What's Included (per data source)
Data source assessment: structure, quality, access patterns, sensitivity
Data preparation: cleaning, deduplication, formatting for AI consumption
Retrieval-augmented generation (RAG) setup on a standard stack: managed vector database, standard chunking and embedding configuration, semantic search enabled
Integration into the target agent
Acceptance validation: retrieval checked against a client-provided question set, with findings reviewed in a working session
What's Not Included
Remediation of materially inconsistent source data: scoped separately
Real-time sync: available as an add-on; standard configuration runs on scheduled refresh
The agent itself: see Agent Setup or Multi-Agent Orchestration
Retrieval tuning beyond the standard configuration: routes to Katalyst Custom™
Custom API development for a source without a supported access method: routes to Katalyst Custom™
Price: fixed for the defined scope, set during scoping.
Qualified Before It's Quoted
Before this service is quoted, the data source passes an intake check during the scoping conversation: structure, volume within the defined ceiling, and an access method on the supported list. A source that doesn't pass isn't quoted as a smaller version of this service; it's quoted as Katalyst Custom™, where the fuller architecture work belongs.
This service is required whenever a data source needs cleaning, deduplication, or retrieval setup before an agent can use it safely. The single integration in Agent Setup assumes the source already meets that bar.
Pattern We See
A retrieval system is only as scoped as the access controls on the data behind it. A common failure mode in early AI deployments: a knowledge base or shared drive that had years of informal, low-stakes access suddenly becomes searchable in plain English the moment an agent sits in front of it. The underlying permissions didn't change; what changed is how easy they are to reach.
FAQ
It depends how messy, and we find out before you're quoted rather than after. Every source passes an intake check during the scoping conversation. If it doesn't pass, we quote the work as Katalyst Custom™ instead of selling you a fixed-scope service that can't succeed.
Retrieval checked against a question set you provide, with findings reviewed together in a working session. You see the system answering your real questions from your real data before the engagement closes.
The service is priced per source, deliberately. One source, prepared properly, integrated into your agent. Several sources or production-scale retrieval across systems belongs in a Custom engagement, where the architecture gets designed rather than assembled.
A managed vector database with standard chunking and embedding configuration. No bespoke retrieval architecture, no infrastructure buildout. That constraint is what makes fixed pricing honest; tuning beyond it routes to Katalyst Custom™.
