mempalace

MemPalace — AI Memory System

Persistent, structured, local-only memory across sessions. 19 MCP tools. Source: https://github.com/milla-jovovich/mempalace

Relationship to OpenClaw's built-in memory: OpenClaw writes raw daily logs to memory/YYYY-MM-DD.md. MemPalace adds a structured, searchable layer on top — semantic search across all sessions, a temporal knowledge graph for network facts, and per-agent diaries. Use daily logs for "what happened today" and MemPalace for "what did we decide and why."

How to Call the Tools

All tools use mcp-call with the mempalace MCP server:

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" <tool-name> '<arguments-json>'

Palace Read Tools (7)

mempalace_status — Palace Overview

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_status '{}'

Returns total drawers, wing/room counts, palace path, memory protocol instructions, and AAAK dialect spec. Call this at session start to load palace context.

mempalace_get_aaak_spec — AAAK Dialect Spec

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_get_aaak_spec '{}'

Get the AAAK dialect specification — the compressed memory format MemPalace uses. Call this if you need to read or write AAAK-compressed memories.

mempalace_search — Semantic Search

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_search '{"query":"why did we configure OSPF area 10 as stub","limit":5}'

Parameters:

• query (required): Natural language search
• limit (optional, default 5): Max results
• wing (optional): Restrict to wing
• room (optional): Restrict to room

Semantic search. Returns verbatim drawer content with similarity scores.

mempalace_list_wings — List All Wings

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_list_wings '{}'

List all wings with drawer counts.### mempalace_list_rooms — List Rooms

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_list_rooms '{"wing":"wing_netclaw"}'

List rooms within a wing (or all rooms if no wing given).

Parameters: wing (optional) — filter by wing.

mempalace_get_taxonomy — Full Taxonomy

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_get_taxonomy '{}'

Full taxonomy: wing → room → drawer count.

mempalace_check_duplicate — Check Before Filing

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_check_duplicate '{"content":"R1 upgraded to IOS-XE 17.12.1"}'

Check if content already exists in the palace before filing.

Parameters: content (required), threshold (optional, default 0.9).

Palace Write Tools (2)

mempalace_add_drawer — Store Memory

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_add_drawer '{"wing":"wing_netclaw","room":"architecture-decisions","content":"Campus OSPF design: Area 0 backbone on core switches, Area 10 Building A, Area 20 Building B. Stub areas on access layer."}'

File verbatim content into the palace. Checks for duplicates first.

Parameters:

• wing (required): Wing name (project/person)
• room (required): Room name (topic/aspect)
• content (required): Verbatim content to store — exact words, never summarized
• source_file (optional): Where this came from
• added_by (optional, default "mcp")

mempalace_delete_drawer — Delete Memory

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_delete_drawer '{"drawer_id":"drawer_wing_netclaw_decisions_abc123"}'

Delete a drawer by ID. Irreversible.

Knowledge Graph Tools (5)

mempalace_kg_add — Add Temporal Fact

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_kg_add '{"subject":"R1","predicate":"runs_version","object":"IOS-XE 17.12.1","valid_from":"2026-03-15"}'

Add a fact to the knowledge graph. Subject → predicate → object with optional time window. E.g. ('Max', 'started_school', 'Year 7', valid_from='2026-09-01').

mempalace_kg_query — Query Entity

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_kg_query '{"entity":"R1"}'

Query the knowledge graph for an entity's relationships. Returns typed facts with temporal validity. Filter by date with as_of to see what was true at a point in time.

Parameters: entity (required), as_of (optional, YYYY-MM-DD), direction (optional: outgoing/incoming/both).

mempalace_kg_invalidate — Mark Fact Expired

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_kg_invalidate '{"subject":"R1","predicate":"runs_version","object":"IOS-XE 17.9.4a","ended":"2026-03-15"}'

Mark a fact as no longer true. E.g. ankle injury resolved, job ended, moved house.

mempalace_kg_timeline — Entity Timeline

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_kg_timeline '{"entity":"R1"}'

Chronological timeline of facts. Shows the story of an entity (or everything) in order.

mempalace_kg_stats — KG Overview

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_kg_stats '{}'

Knowledge graph overview: entities, triples, current vs expired facts, relationship types.

Navigation Tools (3)

mempalace_traverse — Walk the Palace Graph

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_traverse '{"start_room":"ospf-design","max_hops":2}'

Walk the palace graph from a room. Shows connected ideas across wings — the tunnels. Like following a thread through the palace: start at 'chromadb-setup' in wing_code, discover it connects to wing_myproject (planning).

mempalace_find_tunnels — Cross-Wing Connections

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_find_tunnels '{"wing_a":"wing_campus","wing_b":"wing_datacenter"}'

Find rooms that bridge two wings — the hallways connecting different domains. E.g. what topics connect wing_code to wing_team?

mempalace_graph_stats — Graph Overview

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_graph_stats '{}'

Palace graph overview: total rooms, tunnel connections, edges between wings.

Agent Diary Tools (2)

mempalace_diary_write — Write Diary Entry

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_diary_write '{"agent_name":"netclaw","entry":"SESSION:2026-04-08|health.check.R1.R2|CPU.normal|OSPF.adjacency.flap.detected.R1-R3|★★★","topic":"operations"}'

Write to your personal agent diary in AAAK format. Your observations, thoughts, what you worked on, what matters. Each agent has their own diary with full history. Write in AAAK for compression — e.g. 'SESSION:2026-04-04|built.palace.graph+diary.tools|ALC.req:agent.diaries.in.aaak|★★★'. Use entity codes from the AAAK spec.

mempalace_diary_read — Read Diary History

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_diary_read '{"agent_name":"netclaw","last_n":10}'

Read your recent diary entries (in AAAK). See what past versions of yourself recorded — your journal across sessions.

When to Use

• Session start: Call mempalace_status to load palace context
• Before decisions: Search past sessions for prior decisions on the same topic
• After changes: Store architecture decisions, config rationale in palace drawers
• Tracking upgrades: Use knowledge graph to record temporal network facts
• Cross-session troubleshooting: Search past sessions for similar symptoms
• Session end: Write diary entry summarizing observations and lessons

Network Operations Workflow

Step 1: Load Palace Context

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_status '{}'

Step 2: Search for Prior Context

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_search '{"query":"BGP migration campus core"}'

Step 3: Record Network Fact

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_kg_add '{"subject":"R1","predicate":"bgp_peer","object":"R2","valid_from":"2026-04-08"}'

Step 4: Store Decision

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_add_drawer '{"wing":"wing_netclaw","room":"routing-decisions","content":"Migrated campus core from OSPF to eBGP. Reason: multi-vendor support (Arista + Cisco). AS 65001 for core, AS 65010-65020 for distribution."}'

Step 5: Write Diary Entry

python3 $MCP_CALL "python3 -u $MEMPALACE_MCP_SCRIPT" mempalace_diary_write '{"agent_name":"netclaw","entry":"SESSION:2026-04-08|bgp.migration.campus.core|OSPF→eBGP|AS65001.core|multi.vendor.rationale|★★★★"}'

GAIT Audit Trail

Record memory operations in GAIT:

python3 $MCP_CALL "python3 -u $GAIT_MCP_SCRIPT" gait_record_turn '{"artifacts":[],"user_text":"Store BGP migration decision in MemPalace","assistant_text":"Added drawer to wing_netclaw/routing-decisions: Campus core OSPF→eBGP migration rationale. Added KG triple: R1 bgp_peer R2 (valid_from 2026-04-08). Diary entry recorded."}'

Failure Behavior

• If a tool call fails with an authentication or connection error, check that GAIT_MCP_SCRIPT, MCP_CALL, MEMPALACE_MCP_SCRIPT are set and valid before assuming a data or device problem.
• On a tool error (timeout, unreachable host, malformed response), report the failure and its error message directly to the user rather than fabricating or guessing at results.
• For a confirmed read-only call, check connectivity and retry once if appropriate. For any call that changes state or sends a message, a timeout does not prove the action failed: inspect current state or delivery status before retrying, preserve the required approval/change gates, and do not repeat an action whose outcome is unknown.

Audit examples are illustrative. Replace their request and outcomes with observed session evidence; inspect MCP isError, returned ok and recorded GAIT text.