hmem-write

How to use write_memory

Call the MCP tool write_memory to save lessons, errors, decisions, or project insights to long-term memory.

If write_memory is not available:

1. Tell the user: "write_memory tool not found. Please reconnect the MCP server (in Claude Code: /mcp, in other tools: restart the tool)."
2. NEVER write directly to the .hmem SQLite file via shell commands. The database has WAL journaling, integrity checks, and tree-structure logic that raw SQL INSERT will bypass — causing corruption or data loss.

---

Syntax

write_memory(
  prefix: "E",
  content: "Short Title (~50 chars)\n\nL1 body — detailed explanation, can span multiple lines\nsecond body line with more context\n\tL2 node title\n\n\tL2 body text (supports newlines)\n\tmore L2 body\n\t\tL3 detail (2 tabs)\n\t\t\tL4 raw data (3 tabs — rarely needed)"
)

Title + Body convention (git-commit style): Every node has a title (short navigation label) and an optional body (detailed content loaded on drill-down). Separate them with a blank line — just like a git commit message.

• Title: The first line (L1) or first line at a given indent level (L2+). ~50 chars, like a chapter title.
• Body: Everything after the blank line at the same indent level. Freetext, no special prefix needed. Shown only when the node is drilled into, not in listings.
• Legacy > prefix: Still works for backward compatibility, but blank-line separation is preferred.
• Without body: The full text is stored as content and the title is auto-extracted from the first maxTitleChars characters.

L1 example with body:

Short Error Title

SQLite connection failed because .mcp.json used a relative path.
The fix was to use an absolute path in the HMEM_PATH env var.
	Details about reproduction

	Steps: 1. Set HMEM_PATH=./hmem  2. Run hmem serve  3. Observe SQLITE_CANTOPEN

L1 example without body (still works):

SQLite connection failed due to wrong path in .mcp.json
	Fix: use absolute path in env var