Memory Tools

Memory tools let agents store facts that persist across conversations.

remember_user_fact

Stores a fact about the user. User facts are shared across all agents, so any agent can access information stored this way.

remember_user_fact(
  fact: "Prefers dark mode and monospace fonts"
)

Parameters

Parameter	Type	Required	Description
fact	string	yes	A short, atomic fact about the user
overrides	boolean	no	Set to true if this fact contradicts or supersedes a previously stored fact (default: false)

remember_agent_fact

Stores a fact private to this agent. Other agents cannot see these facts.

remember_agent_fact(
  fact: "Last research topic was renewable energy markets"
)

Parameters

Parameter	Type	Required	Description
fact	string	yes	A short, atomic fact for this agent to remember
overrides	boolean	no	Set to true if this fact contradicts or supersedes a previously stored fact (default: false)

update_identity

Updates the agent’s own identity attributes. These are key-value pairs that persist and affect how the agent presents itself.

update_identity(
  attributes: {
    "name": "Research Bot",
    "style": "thorough and detail-oriented"
  }
)

Parameters

Parameter	Type	Required	Description
attributes	object	yes	Key-value pairs of identity attributes to set. Use an empty string value to remove an attribute

Best practices

• Check before storing: read existing memory first to avoid duplicates
• Be specific: store concise, actionable facts rather than vague impressions
• Update, don’t duplicate: use overrides: true when information changes
• Keep it lean: for large data, save to a workspace file and reference it in memory
• Scope correctly: user facts for personal preferences, agent facts for task-specific context

Automatic compaction

Stored facts are periodically compacted by the platform’s insight compaction process. This merges duplicates, removes stale information, and keeps the memory set manageable. See Insights for details.