Onyx Search/RAG Prototype

Onyx is an enterprise AI interface that combines chat with internal search against connected knowledge sources and optional agent capabilities. Onyx’s documentation describes it as a natural‑language interface that integrates with your knowledge and applications, with core features including internal search, agents, and actions.

Onyx relies on a Large Language Model of your choice, either cloud-hosted or locally-hosted. When used with a locall-hosted LLM, Onyx talks to no external source, and can even operate in air-gapped environment. 

Who it’s for

• Teams struggling with long learning curve because knowledge is scattered across tools

• Engineering/product teams drowning in design docs, tickets, PRs, Slack, Confluence, SharePoint

• Internal support/service desks

• Security-conscious organisations that want internal AI but don’t want “everyone pasting into ChatGPT”

Possible indications you need it

• People can’t find “the latest” policy/process/decision and keep using stale PDFs

• You want internal AI search but your security team says “not in the cloud”

• You’re paying for an AI tool and leadership asks “where’s the value?”

• You’ve had a near-miss (wrong summary, wrong email, wrong decision support)

What we build

A working Onyx deployment connected to your real tools so people can:

• search + chat against internal sources with citations

• use role-appropriate prompts (not prompt theory)

• measure whether answer quality and time-saved are real

Deliverables (by end of pilot)

• Onyx deployed (Docker or Kubernetes) in your environment

• 2–4 sources connected (your choice; we keep scope tight)

• Search + chat tuned for your use cases (retrieval settings + prompt templates)

• Evaluation harness: test set, success criteria, baseline vs improved results

• Runbook + handover: how to operate it, what to harden next, what to monitor

• Go/No-Go recommendation: whether to scale, and what will block scaling

Timeline (typical, 10–15 business days)

Days 1–2: scope, sources, success metrics, access constraints, stakeholder interviews

Days 3–7: deploy + connect sources; iterate on retrieval and answer quality

Days 8–12: evaluation; hardening notes; safe-use + governance guidance

Days 13–15: handover session; runbook walkthrough; next steps

What I need from you (prework)

Ideally, to complete Onyx Implementation Planning 1-day workshop that would clarify all the requirements. Alternatively:

• 30–45 minutes with a sponsor + an owner for coordination

• Read access to target sources (and permission to index a limited subset)

• Your constraints: cloud allowed? self-host required? air-gapped? approved LLM endpoint?

• 10–20 real questions/tasks used to test success (not “generic AI demo” questions)

Notes on editions and security

Onyx Community Edition is enough for a first proof-of-value. Once completed and you decide to proceed to implementation, you will need enterprise features like Role-Based Access Control. That can be available from Onyx Enterprise Edition, or from some of the other enterprise-grade platforms. It would require more effort and more funds - however by that time you would have confidence in what you need and how you want it achieved.

What this service is NOT

• Not a knowledge management transformation

• Not “rewrite all your documentation”

• Not a multi-month platform implementation

• Not a vendor pitch deck or a generic demo