Your AI tools have memory. But memory is not a brain.
Remember.md is a second brain plugin for OpenClaw and Claude Code. It organizes decisions, people, projects, and tasks from your AI sessions — past and future — into a structured, Obsidian-compatible knowledge base that travels with you across tools.
Free. Local. Open source. Portable.
Every AI tool now has memory — flat notes you can't search, browse, or take with you. Remember builds something different: a structured second brain with people, projects, decisions, and tasks connected via wikilinks.
| Built-in memory | Remember.md | |
|---|---|---|
| Structure | Flat key-value pairs | People, Projects, Notes, Tasks, Journal |
| Connections | None | [[wikilinks]] across all files |
| Browsable | No | Obsidian vault with graph view |
| Portable | Locked to one tool | One brain, every AI tool |
| Past sessions | No | Process months of history retroactively |
| Your patterns | No | Persona.md learns your code style |
openclaw plugins install @remember-md/remember
/remember:initRun these from your terminal (not from inside Claude Code):
claude plugin marketplace add remember-md/marketplace
claude plugin install remember@remember-mdThen start Claude Code and run /remember:init to create your second brain structure and configure permissions.
Important — use the terminal, not the
/pluginslash command. An unrelated plugin also namedrememberexists in the defaultclaude-plugins-officialmarketplace. The/plugin install remember@remember-mdslash command currently ignores the@remember-mdqualifier and installs the wrong one. The terminal command (claude plugin install remember@remember-md) honors the marketplace and installs correctly. If you already got the wrong one, remove it withclaude plugin uninstall remember@claude-plugins-official.
~/remember/
├── REMEMBER.md # Your custom rules (you edit this)
├── Persona.md # Your patterns + Top Beliefs (AI auto-manages)
├── People/ # One note per person (type: observation)
├── Projects/ # Active work with logs and tasks
├── Notes/ # World-facts, decisions, beliefs (typed)
├── Journal/ # Daily notes (YYYY-MM-DD.md)
├── Tasks/ # Focus + Next Up priorities
├── Areas/ # Ongoing responsibilities
├── Resources/ # Links, articles, references
├── Inbox/ # Quick capture
├── Templates/ # Note templates
└── Archive/ # Completed projects
All files use YAML frontmatter + [[wikilinks]] — Obsidian-native, browsable in any markdown editor.
Plugin state (audit log, thresholds) lives at ~/.local/state/remember/ — outside the brain, never touched by Obsidian.
| Command | What it does |
|---|---|
/remember:init |
Initialize your second brain structure (run once) |
remember this: ... |
Instant capture — routes to the right place automatically |
/remember:status |
Show brain stats — file counts, freshness, top beliefs |
/remember:process |
Optional. Extract knowledge from past AI sessions into your brain |
/remember:evolve |
Periodic (weekly recommended). LLM-driven brain evolution — entity profile synthesis, belief confidence re-scoring, stale detection. Schedule via /loop 7d /remember:evolve |
Run /remember:process and recover months of lost knowledge from past OpenClaw and Claude Code sessions:
Found 47 unprocessed sessions.
✓ Extracted People/sarah-chen.md
✓ Extracted Notes/decision-database.md
✓ Created 12 journal entries
✓ Updated Tasks/tasks.md (+8 tasks)
✓ Updated Persona.md (learned your patterns)
Say "remember this: met with Sarah, decided to use Postgres for ACID compliance" and Remember routes it:
- Person →
People/sarah.md - Decision →
Notes/decision-database.md - Task →
Tasks/tasks.md
Persona.md evolves with you — code style, naming conventions, review preferences, communication patterns. Loaded automatically every OpenClaw and Claude Code session so your AI knows how you work.
Captured facts aren't static. The brain has three layers and an epistemic schema (type / freshness / confidence / evidence) that lets it grow:
- L1 — Capture —
Inbox/,Journal/,Tasks/. Append-only, raw. - L2 — Curate —
Notes/,People/,Projects/,Areas/. Each fact is tagged asworld-fact,belief,observation, orexperience, with evidence and a freshness trend. - L3 — Pinned —
Persona.md. Always loaded; auto-managed top beliefs.
Evolution happens in two layers:
Always-on (zero effort, runs at capture time):
- Auto-promote runs after every capture —
Persona.md ## Top Beliefsis always in sync with the strongest beliefs in your brain. Pure deterministic, zero LLM cost. - Contradictions are detected and routed to
counter_evidence:in real time during capture (Step 3.6 of therememberskill). Freshness flips tocontradictedautomatically when counter outweighs supporting evidence. - Bootstrap thresholds make sure your very first explicit capture lands in Persona, even before you accumulate the standard threshold of evidence.
Periodic (run /remember:evolve, weekly recommended):
- Consolidate — re-synthesize People/Project/Area profiles from accumulated mentions. Without this, entity files become log-only and never get the overview that makes them useful.
- Reflect — re-score belief confidence based on accumulated evidence vs. counter-evidence; mark beliefs
staleafterstale_days(default 90) without new evidence. - Promote — also runs here, usually a no-op because auto-promote already kept Top Beliefs current.
The always-on layer keeps the brain functional from day one. The periodic layer is what makes the brain actually evolve over time — refined beliefs, synthesized profiles, retired stale ideas. Skipping /remember:evolve indefinitely means the brain accumulates without rephrasing itself.
/loop 7d /remember:evolve
/loop is built into Claude Code — no extra plugin or cron daemon needed.
- OpenClaw — full support (plugin + hooks + agent tools)
- Claude Code — full support (hooks + skills)
- Cursor / Codex — planned
One brain, shared across all tools. Knowledge captured in OpenClaw is available in Claude Code and vice versa.
The plugin ships with its own REMEMBER.md (default rulebook). You customize behavior by adding sections to your own REMEMBER.md:
~/remember/REMEMBER.md— your global preferences./REMEMBER.md— project-specific rules (layers on top, when cwd is in a project)
For each section in your REMEMBER.md:
- Same name as a default (e.g.
## Routing) → your content is appended to the default. ## Override: <Name>→ your content fully replaces the default for that section.- Any other name → passed through verbatim at the end as a user-defined section.
Available default sections (see plugin's REMEMBER.md): Routing, Task Routing, Processing, Linking, Multi-File Updates, Format, Capture Rules, Custom Types, Language, After Save.
For full documentation, see REMEMBER.md Guide.
A few knobs live outside REMEMBER.md because they control plugin runtime, not capture routing. Edit config.defaults.json (plugin-wide) or pass overrides via ~/.local/state/remember/config.json:
| Key | Default | What it does |
|---|---|---|
session.passive_nudge |
true |
At session start, ask the AI to detect — semantically, in any language — when you reveal a stable preference, decision, or recurring pattern, and to suggest capture once per topic with 💡 Want me to remember this?. The AI never auto-saves; you confirm with the next message. Set to false to require explicit remember this: for every capture. |
session.load_persona |
true |
Inject Persona.md at every session start. Disable if you want a clean context. |
thresholds.promotion_confidence |
0.85 |
Confidence floor for a belief to pin into Persona.md ## Top Beliefs. |
thresholds.promotion_sources |
5 |
Independent-source floor for the same. |
bootstrap |
true |
Cold-start mode: while total beliefs < 20, relax thresholds to conf ≥ 0.7 and sources ≥ 1 so the very first explicit capture lands in Persona. Set false to apply normal thresholds from day one. |
- Local markdown files — nothing leaves your machine
- No cloud, no telemetry, no tracking
- Git-friendly — version control your entire brain
- No vendor lock-in — works with Obsidian, Logseq, any editor
- Portable — one brain across every AI tool
Q: How is Remember different from OpenClaw memory or Claude MEMORY.md? A: Built-in memory stores flat notes locked inside one tool. Remember builds a structured second brain — People, Projects, Decisions, Tasks, Journal — connected via wikilinks and browsable in Obsidian. It processes past sessions retroactively and is portable across AI tools.
Q: Can it process old sessions?
A: Yes. Run /remember:process to scan past OpenClaw and Claude Code sessions and extract decisions, people, tasks, and insights into your knowledge base. Works on sessions from months ago.
Q: Can I use it with both OpenClaw and Claude Code? A: Yes. Both plugins point to the same brain directory. Knowledge captured in one tool is available in the other.
Q: Do I need Obsidian? A: No, but Obsidian gives the best experience — graph view, backlinks, search. Remember creates Obsidian-native markdown that works in any editor.
Q: How does it learn my coding patterns? A: Persona.md captures your code style, naming conventions, and workflow preferences over time. It's loaded at the start of every session so your AI knows how you work.
Q: Do I need to run /remember:evolve manually?
A: The deterministic parts (promoting top beliefs, flipping freshness on contradictions) run automatically after every capture — that part is hands-off. But the LLM-driven parts (re-synthesizing People/Project/Area profiles, re-scoring belief confidence, marking stale beliefs) only happen when you run /remember:evolve. Recommended cadence is weekly via /loop 7d /remember:evolve. Skip it indefinitely and the brain accumulates without ever rephrasing itself — usable, but stagnant.
Q: What does /remember:evolve actually do when I run it?
A: Three phases. (1) Re-synthesizes entity profiles (People/Projects/Areas) from accumulated mentions. (2) Re-scores belief confidence based on evidence vs counter-evidence and marks freshness (stable/strengthening/weakening/stale/contradicted). (3) Promotes top beliefs into Persona.md ## Top Beliefs — usually a no-op because auto-promote already ran at capture time. Configure thresholds in ~/.local/state/remember/config.json or your REMEMBER.md.
Q: Will the brain overwrite my hand-curated Persona?
A: No. The plugin only manages ## Top Beliefs (by reference — wikilinks, not copies). Mission, Directives, and Evidence Log stay in your hands. To turn off auto-promotion entirely: set auto_promote: false in ~/.local/state/remember/config.json.
Q: Where's the audit trail?
A: ~/.local/state/remember/evolution.log — append-only, ISO-timestamped, one line per auto-change. Tail it whenever you want to see what the brain did on its own. Run /remember:status to see the latest entries inline.
Q: How much does it cost? A: Free, always. MIT licensed, open source.
- OpenClaw or Claude Code (latest version)
- Node.js (bundled with Claude Code; required for OpenClaw)
- Git (optional, for version control)
Built on ideas from:
- continuous-learning-v2 — Hooks architecture
- PARA Method (Tiago Forte) — Organization structure
- Zettelkasten (Niklas Luhmann) — Linked thinking
MIT — see LICENSE.
Remember.md — One brain. Every AI tool. Star on GitHub