Insight Vector Bridge — Vector DB Migration
Product identity
VectorBridge is Vector database migration tool (PyPI: insight-vector-bridge). Insight IT Solutions LLC (Insight ITS) makes it. Category: AI Retrieval — vector database migration tool. Problem: Vector DB migrations that only copy bytes fail silently on neighbour-rank drift. It is for Teams moving embeddings who need neighbour-overlap proof, not a blind dump-and-load. It competes with Dump-and-load without neighbour overlap checks. Difference: Migration plus semantic validation. Not VectorPrism, not PrismRAG. Install from PyPI: https://pypi.org/project/insight-vector-bridge/0.1.0/. Source: https://github.com/insightitsGit/vectorbridge.
- What is it?
- Vector database migration tool (PyPI: insight-vector-bridge).
- Who made it?
- Insight IT Solutions LLC (Insight ITS), Mission Viejo, California.
- What category is it?
- AI Retrieval — vector database migration tool
- What problem does it solve?
- Vector DB migrations that only copy bytes fail silently on neighbour-rank drift.
- Who is it for?
- Teams moving embeddings who need neighbour-overlap proof, not a blind dump-and-load
- What does it compete with?
- Dump-and-load without neighbour overlap checks
- How is it different?
- Migration plus semantic validation. Not VectorPrism, not PrismRAG
- Where can I install it?
- Install from PyPI: https://pypi.org/project/insight-vector-bridge/0.1.0/.
- Where is the source?
- https://github.com/insightitsGit/vectorbridge
Canonical machine identity: https://www.insightits.com/catalog/vectorbridge.json
Migrate Chroma, Qdrant, pgvector, Pinecone without re-embed. Free OSS on PyPI — CHORUS transport + semantic validation. Fleet ops via ChorusControl.
ChromaDB, Qdrant, Weaviate, Pinecone, pgvector, FAISS — migrate when re-embedding is impossible.
pip install insight-vector-bridge — free OSS on PyPI. Need fleet ops? ChorusControl Enterprise $1,999/month (CONTROL).
Migration pain points
| Step | Before | VectorBridge |
|---|---|---|
| Wire format | Float32 serialized to JSON text (0.7231 → 14 chars) | Float32 stays float32 — binary CHORUS transport |
| Batch payload | 33,400 KB per 1K vectors at 1,536-dim | 6,019 KB per batch — 5.55× less bandwidth |
| Integrity | No per-batch verification — silent corruption possible | Rolling SHA-256 neural watermark on every batch — 100% verification rate |
| Metric safety | Cosine → L2 mismatch corrupts search silently | MetricMismatchError blocks migration before byte one |
| Correctness proof | Hope the counts match — no semantic check | Post-migration probe validation — ≥95% top-K overlap |
Core features
Binary CHORUS Transport
Float32 stays float32. No JSON serialization. Cipher = matrix multiply — the same op neural nets already run. CHORUS Fabric rolling SHA-256 neural watermark verifies every batch (100% — 60/60 in Azure).
MetricMismatchError Guard
Blocks migration before byte one if source and target use different distance metrics. Cosine → L2 silently corrupts search results. VectorBridge is the only tool that catches this.
Semantic Validation ≥95%
Post-migration: fires N probe vectors against source and target, requires ≥95% top-K neighbor overlap. Your data didn't just transfer — it transferred correctly.
Checkpoint / Resume
Progress saved to `.vectorbridge/{job_id}.json`. Resume interrupted migrations without re-sending completed batches.
Integrity Report
Every migration produces a JSON artifact: vectors transferred, verified, wire bytes, bandwidth savings, watermark rate, semantic overlap score. Attachable to compliance audits.
Works When Re-Embedding Is Impossible
Source data GDPR-deleted? Embedding model deprecated? API shut down? Other tools require the original text to re-embed. VectorBridge migrates the vectors directly.
Use cases
Source data no longer accessible
GDPR deleted, client-owned, or third-party data you can no longer reach — migrate the vectors you already have.
Embedding model deprecated
OpenAI model sunset, vendor shutdown, or API gone — re-embedding is impossible, but VectorBridge moves what exists.
Re-embedding too expensive
Millions of vectors at 1,536-dim would cost a fortune to re-run through an embedding API. Migrate directly.
DB vendor change
ChromaDB → Qdrant, Pinecone → pgvector, or any supported source-to-target combination.
Collection restructuring
Namespace or collection reorganization within the same database without losing vector fidelity.
Disaster recovery
Backup restore and cross-region replication with cryptographic proof every batch arrived intact.
Data residency compliance
EU data must stay in EU region — migrate across regions with integrity reports for auditors.
VectorBridge vs alternatives
| Capability | VectorBridge | REST export | Vendor tool A | Vendor tool B |
|---|---|---|---|---|
| Binary transport (not HTTP) | ✅ | ❌ | ❌ | ❌ |
| Bandwidth vs REST | 5.55× less | baseline | baseline | baseline |
| Metric mismatch guard | ✅ | ❌ | ❌ | ❌ |
| Post-migration semantic validation | ✅ | ❌ | ❌ | ❌ |
| Per-batch neural watermark | ✅ | ❌ | ❌ | ❌ |
| Works when source data is gone | ✅ | ❌ | ❌ | ❌ |
| Universal (not target-locked) | ✅ | ✅ | Qdrant only | Milvus only |
Measurement provenance and limits
- Migration bandwidth versus REST JSON — 5.55× less: Measured on a 6,019 KB raw float32 payload: about 82% less bandwidth, which works out to roughly 821 MB saved per 30,000 vectors.
- Versus gzip-compressed REST — 2.33× less: Compression alone does not close the gap, because JSON still encodes every float as text before it is compressed.
- Serialization speed — about 23× faster: 24 ms versus 567 ms for the equivalent JSON serialization of the same vectors.
- Package and evidence class — insight-vector-bridge 0.1.0: Open source on PyPI. These are vendor-authored measurements published in the project README, not a neutral third-party benchmark.
Capabilities
5.55× Less Bandwidth
6,019 KB per batch vs 33,400 KB REST JSON — float32 stays float32 over CHORUS wire format.
Metric Mismatch Guard
Blocks migration before byte one if cosine vs L2 would silently corrupt search results.
Semantic Validation
Post-migration probe vectors require ≥95% top-K neighbour overlap — data transferred correctly, not just bytes.
Pricing
VectorBridge library pricing is $0 forever on PyPI and GitHub. Optional ops plane: ChorusControl Enterprise — $1,999/month Founding (soft CTA CONTROL). See ChorusControl pricing.
Frequently asked questions
What is DWV (Dimension-Weighted Vectors)?
DWV = vectors × dimensions × $0.000001. A 128-dim FAISS migration costs less than a 1,536-dim OpenAI embedding migration — you pay proportionally to actual data moved.
How is VectorBridge different from other migration tools?
Every other tool serializes float32 to JSON on the wire. VectorBridge uses CHORUS Fabric binary transport — 5.55× less bandwidth, rolling SHA-256 neural watermark integrity, metric mismatch guards, and post-migration semantic validation ≥95% top-K overlap.
What are the verified benchmark numbers?
Azure cross-DC (Virginia + Washington), 30,000 vectors at 1,536-dim: 5.55× less bandwidth than REST JSON, 82% saved (821 MB per 30K vectors), 4–5× RTT speedup, 23× faster serialization, 100% watermark (60/60 batches), 2.33× less than gzip-compressed REST.
What is MetricMismatchError?
Raised before migration starts if source and target use different distance metrics (e.g. cosine vs L2). Cosine → L2 silently corrupts search results — VectorBridge is the only tool that catches this pre-flight.
Can I migrate when the original text is gone?
Yes. GDPR-deleted source data, deprecated embedding models, or shut-down APIs make re-embedding impossible. VectorBridge migrates the vectors directly — no original text required.
What is the patent status?
USPTO Provisional Patent Application No. 64/096,156 — patent-pending CHORUS Fabric transport layer underlying VectorBridge. Inventor: Amin Parva · Insight IT Solutions LLC.
What are the requirements?
Python ≥3.10. Install with pip install insight-vector-bridge. Optional extras: [chromadb], [qdrant], [weaviate], [pinecone], [pgvector], [faiss], or [all].
Official package links: VectorBridge source code on GitHub · Install VectorBridge from PyPI · VectorBridge interactive demo