Skip to content

Willder documentation

Your first useful answer.

Save a decision in one Claude Code session. Start a new session and it already knows. One scope, one token.

What you see after sign-up

  1. Choose Start free on willder.ai. Sign up with Google, or with an email and a password. Email sign-up sends you a 6-digit code to confirm.
  2. You land on Dashboard. A Set up your workspace window may ask four questions about your work. You can skip it.
  3. If Dashboard asks you to create an organization, choose Set up your workspace, give it a name and skip inviting people. You are its admin.
  4. Open Access in the left menu. Its Tokens tab is where your token comes from. A token is shown once, when you issue it.

1Make one scope

A scope is the part of memory an agent can reach. Open Access, then Scopes, and choose + New scope. Name it Coding, leave the parent scope blank and choose Create scope.

2Issue one token and connect

Open Access, then the Tokens tab, and choose + Issue token:

  • Agent: Other client (not an agent here), Client ID claude-code.
  • Scope: Coding. Permission: write.
  • TTL hours: 720, which is 30 days. The default, 168, ends after 7 days.

Copy the token. Then follow Connect Claude Code, steps 1 to 3: load the token, connect, and add the session hook. Using another MCP client? Connect an AI agent covers it, but only Claude Code gets the hook.

3Save it, then start a new session

Start claude in your project and say:

Session 1 · Say this to Claude Code
Record this as a decision in Willder: we store sessions in Postgres, not Redis, so they survive a restart.

Claude calls memory_ingest. Willder’s reply starts with “Recorded decision” and gives Claude a link to show you. Open it, signed in to Willder, and choose Confirm decision. Until a person confirms it, a new session is told the decision is a proposal to check, because an agent recorded it.

Type /exit, then start claude again. Without naming Willder, ask:

Session 2 · Ask Claude Code
Should we store sessions in Redis or Postgres?

The hook hands Claude your decision before it reads the question, so it should answer Postgres, with your reason. That is the product: a decision made once reaches the next session without anyone repeating it.

A decision you confirm is a standing rule. Every new session in Coding gets it until a newer decision replaces it. Swap the example for a real decision from your project, or ask Claude to record a new decision that supersedes this one when you are done.

Next: share Coding with a second agent

In Access, then Tokens, issue a second token: scope Coding, permission read, a short lifetime such as 24 hours. Give it to another agent. It can read the decision but not write to Coding.

To take it back, choose revoke on that token. Its next request returns HTTP 401 with “Invalid or unknown capability token.” Revoking does not remove what the agent already read. See access and lifecycle for permissions and lifetimes, and the tools reference for every MCP tool.