Inspect, back up, search, export, and update OpenClaw long-term memory stored with MemoryOS. Use when Codex needs to manage MemoryOS memory files for an Open...
Memory
Memkeeper for OpenClaw
Try itInstall, configure, verify, or remove local-first Memkeeper memory for OpenClaw. Use when an OpenClaw user needs Memkeeper's local embedding and reranking mo...
What it does
Install, configure, verify, or remove local-first Memkeeper memory for OpenClaw. Use when an OpenClaw user needs Memkeeper's local embedding and reranking models, a safe MCP tool profile, or diagnosis of a failed Memkeeper MCP connection.
The skill document
Memkeeper for OpenClaw
Configure Memkeeper as a local stdio MCP server. Keep embedding and reranking local by default. Do not select an OpenClaw reasoning model, configure a model provider, install a Gateway service, or enable a messaging channel unless the user explicitly asks.
Set up local semantic memory
Use a user-level store unless the user requests project isolation.
MEMKEEPER_BIN="${MEMKEEPER_BIN:-$HOME/.local/bin/memkeeper}"
MEMKEEPER_STORE="${MEMKEEPER_STORE:-$HOME/.memkeeper/store.sqlite}"
If MEMKEEPER_BIN is absent, download the public installer to a temporary file,
inspect it, and run it. Do not paste secrets into a command or config file.
curl -fsSL https://raw.githubusercontent.com/teflon07/memkeeper/main/install.sh \
-o /tmp/memkeeper-install.sh
bash /tmp/memkeeper-install.sh
Download the local models, initialize the store, and require semantic retrieval:
"$MEMKEEPER_BIN" pull-models
"$MEMKEEPER_BIN" init --store "$MEMKEEPER_STORE" --json
MEMKEEPER_REQUIRE_SEMANTIC=1 \
"$MEMKEEPER_BIN" doctor --store "$MEMKEEPER_STORE" --json
Require the final doctor output to report semantic models as ready. Treat a missing model as a setup failure, not as permission to use lexical fallback.
Register the MCP server
Register Memkeeper through OpenClaw's MCP command. Expose the tested 12-tool profile first:
openclaw mcp add memkeeper \
--command "$MEMKEEPER_BIN" \
--arg mcp \
--env "MEMKEEPER_STORE=$MEMKEEPER_STORE" \
--env "MEMKEEPER_REQUIRE_SEMANTIC=1" \
--include "stats,search,get,memory_list,entity_search,graph_neighbors,graph_context,dream_graph,remember,pack,candidate_submit,candidate_list" \
--timeout 30 \
--connect-timeout 30
Keep forget, entity_upsert, relationship_upsert, and verify disabled
until the operator explicitly needs deletion or direct graph/metadata mutation.
Verify and report
Prove the live stdio connection after registration:
openclaw mcp doctor memkeeper --probe --json
Report the configured store path, enabled tool count, and probe result. Do not print stored memory contents, API keys, tokens, or other sensitive environment values.
The MCP connection test does not require an OpenClaw reasoning provider, a Gateway daemon, or messaging channels. Those are independent, user-selected choices.
Handle non-local embedding only on request
Do not change the local mxBai embedding and reranking defaults without explicit operator direction. An API embedding or reranking provider receives memory text outside the machine and changes the store's vector backend. Use a separate store or deliberately re-embed before switching.
For removal, unregister the MCP server with openclaw mcp unset memkeeper.
Leave the local store and models in place unless the user explicitly requests
their removal.
Related skills
Local-first MCP memory backend and governance console for coding agents. Auto-organize, quarantine, supersede, and rollback shared memories across multiple a...
Privacy-first, self-hosted chat memory for OpenClaw — save and recall conversation history across sessions without any cloud dependency.
Scaffold, sanitize, or share an OpenClaw multi-agent memory system with a reusable workspace, memory-lancedb-pro configuration, role prompts, task-board conv...
Map of every agentmemory MCP tool, what each does, and its parameters. Use when choosing which memory tool to call, when a tool name or argument is unclear,...
Reuse, inspect, search, or append shared OpenClaw memory on the current machine. Use when a request depends on remembered local servers, deploy hosts, intern...