Skip to content

Quick Start

That is the whole loop. Write memory, retrieve memory, inspect memory. The background pipeline handles the rest.

# 1. Initialize
/memory-onboard
# 2. Teach it something
graph_memory(
action="remember",
path="preferences/deployment",
gist="Always use blue-green deploys for production services",
content="Blue-green for prod. Canary for staging. Never direct push.",
tags=["preferences", "deployment"],
confidence=0.9
)
# 3. Recall it next session, or next week
/recall deployment strategy
# 4. Check what your agent knows
/memory-status
# 5. See the full history
graph_memory(action="history")

Teach your agent a durable fact, decision, or pattern with the remember action:

graph_memory(
action="remember",
path="decisions/auth-strategy",
gist="Use Clerk for auth across web and mobile",
content="Clerk chosen over Auth0 for its SDK quality and org support...",
tags=["decisions", "auth"],
confidence=0.9,
edges=[{ target: "patterns/auth", type: "supports" }]
)

Retrieve relevant memory with edge traversal:

graph_memory(action="recall", query="deployment strategy", depth=1)

You can also run /recall <query> as a slash command in Claude Code.