Memory CRUD & History
Memron exposes four consolidated verbs so agents can work with memory without loading dozens of overlapping tool names. Administrative and analytical workflows stay in the dashboard.
Tools Catalog
| Verb | Primary Parameters | Purpose | Lifecycle |
|---|---|---|---|
memory_store | content, type?, source?, metadata?, tags? | Encrypts, embeds, and stores content. | New items default to Inbox / untriaged. |
memory_recall | query, mode?, category?, limit?, tokenBudget? | Hybrid retrieval with pinned rules and token packing. | Read-only; returns decrypted context. |
memory_manage | action, pointerId, status?, feedback? | Updates, triages, pins, archives, or records feedback. | Human dashboard can perform the same triage actions. |
memory_validate | action, proposedPlan, context? | Checks plans against pinned constraints and known risks. | Run before consequential agent actions. |
Example: Invoking memory_store
Here is an example JSON-RPC payload sent by an MCP agent to store a memory:
{
"jsonrpc": "2.0",
"method": "tools/call",
"params": {
"name": "memory_store",
"arguments": {
"title": "Database Schema Convention",
"content": "All tables must include id SERIAL PRIMARY KEY, created_at TIMESTAMPTZ DEFAULT NOW(), and updated_at triggers.",
"tags": ["database", "postgresql", "standards"],
"bucket": "engineering"
}
},
"id": 101
}