admin_reindex_all

Backfill Cloudflare Vectorize embeddings for all existing contexts across every project in one call.

Overview

admin_reindex_all is a cross-project version of reindex_project. It iterates every context in D1 — regardless of project — and writes vector embeddings to Vectorize. Run it once after initial setup or after adding Vectorize to an existing deployment.

Added: v3.6.0

Layer: Infrastructure / Admin

Purpose: One-shot Vectorize backfill across all projects

---

Parameters

None. No arguments required.

---

Returns

A count of indexed contexts broken down by project:

Reindexed 47 contexts across all projects — semantic search now covers all historical snapshots.

By project:
  - wake-intelligence: 12
  - api-service: 23
  - mobile-app: 8
  - home-wake-test: 4

If the Vectorize index is not configured or no contexts exist:

No contexts reindexed. Either no contexts exist or the vector index is not configured.

---

When to Use

After Initial Deployment

The first time you connect Wake Intelligence to Vectorize, existing contexts have no embeddings. admin_reindex_all seeds the index in one call.

After a Vectorize Outage

If Cloudflare Vectorize was unavailable when contexts were saved, those contexts fell back to keyword-only search. Reindex to restore semantic coverage.

When Semantic Search Returns Nothing

If natural language queries consistently return no results, the Vectorize index is likely empty or sparse. Run this to restore full semantic search.

---

Example

admin_reindex_all()

Reindexed 47 contexts across all projects — semantic search now covers all historical snapshots.

By project:
  - my-project: 31
  - side-project: 16

---

How It Works

Internally, admin_reindex_all:

1. Fetches all contexts from D1 (up to 2,000)
2. For each context, generates a vector embedding from its AI-summarized text using @cf/baai/bge-base-en-v1.5
3. Upserts the vector into the wake-context-embeddings Vectorize index with the project as metadata
4. Returns a count per project

Contexts that already have embeddings are overwritten with fresh ones (upsert is idempotent).

---

Performance

• Processes up to 2,000 contexts per call
• ~500ms per 10 contexts (embedding generation via Workers AI dominates)
• For large deployments, prefer running during low-traffic periods

---

vs. reindex_project

	reindex_project	admin_reindex_all
Scope	Single project	All projects
Arguments	project (required)	None
Use case	Targeted backfill	Initial setup, full reset

---

See Also

• reindex_project — Single-project backfill
• search_context — Semantic search that benefits from reindexing
• get_memory_health — Check overall system health