Your first memory.

Record a memory

Use MCP tools inside your coding agent, or the CLI directly. Both write to the same memory graph.

# MCP (in your coding agent)
record_graph_memory({
  category: "preference",
  summary: "Prefer terse, sentence-by-sentence delivery."
})

# CLI equivalent
empirical memory record --category preference --summary "Prefer terse, sentence-by-sentence delivery."

Query memory

Run a semantic search and confirm your new memory surfaces in the top results.

# MCP
query_memories({ query: "writing style preference", topK: 3 })

# CLI
empirical memory query --match "writing style preference" --top-k 3 --pretty

Validate in workflow

Once the memory surfaces in query results, use it during a real task. If recall returns nothing, run empirical doctor to confirm API connectivity and token health.

Keep it durable: Do not record secrets, transient conversation details, or facts that will be wrong after the current task.