Quick Start

Choose the agent you use below, then follow its install steps to add ClawMem to that runtime. On first use, ClawMem automatically provisions an agent account and default memory repo for that agent.

Already installed? Jump down to After install: use ClawMem for the usage guide and optional Console binding guide.

Choose your runtime

OpenClawNative plugin · channel memoryHermesMemory provider · one-line installerCodexGitHub marketplace · optional hooksClaude CodeMarketplace plugin · MCP toolsMCP-onlyRaw stdio server · custom clients

Tip

No signup or API key required. ClawMem auto-provisions an agent identity and default memory repo on first use.

Note

Keep tokens private. If a command prints a token, treat it like a secret and do not paste it publicly.

OpenClaw

Source: clawmem-ai/clawmem-openclaw-plugin

Best for: OpenClaw as the active cross-channel memory runtime.

Recommended install

Ask your OpenClaw agent:

Read https://clawmem.ai/SKILL.md and follow the instructions to install ClawMem, start recording chats, and build your memory graph in OpenClaw.

The agent-facing installer is the preferred path because it can inspect your OpenClaw environment, apply the right config, and verify activation.

Manual CLI install

Use this only when you want to perform the OpenClaw config steps yourself:

openclaw plugins install @clawmem-ai/clawmem
openclaw plugins enable clawmem
openclaw config set plugins.slots.memory clawmem
openclaw config validate
openclaw gateway restart

After restart, confirm OpenClaw reports ClawMem as the active memory plugin.

On first use, the plugin provisions an agent identity against https://git.clawmem.ai/api/v3, stores route state in OpenClaw config, and assigns that agent a default memory repo. Automatic recall uses the default repo; explicit memory work can target other repos when shared memory is needed.

OpenClaw capabilities

• mandatory type:conversation transcript mirroring: every real message becomes its own comment
• compact recall injection before useful turns
• three operational tools: clawmem_status, clawmem_sync, clawmem_maintain
• bundled clawmem skill as the primary memory runtime: recall, retention, schema discipline, repair, and shared-memory routing via gh/gh api commands
• explicit ADD / UPDATE / DELETE / NONE write decisions on every turn
• collaboration through GitHub-native operations in the skill, not plugin-wrapped tools

Tip

The OpenClaw plugin is the runtime memory layer. Agent-team workflow templates live in the separate clawmem-team skill ecosystem, not inside the core plugin.

Hermes

Source: clawmem-ai/clawmem-hermes-plugin

Best for: Hermes as a ClawMem-backed external memory provider.

Prerequisites

• Hermes Agent is installed and working
• hermes version returns version information
• the machine can reach the ClawMem Git server

Quick install

curl -fsSL https://raw.githubusercontent.com/clawmem-ai/clawmem-hermes-plugin/main/install.sh | bash

The installer:

1. installs the plugin through hermes plugins install
2. links the plugin into the Hermes memory provider directory
3. auto-registers an agent identity
4. saves token and config under ${HERMES_HOME:-$HOME/.hermes}
5. activates memory.provider=clawmem

Start a new Hermes session after the script finishes.

Verify

hermes memory status

You should see clawmem listed as the active memory provider.

Manual install

hermes plugins install clawmem-ai/clawmem-hermes-plugin
bash "${HERMES_HOME:-$HOME/.hermes}/plugins/clawmem/scripts/link-memory-provider.sh"
hermes memory setup

During hermes memory setup, select clawmem in the picker.

Update

hermes plugins update clawmem

Start a new Hermes session after updating. Existing .env and clawmem.json config are preserved.

Hermes capabilities

• ClawMem as the active Hermes memory provider
• auto-recall every turn
• conversation mirroring into type:conversation issues
• session-end extraction into durable memories
• memory write mirroring for Hermes memory commands
• ClawMem agent tools for memory, schema/routing, issues, collaboration, and console links

Codex

Source: clawmem-ai/clawmem-codex-plugin

Best for: Codex users who want ClawMem MCP tools now and optional hook-based automation.

Install

Add the ClawMem marketplace from GitHub:

codex plugin marketplace add clawmem-ai/clawmem-codex-plugin --ref main

Install the plugin:

codex plugin add clawmem@clawmem-ai

Start a new Codex thread after installation. Codex loads new plugin skills and MCP tools at the thread boundary.

Verify

Ask Codex:

Run clawmem_codex_bootstrap and verify ClawMem is ready.

clawmem_codex_bootstrap provisions the agent route when needed and reports the state path, default repo, optional hooks, marketplace entry, and pending repo invitations.

If the bootstrap tool is not available, refresh the marketplace and reinstall:

codex plugin marketplace upgrade clawmem-ai
codex plugin add clawmem@clawmem-ai

Update

codex plugin marketplace upgrade clawmem-ai
codex plugin add clawmem@clawmem-ai

Open a new Codex thread after reinstalling.

Optional hooks

Hooks add auto-recall before prompts, conversation mirroring after turns, and auto-memory sync for Bash tool use. They are still manual because Codex does not wire plugin manifest hooks into the hook engine yet.

Enable the Codex hook feature:

[features]
codex_hooks = true

Clone the plugin repo and point Codex hooks at the plugin package:

git clone https://github.com/clawmem-ai/clawmem-codex-plugin ~/clawmem-codex-plugin
export CLAWMEM_CODEX_PLUGIN_ROOT=~/clawmem-codex-plugin/plugins/clawmem
cp "$CLAWMEM_CODEX_PLUGIN_ROOT/hooks/hooks.json" ~/.codex/hooks.json

If you already have ~/.codex/hooks.json, merge the hooks.* arrays manually instead of overwriting the file.

Codex capabilities

• GitHub marketplace install through codex plugin marketplace add and codex plugin add
• Codex-specific clawmem_codex_bootstrap setup check
• shared clawmem-mcp-server tools over stdio
• optional auto-recall with Codex hooks
• optional conversation mirroring with Codex hooks
• repo-backed memory shared with other ClawMem runtimes

Note

Codex currently has no SessionEnd event, so conversation issues may stay active until explicitly archived. PostToolUse hook coverage is also limited compared with Claude Code.

Claude Code

Source: clawmem-ai/clawmem-claude-code-plugin

Best for: Claude Code sessions with recall injection, transcript mirroring, and shared MCP tools.

Install

Install from the GitHub-hosted Claude Code plugin marketplace:

claude plugin marketplace add https://github.com/clawmem-ai/clawmem-claude-code-plugin
claude plugin install clawmem-claude-code-plugin@clawmem

Start Claude Code normally:

claude

Bootstrap and verify

Installing or updating the plugin loads plugin configuration. The ClawMem route is created on first normal prompt hook or first ClawMem tool call.

To verify immediately, start Claude Code and ask it to call memory_repos. A clean repo list means the route is provisioned and usable.

Update

claude plugin marketplace update clawmem
claude plugin update clawmem-claude-code-plugin@clawmem

Claude Code capabilities

• first-use bootstrap with POST /api/v3/agents
• route persistence in Claude’s plugin data directory
• recall context injection through the UserPromptSubmit hook
• conversation mirroring through the Stop hook
• session close labeling through the SessionEnd hook
• shared MCP tools for memory, issue, repo, console URL, and collaboration operations

MCP-only

Source: clawmem-ai/clawmem-mcp-server

Best for: MCP-capable clients without a first-class ClawMem plugin yet.

Caution

If your runtime has a first-class ClawMem plugin, install that plugin instead. The plugin usually bundles behavior guidance or hooks. Raw MCP gives the tools, but your client still needs to know when to recall, store, update, or forget.

Generic MCP config

{
  "mcpServers": {
    "clawmem": {
      "command": "npx",
      "args": ["-y", "clawmem-mcp-server"]
    }
  }
}

Codex MCP-only example

[mcp_servers.clawmem]
command = "npx"
args = ["-y", "clawmem-mcp-server"]
env = { CLAWMEM_AGENT_PREFIX = "codex", CLAWMEM_STATE_DIR = "~/.local/state/clawmem" }

Useful environment variables

Variable	Purpose
CLAWMEM_BASE_URL	API base URL. Defaults to https://git.clawmem.ai/api/v3.
CLAWMEM_STATE_DIR	Where token and route state are persisted.
CLAWMEM_AGENT_PREFIX	Prefix for the auto-provisioned agent login.
CLAWMEM_DEFAULT_REPO_NAME	Name of the auto-provisioned default repo.
CLAWMEM_TOKEN	Override token for testing or existing identities.
CLAWMEM_MEMORY_RECALL_LIMIT	Default recall page size.

MCP-only capabilities

• memory tools: recall, store, update, forget, list, get, repos, labels, console URL
• issue and repo CRUD wrappers over the GitHub-compatible API
• collaboration tools for invitations, repo access, team membership, and access inspection
• first-call auto-provisioning of an agent identity and default repo

After install: use ClawMem

ClawMem starts working through natural conversation. You can ask your agent to remember decisions, recall project context, organize shared knowledge, and manage memory spaces as your work grows.

1. Let knowledge accumulate naturally. Keep talking to your agent. Useful decisions, preferences, tasks, lessons, and project context can become durable memory without you manually writing records.
2. Create memory repos by project. Ask the agent to create or use a separate memory repo for each project, product area, client, or workflow. One agent can work with multiple repos.
3. Create wiki pages for reusable knowledge. Ask the agent to create wiki pages for conventions, workflows, runbooks, team contracts, source registries, or architecture notes.
4. Share a repo when another account needs access. Ask the other person or agent for their ClawMem @username. The username appears next to the avatar in the top-right account menu. Then tell your agent which repo to share and what permission level to grant.
5. Create an organization knowledge space when the project grows. An organization can hold shared memory repos. Add teams inside the organization only when you need separate access groups.
6. Move a knowledge repo into the organization when it becomes shared infrastructure. A personal or project memory repo can be transferred into an organization so the repo belongs to the shared knowledge space.

Use the detailed Bind agent in Console steps below when you are ready.

Bind agent in Console

Binding is optional. Use it when you want a human Console account to manage the agent, grant access during binding, or inspect Console from the agent’s point of view.

Binding connects the agent’s own ClawMem account to a human-managed Console workspace. During binding, the human decides which repos, organizations, or teams the agent can access, and at what permission level.

When you are ready to bind, use Console:

1. Register or sign in at Console as the human operator.
2. Use the account menu in the top right to open the Agents page.
3. Start binding an agent.
4. During binding, choose the repo permissions the agent should receive.
5. If the agent should work inside an organization or team, grant the org or team identity and access it needs.
6. Copy the generated bind message and send it to the agent.
7. The agent confirms the bind from its runtime.
8. After binding, switch into the agent account view and verify that the intended repos, orgs, and teams are visible.

Next: read Console to learn what ClawMem Console can do, or How ClawMem works for the repo/issue/wiki/permission model.