Set up Remember

Install from a local Remember checkout. That checkout contains the integration, the embedded daemon manager, and the API the daemon runs. Keep it in a stable location.

Prerequisites

Git, Node.js, npm, uv, and Python 3.11 or newer. The checkout pins its development Python version in .python-version. On macOS, a current Rust toolchain may also be needed for Python dependencies.
Codex CLI or Claude Code. Both clients can share the same Remember checkout and database.
A configured memory model. You can use your signed-in Codex or Claude Code CLI, use Ollama on the same device, or supply a hosted provider's credentials.

Install for Codex

Run from the directory where you keep your checkouts. If you've already cloned Remember, start at cd remember/integrations/agents.

git clone https://github.com/nazmirket/remember.git
cd remember/integrations/agents
npm ci
npm run build
node dist/installer.js install codex

Or ask Codex to help with setup:

Set up Remember for Codex from https://github.com/nazmirket/remember. Reuse my checkout if present. Show the hook and MCP changes and how to test recall in a fresh session. Ask which memory model to use.

Check the installer's output, make sure Codex hooks are enabled, and review any hook trust prompt when you start a fresh session. The installer records the checkout's embedded package path.

Install for Claude Code

If you already installed for Codex from this checkout, run only the last line from integrations/agents.

git clone https://github.com/nazmirket/remember.git
cd remember/integrations/agents
npm ci
npm run build
node dist/installer.js install claude-code

Or ask Claude Code to help with setup:

Set up Remember for Claude Code from https://github.com/nazmirket/remember. Reuse my checkout if present. Show the hook and MCP changes and how to test recall in a fresh session. Ask which memory model to use.

Review the changes to Claude Code's hooks and MCP configuration, then start a fresh session. For either harness, uninstall with node dist/installer.js uninstall codex or node dist/installer.js uninstall claude-code from the agent integration directory.

Choose a memory model

Set provider variables in the environment that launches your agent. To use a signed-in CLI explicitly, set REMEMBER_API_LLM_PROVIDER to openai-codex for Codex or claude-code for Claude Code. The selected CLI needs to be on your agent's PATH. Without an explicit provider, Remember can discover supported API keys or a CLI; if you have both CLIs and want a specific one, choose it explicitly.

For Ollama, run it on your device, pull the model you want, and export the settings before starting your agent. For example:

ollama pull llama3.2
export REMEMBER_API_LLM_PROVIDER=ollama
export REMEMBER_API_LLM_MODEL=llama3.2

If the daemon is already running, stop it safely before starting the agent with the new settings. For hosted models, use the provider settings and keep API keys out of project files and conversation. A CLI adapter can still use a hosted model. See the inference and storage boundary.

Try a handoff

Open your agent in a Git project. Give it a small, distinctive fact, such as “For this project, the smoke fixture is named billing-v2.”
Let the turn finish, then open a new session in the same project or a linked worktree.
Ask “Which fixture did I say to use for the smoke test?” You can use the other agent if you installed both.
If recall is empty, ask the agent to run remember_sync_status and remember_diagnose, or follow these checks.

The first session may need extra time to start the daemon and load models. Recall is best effort, so use a distinctive detail and verify the original evidence before acting on a high-impact decision.