Employee AI
ChatGPT energy.
Your documents.
On your models.
Give people a chat box they’ll use, grounded in policy manuals, policy manuals, off Reddit vibes from a US data centre from a US data centre.
Banning ChatGPT without a replacement is how you invent shadow IT. Corvux ships a private ChatGPT: same muscle memory, different physics, retrieval over your corpus, models on Kenya GPUs, logs your security team can open.
$ corvux chat ask --org insurer-ke \ "What's our waiting period on outpatient claims?" SOURCE Product_Wording_v12.pdf · p.14 ANSWER Outpatient waiting period is 30 days… HOST kenya-gpu-example · Nairobi ✓ cited · logged · no foreign API
Inside the box
UI people recognise. Plumbing auditors like.
Chat UI
Streaming answers, history controls, department spaces.
RAG
Chunk, embed, retrieve, cite, embeddings stay local.
SSO
Hook into your identity provider; kill anonymous use.
Admin
Publish docs, set retention, watch usage by team.
Host on your GPUs or managed Nairobi capacity. Same product. Different CapEx story.
Best first departments: compliance, credit ops, claims, HR. Worst first department: “everyone in the company with the entire file server.” Start narrow, win trust, expand.
Rollout
Four weeks, one department, hard numbers.
FAQ
From IT and the business.
Will staff notice it’s “worse” than GPT-4?
For internal policy and process questions with RAG, private open weights often win because answers cite your sources. For random internet trivia, keep public tools on non-sensitive machines.
SharePoint / file shares?
Yes, we connect approved corpora and respect folder permissions where your IAM allows.
How do we stop the old habit?
Launch chat + optional AI gateway to log/block foreign LLM domains after adoption.
Mobile?
Hardened web access to your Kenya endpoint, same residency story.
Pilot private chat
One department. Clean corpus. Adoption dashboard. That’s the deal.
Start a private chat pilot