Why we built Nexora self-hosted

Most AI-agent platforms ask you to do something that, on reflection, is strange: hand a third party a standing connection to your most sensitive systems. To be useful, an agent needs access to your code, your ticketing system, your cloud accounts, your databases, your internal chat. The pitch is convenience. The cost is that a vendor now sits in the middle of everything an agent touches.

We did not want to build that, and we did not want to use it. So the Nexora agent product — the part that actually reads your data and executes against your systems — runs entirely on your own infrastructure. That is not a deployment option we offer alongside a hosted version. It is the only way it runs.

The agent product is never a third-party SaaS

Parendum does host some supporting services: the marketing site, the documentation, the licensing gateway, and the package marketplace. None of those ever see your agents, your prompts, your knowledge bases, or your production systems. The licensing gateway verifies a signed license and collects coarse usage counts over a heartbeat — that is the entire surface.

The core platform is open source under the MIT license. You can read every line that touches your data. The paid product, Nexora Cloud, builds on that same open core and is license-gated, but it still runs on your hardware. License validation is offline: the instance verifies a signed JWT against a public key it already holds. No call home is required to start working, and an air-gapped deployment is a supported configuration, not a workaround.

Data and production access stay where they belong

When an agent in Nexora opens a pull request, queries a database, posts to Slack, or applies a Kubernetes manifest, every one of those calls originates from inside your network, using credentials you provisioned, against endpoints you control. The bytes do not transit our servers because our servers are not in the path.

• Your LLM keys are yours. Nexora speaks to roughly 25 providers — including local Ollama and any OpenAI-compatible endpoint — so you decide which models see your prompts, and you can keep them fully on-premise if you want to.
• Secrets are encrypted at rest with a key you own. Provider keys and integration tokens are stored with Fernet encryption tied to your ENCRYPTION_KEY, never ours.
• Knowledge bases and semantic memory live in your Postgres with pgvector. The embeddings of your internal documents sit in your database, not a vendor index.

Why this matters for real teams

For regulated industries, "the data never leaves our infrastructure" is not a feature — it is the precondition for using agents at all. But even outside compliance, the calculus is simple: the blast radius of an agent platform is enormous, and you should be the only one holding the keys. Self-hosting means there is no third-party outage that takes your automation down, no vendor data breach that exposes your internal systems, and no quiet change to terms that suddenly lets someone train on your prompts.

We think the right default for AI that acts on your behalf is sovereignty: your infrastructure, your data, your control. That belief is the reason Nexora is shaped the way it is, and it is not something we plan to compromise on.