This page covers what it takes to run.
What you run
eavantage ships as a Docker Compose stack:
- a graph database (Neo4j) — the authoritative model
- a document store (MongoDB) — engagements, candidates, the gap register, recommendations, standards, and the audit and chat logs
- an embedded vector store — for semantic retrieval over standards and engagement context
- the API service and the web interface
- a one-time bootstrap that initialises the databases on first run
It runs on your compute. Specific sizing and prerequisites come with the deployment guide.
What you provide
- The infrastructure — compute, storage, and network — and its running cost. Because there is no shared hosting, the vendor carries none of this for a licensed customer.
- Encryption at rest, if you need it. eavantage’s databases don’t encrypt at rest natively, so this is a host-disk or volume decision you make at deploy time. See Security.
- A language-model provider connection — see below.
The language-model provider
eavantage’s AI features call a language-model provider — Anthropic, OpenAI, or Azure OpenAI — through an API key. Whether that key is yours or one provided under your licence is part of thelicensing conversation.
If no key is configured, eavantage runs in a manual mode: every AI feature falls back to human input. The product still works end to end — you lose the acceleration, not the workflow.
A fresh instance is empty
A new deployment starts with nothing in it — no pre-loaded standards, no clients, no users beyond a single administrator account. Your standards library builds up as you run engagements, and anything worth keeping is promoted into it at closure. See How it works.
Updates
While eavantage has a small number of customers, updates are hands-on: we prepare each release and work through the upgrade with you. Nothing updates on its own, and no client or engagement data ever leaves your environment.
An update is a short maintenance window. Each release is a version-tagged bundle that upgrades in one packaged step: an encrypted backup is taken first, the database migrations are applied forward-only, the stack restarts on the new version, and a health check runs before it’s considered done. If any step fails, the previous version is restored rather than leaving a half-upgraded system. Delivering releases through a self-service channel is planned as the customer base grows.
What isn’t supported yet
Docker Compose is the one verified deployment target today. Kubernetes and air-gapped installation have been evaluated but are not buildable yet — if either is a hard requirement for you, raise it early.
The licence check
A licensed deployment checks in with our licensing service about once a day to confirm the licence is current and has not been revoked. The check-in sends the licence key and nothing about your work — never client, engagement, or model data. An evaluation trial box does not check in at all.
If a check-in fails — a network outage, a firewall change, or a lapsed licence — the deployment keeps running at full function for a grace period of seven days. After that, a lapsed deployment moves to a reduced state: you can still open and read existing engagements, run every export, sign in, and back up or restore. What pauses is new work — new engagements and new AI extraction — until the licence is restored.
A licence problem never deletes data, never locks you out of what you have already built, and never shuts the deployment down — your data stays in your environment either way. Stricter enforcement is available as a per-contract option but is never the default. Any notice period and wind-down terms are set in the licence agreement, not here.
Security covers data handling and what’s protected. Licensing covers terms and support.