CheriMind

Projects - Knowledge spaces across all clients

Project	Indexed chunks	Sources	Users	LLM

Platform scorecard - Adoption, cost, reliability & security across all tenants

Tenant	Queries	Active users	Grounded	Gaps	p95	CSAT	LLM cost	Open fails	Key mode

Cost intelligence - Forecast, top spenders & model mix across all tenants

Encrypt a value - Turn an API key or secret into its stored, encrypted form

Paste a plaintext value (for example an AI provider key). It is encrypted with this server's key so you can store it safely in config or the database. Nothing is saved here, and there is no decrypt.

Demo cloud model - The fast option on the demo login (Azure OpenAI)

Set an Azure OpenAI model for the demo's "Fast cloud" choice. The key is encrypted at rest and used only for the demo, capped per visitor. Leave the key blank when updating to keep the stored one. This does not affect any real project.

RAG evaluation - Measure retrieval quality, do not eyeball it

Runs test questions through the real search pipeline and scores it, so every change is judged by numbers. Leave Cases blank to use the built-in set, or paste your own. Off-topic cases (shouldFind:false) must return nothing.

Embedding index migration - Bless a known-good legacy index

If queries are blocked because chunks were indexed before embedder fingerprinting, this stamps the current fingerprint onto unstamped chunks - but only when their stored vector dimension matches the running embedder. Chunks with a different dimension are a real model change and are reported as skipped (re-index those). Does not re-embed.

Users

Username	Project	Role	Actions

Support tickets - Demo/client requests; reply and the user sees it in their Support page

Pipeline failures - Retry from scratch (query) or resume from the staged file (upload)

Time (UTC)	Project	Op	Failed stage	Error	Actions

Activity - Recent queries across all clients

Time (UTC)	Project	Question	Persona	Tokens

On-premises installation - One installer to share with a client

A client can host the whole platform (app, database, and AI service) on their own machine, so nothing leaves their network. You share one file - CheriMind-Installer.zip - which contains everything pre-built. No source code is shared. The client installs Docker once, unzips it, and double-clicks a start file.

Download CheriMind-Installer.zip Downloads the ready-to-share installer (available once it has been built and placed on the server).

Step 1 - Create the installer (you do this once)

On a machine that has Docker and the CheriMind source, run the build script - it builds the images and produces the single installer file:

• Windows (PowerShell): deploy\onprem\build-installer.ps1
• Mac / Linux: deploy/onprem/build-installer.sh

It creates CheriMind-Installer.zip (a few hundred MB - it bundles the app, the pgvector database, and the AI service as ready-to-run images). That one file is the only thing you send the client.

To enable the Download button above, put that zip on this server at <app-folder>/installer/CheriMind-Installer.zip (default /opt/cheri-rag/installer/), or set CHERIRAG_INSTALLER_PATH to its location. Then any admin can download it from here.

Step 2 - The client installs it

1. Install Docker Desktop (Windows/Mac) or Docker Engine (Linux) - one time.
2. Unzip CheriMind-Installer.zip.
3. Double-click start.bat (Windows), start.command (Mac), or run ./start.sh (Linux). First run loads the images and generates secrets (2-5 min).
4. Open http://localhost:5099, sign in as admin / admin123 (forced to change), connect an AI model, and index documents.

Do not share your own .env - it holds this server's secrets; the client's start script generates their own. The full per-OS guide (Docker install, first-time setup, backups, troubleshooting) ships inside the zip as README.md.