Deliverables Tools

Why this matters

Cached results are ideal for in-session chaining, but teams also need durable outputs they can revisit later. Deliverables give agents a clean public surface for preserving analyses, exports, charts, and packaged datasets without requiring callers to understand the internal run history.

Three tools

• ✓deliverables.list — Search or browse previously saved work products by query, type, and recency
• ✓deliverables.get — Retrieve the full payload and metadata for a specific deliverable
• ✓deliverables.register — Promote a cached result or packaged output into persistent storage for later reuse

From working memory to durable output

Use ephemerals and cache_ref values while an agent is actively working. When the output becomes something worth keeping, register it as a deliverable so it can be searched and reopened later without rerunning the original tool chain.

Example: search and promote

// Semantic search over past work products
{ "name": "data-grout@1/deliverables.list@1",
  "arguments": { "query": "revenue analysis from last quarter" }
}

// Register a result as a permanent deliverable
{ "name": "data-grout@1/deliverables.register@1",
  "arguments": { "title": "Q1 Revenue Summary", "type": "data", "cache_ref": "rc_abc123" }
}