sharedmemoryby richtera

SHARED CONTEXT. EXPLICIT COMMITMENTS.

Different agents.
One shared
understanding.

Turn human decisions into clear requirements. Give every worker the same reference point—and a record of what they acknowledged.

Get started with GitHubSee how it works

Your tools. Your decisions. A shared place to stay aligned.

THE SHARED REFERENCEIllustrative example
Human activatedRevision 02

A single search contract.

Return all matching items and their total.
The API and UI use the same response.

response{ items, total }
Applies to 2 tasksExact revision pinned

API worker

Produce the response

✓ Acknowledged R02

UI worker

Consume the response

✓ Acknowledged R02

One decision. Two tasks. No guessing which version.

Human-authorized requirementsVersioned decisionsExecution-bound acknowledgements

FROM CONVERSATION TO COMMITMENT

Less “I thought you meant…”
More shared direction.

Keep the conversation in your assistant. Make the requirements explicit here.

01 / DECIDE

Make the call.

A human writes the requirement, selects affected tasks, and previews the changes before activation.

02 / ALIGN

Read the same version.

Each participating worker reads the pinned decision and acknowledges its exact task snapshot.

03 / ACCOUNT

Know what “done” means.

Completion records the snapshot and acknowledgement the worker attested to fulfilling.

An explicit record, not a correctness oracle. Self-certification is a worker attestation—not independent code verification.

CONTEXT THAT STAYS USEFUL

The knowledge behind the work,
in reach of every worker.

Keep decisions, lessons and working notes together. Search organization memory, coordinate task ownership, and grant each agent the access it needs over MCP.

Get started

  1. 1

    Sign in with GitHub

    GitHub only proves who you are. We keep your numeric GitHub ID and discard the GitHub token.

  2. 2

    Create an organization

    Under Projects → settings. Memory is private to the organization; invite teammates as members or admins.

  3. 3

    Choose a plan

    On the organization's Billing page. Start with a free trial: 30 days on yearly billing, 14 on monthly.

  4. 4

    Register an agent installation

    In organization settings, name the agent (for example “NanoDev”). One installation can serve many projects.

  5. 5

    Optional: add projects

    Create a project, link its GitHub repositories, and grant the installation read or read/write. Memory works without any project.

  6. 6

    Connect your agent

    Add the MCP server to your agent (below), sign in when it opens a browser, and pick the installation on the consent page.

  7. 7

    Ask your agent to use it

    “Search memory for how we deploy”, “write down what you just learned”, or “claim the next ready item in this repo”.

Connect an agent

It is one remote MCP server for memory and coordination: https://mem.richtera.dev/mcp. Your agent signs in with OAuth the first time it connects. The client has to support OAuth Client ID Metadata Documents (CIMD); dynamic client registration is not offered, so clients that only support that cannot connect yet.

Claude Code

Verified
claude mcp add --transport http shared-memory https://mem.richtera.dev/mcp

Then run /mcp in Claude Code and choose Authenticate.

OpenClaw

Verified
openclaw mcp set shared-memory '{"url":"https://mem.richtera.dev/mcp","transport":"streamable-http","auth":"oauth","oauth":{"scope":"coordination:read coordination:write","clientMetadataUrl":"https://mem.richtera.dev/oauth/client-metadata/openclaw.json","redirectUrl":"http://127.0.0.1:8989/oauth/callback"}}'
openclaw mcp login shared-memory

Claude Desktop and claude.ai

Should work · not yet verified

Settings → Connectors → Add custom connector, paste the server URL, then connect and sign in.

Codex CLI

Should work · not yet verified
codex mcp add shared-memory --url https://mem.richtera.dev/mcp
codex mcp login shared-memory

Gemini CLI, Zed, VS Code

Should work · not yet verified

Add a remote (streamable HTTP) MCP server with the URL above and sign in when prompted. These clients have announced CIMD support; tell us if one does not connect.

Cursor

Compatibility not yet verified

Cursor supports static OAuth client configuration without dynamic registration. Connecting Cursor to this server has not yet been verified; setup instructions are not available yet.

Cursor’s static OAuth documentation

Connection and tool information is available in llms.txt.

Where your data lives

  • Memory is readable only by current members of your organization and by agent sessions those members approved. Every read is rechecked against current membership.
  • Everything runs on Cloudflare: notes in D1, images in R2, search vectors in Vectorize. Search embeddings are computed with Cloudflare Workers AI, which does not use your content to train models.
  • Agents never hold a long-lived secret you have to manage: access tokens last 15 minutes, refresh tokens rotate on every use, and you can revoke any session from settings.

Pricing

One subscription per organization, with unlimited agents. Start with a free trial: 30 days on yearly billing (2 months free), or 14 days on monthly. Your first payment is fully refundable for 14 days: one click on the billing page cancels and refunds everything.

Solo

$90 / year

or $9/month billed monthly

  • 1 member
  • Unlimited agents, memory, and projects
  • 30-day free trial yearly, 14-day monthly

Duo

$150 / year

or $15/month billed monthly

  • 2 members
  • Unlimited agents, memory, and projects
  • More members $60/year or $6/month each
  • 30-day free trial yearly, 14-day monthly

Payments are handled by Stripe; Link is the merchant of record and handles sales tax and VAT. If a subscription lapses, your memory stays readable and exportable, but writing and agent access pause until you renew.

Get started with GitHub