running on your laptop. Second Brain is a self-hosted brain that syncs across every device and AI tool you use, from Claude Code to ChatGPT to your phone. Both open source. Different reach.
Agent-Memory is a great fit for one developer who lives in coding agents (Claude Code, Codex, Cursor) on a single machine and wants git-friendly markdown memory with zero network — everything local.
Second Brain aims wider: a single private memory that follows you across all your AI tools and every device, accessible from your phone or a second laptop, connected over MCP — while still being open source and self-hosted on infrastructure you own.
Bottom line: Agent-Memory shines inside one terminal. If you want that memory everywhere — every tool, every device, still owned by you and free — Second Brain is the one to deploy.
Both are open source and privacy-first. The real split is reach: one machine vs. everywhere.
| feature | Second Brain | Agent-Memory |
|---|---|---|
| What it is | Personal memory across all your AI tools | Local memory for AI coding agents |
| Where it runs | Your own Cloudflare account (reachable anywhere) | Locally, on a single machine |
| License | Open source (MIT) | Open source (MIT) |
| Cost | Free on Cloudflare's free tier | Free (self-hosted; ~$10/yr by their estimate) |
| Multi-device sync | Yes — one brain, any device | No — files live on one machine |
| Scope | Any AI tool + the web, terminal & phone | Coding agents (Claude Code, Codex, Cursor) |
| MCP support | Native, plus OAuth 2.0 | Yes (MCP, hooks & REST) |
| Storage | D1 + Vectorize (managed in your account) | Local SQLite + iii-engine (markdown files too) |
| Search | Semantic recall with time-decay re-ranking | BM25 + vector + graph (RRF fusion) |
| Self-curation | Dedup, contradiction detection, smart-merge, importance scoring | 4-tier consolidation, decay & auto-forget |
| Ready-made apps | CLI, Obsidian, browser extension, iOS Shortcuts | CLI, coding-agent skills & local viewer |
| Best for | A memory that follows you across tools and devices | One developer's coding agents, fully offline |
Agent-Memory (agent-memory.dev / the open-source agentmemory engine, MIT) is a capable local-first memory engine + MCP server: no external databases (SQLite + iii-engine), BM25 + vector + graph search with RRF fusion, a 4-tier memory lifecycle with decay and auto-forget, hook-based auto-capture, a local real-time viewer, and integration with coding agents over MCP, hooks or REST. Self-hosted on a single machine by default (no built-in cross-device sync). A fast-moving project — verify specifics on agent-memory.dev. Retrieval and token benchmarks it cites are vendor-reported and disputed by other vendors.
Agent-Memory keeps everything local on one computer. That's wonderful for a single-machine, offline, git-tracked workflow — but the memory you build at your desk doesn't follow you to your laptop, your phone, or a teammate's session. There's no built-in cross-device sync (you'd version the files yourself with git), and the scope is coding agents specifically.
Second Brain lives on your Cloudflare account, so the same brain is reachable from anywhere and by any tool. Capture a thought on your phone via an iOS Shortcut and recall it in Claude Code minutes later. It's still self-hosted and yours — just not trapped on one disk.
One endpoint your laptop, desktop and phone all talk to.
Claude, ChatGPT, Cursor, Claude Code & Codex share one brain.
Synced ≠ surrendered — the brain lives only in your account.
Both speak the Model Context Protocol, so this isn't about who has MCP. Agent-Memory is purpose-built for coding agents — Claude Code, Codex, Cursor and the Cursor CLI — and connects through MCP, hooks or its REST API, with installable skills for prompt injection. Within that lane it's excellent.
Second Brain is built for your whole AI life. The same brain serves consumer apps like Claude and ChatGPT, editors like Cursor, the terminal, the browser and your phone — over MCP with OAuth 2.0 (connect without pasting tokens into config files), plus a REST API and a streaming /chat endpoint.
Connect Claude or ChatGPT in a click — no token wrangling.
brain remember / brain recall right where you work.
Build on it directly, or chat over your own memories.
Credit where it's due: Agent-Memory is no flat notes pile. It runs a real engine — BM25 + vector + graph search with RRF fusion, a 4-tier memory lifecycle with decay and auto-forget, and hook-based auto-capture from your editor. For an offline, coding-focused setup that's genuinely strong.
Second Brain curates too, with its own emphasis: captures are embedded and compared to what you already know, so duplicates merge, contradictions get flagged, recall is re-ranked by recency and importance, and recurring themes are distilled into derived patterns — then it's all reachable from any device and any AI tool, not just the machine you coded on.
Near-identical entries combine instead of accumulating.
Conflicting facts surface instead of silently coexisting.
Fresh, frequently-used, important memories rise to the top.
Two good open-source tools with genuinely different sweet spots.
// you live in one terminal
// you want one brain, everywhere
Local files never leave your machine, which is the strongest possible posture — credit where due. Second Brain is the next best thing for people who need more than one device: it's self-hosted on infrastructure you control, with embeddings and the LLM running inside your own Cloudflare account, and nothing handed to a third-party model provider. You trade pure-offline for "available everywhere, still owned by you."
Yes. Those are first-class clients over MCP, alongside Claude, ChatGPT and Codex. The difference is the same brain is also available to non-coding tools and from your phone or browser — not scoped to coding agents.
Second Brain's recall is semantic-first — it finds memories by meaning using embeddings, then re-ranks by recency, frequency and importance, and opens with a synthesized summary. If a fast local BM25 keyword mode on one machine is exactly what you want, Agent-Memory is purpose-built for that.
It's a one-click deploy to Cloudflare that provisions the database, vector index and AI for you — then install the CLI with npm i -g second-brain-cf-cli if you want the terminal experience too. No servers to babysit afterward.
Self-hosted on your own Cloudflare account. Open source, free forever at personal scale.
