Agent tools reference

Agents read and write Lore through these tools. Parameter names are the same in every runtime; only the tool name prefix changes. Required parameters are marked required, and everything else is optional.

Memory toolsCopy Anchor Link

lore_guidanceCopy Anchor Link

Loads the full usage rules. Call it when your context does not already contain them.

Parameters: none.

{}

lore_statusCopy Anchor Link

Checks the memory backend. It returns Lore online when the database answers, or Lore offline with the error.

Parameters: none.

{}

lore_bootCopy Anchor Link

Loads the fixed boot memory for the runtime that calls it. Call it when boot memory did not load at session start.

Parameters: none.

{}

lore_get_nodeCopy Anchor Link

Opens one node. Call it before you rely on a memory from a <recall> block, and pass the session_id and query_id from that tag so that Lore records the use. A path without a scheme, such as webhooks, uses the default domain. Use core:// or project:// to browse a domain root.

{
  "uri": "project://checkout_api/webhooks",
  "session_id": "sess_8f3c",
  "query_id": "q_41"
}

lore_searchCopy Anchor Link

Searches memories by keyword, meaning, or both. It returns full content for the top results, so you do not need a separate lore_get_node call for them. An empty query, or * together with a domain, browses that domain's root. Without a domain, * is not a wildcard, so use a keyword or phrase.

{
  "query": "refund webhook",
  "domain": "project",
  "limit": 5
}

lore_list_domainsCopy Anchor Link

Lists the top-level domains, such as core, preferences, and project, with the number of root nodes in each.

Parameters: none.

{}

lore_create_nodeCopy Anchor Link

Creates a memory. Give either a full uri, or domain with parent_path and title. For a multi-segment path, the parent must already exist, with its own content, disclosure, and glossary. Lore can block the create when the priority budget or disclosure check fails. When priority 0 or 1 is full, the error gives the count and asks you to lower an existing node's priority first.

{
  "domain": "project",
  "parent_path": "checkout_api/webhooks",
  "title": "refund_idempotency",
  "content": "Refund webhooks are idempotent by event.id. Tests must replay the same event twice and assert a single refund row.",
  "priority": 1,
  "glossary": ["refund webhook", "idempotency key", "stripe"],
  "disclosure": "When changing refund or payment webhooks in checkout-api"
}

The example assumes that project://checkout_api/webhooks already exists, with its own content, disclosure, and glossary.

lore_update_nodeCopy Anchor Link

Changes fields on an existing node. Fields you omit stay unchanged. Use glossary_add and glossary_remove to change keywords without replacing the whole list.

{
  "uri": "project://checkout_api/webhooks/refund_idempotency",
  "glossary_add": ["event.id"]
}

lore_delete_nodeCopy Anchor Link

Removes a path that is obsolete, duplicated, or no longer wanted. Every change is recorded in the node's history, so you can roll the node back from its history page.

{
  "uri": "project://my_app/old_retry_policy"
}

lore_move_nodeCopy Anchor Link

Moves or renames a node. The new parent must already exist, because it becomes the node's new home.

{
  "old_uri": "project://my_app/refund_flow",
  "new_uri": "project://checkout_api/refund_flow"
}

The example assumes that project://checkout_api already exists, with its own content, disclosure, and glossary.

Tool names by runtimeCopy Anchor Link

In OpenCode, a name such as lore_lore_search means that you have an MCP entry instead of the native plugin. See Agent connection issues.

Pi, OpenClaw, and Hermes do not register lore_guidance. Their session-start context carries the same guidance text.

Skills tools (Loremem Cloud)Copy Anchor Link

Skills tools appear only when the connected server advertises Skills. Loremem Cloud does. The open-source server does not, so these tools do not appear there.

Notes:

• files is a list of objects. Each object has a path (required), either content (UTF-8 text) or content_base64 (binary), and an optional media_type. A new skill's files must include SKILL.md.
• Use lowercase letters, numbers, and hyphens in name. You cannot change it later.
• expected_version is the version you last read. If someone changed the skill first, the update fails. Read the current version with lore_skill_list, then try again.
• The agent doesn't load skills through these tools: each host syncs its own skills directory at session start. See Skills.
• Server-managed files in a synced directory are read-only. Any other file there is an artifact: Lore never checks or uploads it, and it survives updates.
• lore_skill_delete removes the skill's managed files from each agent at the next sync; artifacts stay.
• lore_skill_import never overwrites existing skills: a skill with the same name is reported as exists and left alone. Run it with dry_run first to list what would be imported.

Create a skill:

{
  "name": "release-checklist",
  "files": [
    { "path": "SKILL.md", "content": "# Release checklist\n\nRun the checks in order.\n" }
  ]
}

Update a skill's files. Use the skill_id that lore_skill_list returns:

{
  "skill_id": "your-skill-id",
  "expected_version": 3,
  "upsert_files": [
    { "path": "checklists/rollback.md", "content": "# Rollback\n\n1. Pause the deploy.\n" }
  ]
}

Import skills from a public GitHub repository. Ask the agent to run it with dry_run first, then without it:

{
  "source": "vercel-labs/agent-skills",
  "skills": ["frontend-design"],
  "dry_run": true
}