Tutorial 00: Check your durmem connection

durmem gives coding agents shared, durable memory. Decisions, instructions, and project knowledge survive between sessions and return when relevant. This tutorial checks that the connection you set up works end to end, then opens the practice repository that later tutorials use.

Start with the Quickstart. It installs the durmem command, runs durmem setup, and saves and recalls a first memory. This page assumes you finished it. If you skipped it, do the Quickstart first; this page does not repeat its install steps.

Addresses on this page use durmem.ai, your durmem server's address. On the durmem website it is filled in for you. If you are reading the raw file, replace durmem.ai with your server's host name (for Durmem, durmem.ai), including any :port your administrator gave you.

What `durmem setup` connected

durmem setup adds three pieces to your coding harness (Claude Code, Codex, or OpenCode). Each one has its own check below, because one can work while another does not:

Skills teach the agent how to use memory well, for example when to save a decision and how to read project context.
The MCP connection is the bridge the agent uses to read and write your stores.
Lifecycle hooks run automatically at points such as session start and each prompt, and bring relevant memory into the conversation without you asking.

A store is one memory boundary on the server. A new account starts with a few: a coding store for real project memory, loom_demo for practice, and kanban for optional agent coordination. Accounts created before 2026-10-02 may not have a ready kanban store. This page does not need it. ~/.loom/stores.toml (on Windows, %USERPROFILE%\.loom\stores.toml) records how this computer reaches them. It holds credentials, so never paste it into chats, issues, commits, or screenshots.

1. Open the tutorial repository

The coding tutorials use the public palmerpenguins Python repository. It includes package code, tests, and bundled CSV data, so you do not need access to a private repository. Clone it into a folder named demo:

git clone https://github.com/mcnakhaee/palmerpenguins.git demo
cd demo

Open the cloned demo folder as your agent's workspace, or start your agent from that directory. Keep that workspace open as you continue through the tutorials.

YouConfirm that this workspace is the cloned palmerpenguins repository. Show the repository root and configured Git remote, but do not modify any files.

AgentThe agent checks the workspace root and Git remote without writing to the repository.

> Illustrative agent output: This workspace is the cloned palmerpenguins repository. Its root is the open demo folder, and its configured remote points to the public upstream repository. No files were changed.

Only continue after the repository is open in your agent. It gives durmem a real project for the memory exercises that follow.

2. Check that the hooks run

MCP calls can work while hooks are disabled, so a store listing alone does not prove the hooks are on. Check them directly.

For Codex, restart Codex fully or open a fresh task after setup. If Codex asks you to trust a durmem plugin hook, confirm that it comes from loom@loom and choose the persistent Trust/Enable action. Trust and enable all six durmem hook events: SessionStart, UserPromptSubmit, PreToolUse, PostToolUse, PreCompact, and PostCompact. In the Codex desktop app, review them under Settings → Hooks.

For Claude Code, start a fresh session, then open the read-only hook browser:

/hooks

In /hooks, verify that durmem appears with source Plugin under SessionStart, UserPromptSubmit, PreToolUse, PostToolUse, and PreCompact. You can also check them in the Claude desktop app under Settings → Hooks. Do not launch Claude Code with --bare, and make sure disableAllHooks is absent or false.

Then start another fresh session in the tutorial repository. Its opening context should include durmem's standing-memory summary. Submit an ordinary prompt and confirm that the harness reports the durmem prompt hook running. If either check fails, stop: re-enable the plugin, complete the trust step for your harness, and check again.

3. Check that the agent can see your stores

Store names come from your account. Ask the agent to list them rather than assuming any name:

YouUse durmem to list every store this computer is connected to. Show each store's exact name and description, and tell me which one is the default. Do not load or modify a store yet.

AgentThe agent asks durmem for the configured stores and reports their names and descriptions only, never credentials.

> Illustrative agent output: The connection is working. Here are the exact store names and descriptions on this computer. The default is <exact name>, the practice store is loom_demo, and the coordination store is kanban. I will use these exact names in later prompts.

If the list is empty, a store you expect is missing, or the connection to your stores fails, sign in again with durmem login (for a private server, durmem login --base https://durmem.ai) and repeat this check. Login signs in to the same server and account. Use durmem setup only when a skill or the MCP integration is missing, and for a private server pass the same --base https://durmem.ai to it. Do not substitute a store just because its name looks close.

Make `loom_demo` this session's working store

The tutorials write to the practice store, loom_demo, so nothing real is at risk. Each tutorial starts by making it the working store for that session:

YouLoad the configured store named loom_demo and make it the default for this session only. Confirm the effective session default, and do not edit ~/.loom/stores.toml.

AgentThe agent connects to the existing loom_demo store and changes only this session's default.

> Illustrative agent output: loom_demo is now the default for this session. Persistent configuration was not changed.

A new session goes back to the default in stores.toml until you repeat this prompt.

4. Check the coding-memory skill

The loom-coding skill is the main coding-agent discipline: it loads relevant project context, checks standing instructions, and saves durable decisions as you work. This check separates "the agent can reach a store" from "the agent knows how to use project memory safely."

YouCheck that the loom-coding skill is installed and can load its current guidance. Do not write a fact yet.

AgentThe agent checks the available skills, loads loom-coding, and confirms that its guidance is readable.

> Illustrative agent output: The loom-coding skill is available and active. I can use its read and save discipline in the repository tutorials.

If the skill is missing, run durmem setup again and choose your harness. For a private server, add --base https://durmem.ai.

5. Takeaways

Finish the Quickstart first; this page checks the result rather than repeating the install.
Skills, the MCP connection, and lifecycle hooks are separate pieces; check each one.
A working store list does not prove the hooks run.
Ask durmem for your exact store names instead of guessing.
Make loom_demo the session's working store before each tutorial exercise.
Treat ~/.loom/stores.toml as private.
Continue to Tutorial 01: Build your first memory.