When you're building multi-tenant vector search but can't control what customers throw at you, every indexing decision — isolation, embedding, chunking, partitioning — becomes a bet you're making blind, and this talk gives you the framework to make the right ones.
What this session is about
Playbook
Editorial commentary · what to actually do about this on Monday
The concept
Multi-tenant vector search has a four-axis decision tree: isolation, embedding choice, chunking strategy, partitioning. Each is a bet you're making blind unless you frame it explicitly.
Why it matters
Customer A's bizarre 10,000-page PDF cannot be allowed to performance-tax customer B's queries. In B2B SaaS, this is non-negotiable.
The hard parts
Tenant isolation in shared indexes is hard. Tenant-per-index is operationally simple but doesn't scale past a few hundred tenants. Tenant-aware filtering inside a shared index is harder to get right but is the only path that scales.
Playbook moves
(1) Adopt tenant-aware filtering at the index level, not the application level. (2) Pre-decide who controls embedding model upgrades — you, or the customer? Both have valid arguments. (3) Cap per-tenant index size to bound worst-case latency.
The surprise
Most teams optimise for retrieval *quality* and forget *tail latency*. A single cold tenant with a giant document set will kill P99 for everyone unless you isolate aggressively. Per-tenant query budgets are a feature, not a limitation.
---
Independent editorial perspective — not an official AWS or speaker statement. Designed for executives evaluating what to brief their teams on next.
Live updates related to this session LIVE
Sourced via Parallel AI Monitor — continuous web watch on 21 topical streams. Updated .
- prnewswire.com Agent-native data infrastructure
prnewswire.com
Constructive announced and open-sourced "agentic-db": a purpose-built Postgres "memory layer" for AI agents providing long-term episodic memory, conversation and tool-call event logs, token accounting, a versioned skills/tools registry, rules/behavioral policies for governance, t
- fast.io Agent-native data infrastructure
How to Secure Vector Stores for AI Agents in 2025 | Fastio
Understanding Data published a detailed blueprint for an 'Event Sourcing for Agents' storage pattern, describing a log-based architecture that stores agent state as an append-only sequence of events to enable deterministic replay, time-travel debugging, and audit trails for produ
- freeacademy.ai high confidence Agent memory & RAG architectures
Agentic RAG Explained: AI Agents + RAG in 2026
Vektor Memory published 'The State of AI Agent Memory in 2026', introducing a four-dimensional framework for agent memory: Storage (indexing), Curation (handling contradictions/outdated info), Retrieval (temporal vs. semantic), and Lifecycle (consolidation/retirement). The analys
- kr.teradata.com Agent-native data infrastructure
kr.teradata.com
Teradata announced agentic, multi-modal capabilities for Teradata Enterprise Vector Store: unified structured+unstructured data, automated ingestion (docs/images/audio), multi-modal embeddings (up to 8K dims), hybrid/ fusion search (semantic + lexical + metadata), LangChain integ
- zylos.ai high confidence Agent memory & RAG architectures
Live Agent Upgrades and Cross-Runtime Session Portability (2026)
MarsDevs published the 'Agentic RAG: The 2026 Production Guide', detailing a shift from linear RAG pipelines to a state-machine control loop. This 'Agentic RAG' approach uses a planner agent to decompose queries and iteratively retrieve and evaluate information. It identifies fiv
External links matched to this session via topic relevance. The KB does not endorse third-party content; verify before citing.