We're excited to announce Workspace Memories—short, authoritative team facts that load into every member's Claude Code session automatically.

The Problem: Tribal Knowledge

Every team has facts that live only in people's heads or buried in a Slack thread:

  • "We use pnpm, never npm."
  • "Our API errors always use the AppError class."
  • "Deploy Monday through Thursday—never Fridays."
  • "The auth service is owned by the platform team—ask before touching it."

New hires learn these the hard way. And your AI assistant? It has no idea. So everyone re-types the same caveats into chat, over and over, and the AI still guesses wrong half the time.

The Fix: Facts That Travel With the Workspace

A workspace memory is one short, factual statement that an admin writes once in the SuperClawd app. From then on, it loads into every workspace member's Claude Code session at startup—no copy-pasting, no reminding, no onboarding doc to chase down.

Think of memories as the small set of always-true things your team wishes the AI just knew.

How It Works

Authored Centrally, Delivered Automatically

Memories are managed in a dedicated Memories page in your workspace. Each one has a short title, a one-fact body, and an enable/disable toggle. When a member starts a Claude Code session, the enabled memories are composed into a small block and injected at the start of their session—and re-injected during long sessions so they stay salient.

The app is the single source of truth. These are team memories: members receive them read-only, and only owners, admins, and maintainers can create, edit, or reorder them.

One-Way and Advisory

Two principles make this safe and predictable:

  • One-way only. Memories are pushed down into the session. SuperClawd never reads your machine, your native Claude Code memory, or your session content. Every memory is text a human typed into the app.
  • Advisory, not enforced. A memory is team context, not a hard rule. If you explicitly ask your AI to do something different in the moment, your request always wins. A memory nudges the default; it never overrides you.

Examples

Good memories are short and factual. A few that work well:

Title: Package manager
Body: We use pnpm + turbo. Never npm or yarn.

Title: Error handling
Body: All API errors must use the AppError class with a status code and message.

Title: Commit style
Body: Conventional Commits. Prefix with feat / fix / chore / docs.

Title: Deploy windows
Body: Deploy Mon–Thu only. Never ship to production on a Friday.

Notice the pattern: each is a single, durable fact—not a paragraph, not a tutorial.

Managing Memories

The Memories page is a searchable, paginated list. For each memory you can enable/disable, edit, favorite (star), or delete it. Disabled memories aren't injected and don't count against your budget, and starred memories sort to the top so the ones you reach for most are always in view.

To keep sessions fast, memories stay short by design: a title up to 60 characters and a body up to 256. There's a per-workspace cap (1 on the free tier, 100 on paid) and a 4,000-character budget for the combined enabled set—with a meter that warns you as it fills. The editor nudges you toward short, single-fact entries, because that's what works best.

Experimental Memories

Memories carry the same experimental flag as skills, agents, commands, and workflows. Flag a memory experimental and it's injected only into experimental-mode sessions—turn on the global Experimental mode toggle in the CLI to see it on your machine, while everyone else's session skips it. There's no draft/release split for memories, so the gate is simple: trial a memory with yourself first, then unflag it to roll it out to the whole workspace.

Why Short and Factual Wins

The memory block is re-injected throughout a session to stay top-of-mind, so every byte is "paid for" repeatedly. That's by design: it keeps your team's most important facts salient. But it's also why brevity matters. One crisp fact per memory beats a wall of text—it's easier for the AI to honor and cheaper to carry.

If something is long, link to a full doc and keep the memory itself to the takeaway.

Use Cases

  • Onboarding: new hires' AI sessions already know your conventions on day one.
  • Consistency: everyone's AI follows the same package manager, error handling, and commit style.
  • Guardrails (soft): "ask before changing the auth service" shows up as context, every session.
  • Tribal knowledge: the facts that used to live in one senior engineer's head are now shared.

Get Started

Workspace memories are available now in all SuperClawd workspaces.

  1. Go to app.superclawd.com
  2. Open the Memories page in your workspace
  3. Add your first fact—keep it short
  4. Your team's next Claude Code session picks it up automatically

Want the full details on how memories are composed and injected? Read the docs at docs.superclawd.com.

Pro tip: Start with the three things you find yourself re-explaining to your AI most often. Those are your first three memories.