BLOG
Notes on owning your agent runtime.
Sovereign / Self-Hosted AI

Ask the flow what it did
A run took an odd branch and nobody can say why. Six checks for debugging an AI agent workflow run without reading every step by hand.

It worked in test. Moving it to production broke it.
An agent that works on the test installation can fail on the live one. Seven checks for moving an AI agent or workflow from test to production.

The run history filled the disk
Stored run data can grow until the disk is full and the server stops. Seven checks: what grows on your agent server, and who notices before it is full.

The agent kept answering. It had lost its knowledge.
An AI agent can lose its knowledge base and keep answering, with no error. Seven checks to know your agent still answers from your own documents.

It works for the builder. It fails for the people you share it with.
An agent that works for its builder can fail for the people it is shared with. Seven checks to run before you share an AI agent or workflow.

Before you upgrade your self-hosted agent server
Four self-hosted agent servers broke after a normal upgrade. Seven checks before you upgrade yours: version, what changed, backup, database changes, roles, the way back, a test copy.

Whose key is it? When the AI in your tools runs on the vendor's account
On a hosted tool, the AI often runs on the vendor's key: its quota, its gateway, its model, its bill. What to check before you depend on it.

How much load can it take? Runs at the same time
Long jobs time out or hold up other work, and nobody knows how much load the tool takes. Five questions to ask about runs at the same time.

Data residency for AI agents: which data sits where, and what to put in the contract
An AI agent produces five kinds of data, and they can sit in five different places. Ask about each one separately, then write the answers into the contract.

Who decides when the model behind your agent changes?
A vendor model update can change a live agent that nobody touched. What a pinned model version gives you, what it cannot, and who decides the change.

The screen said success. The record said something else.
The screen says finished, the database says running, the error names the wrong cause. Why status and record disagree, and which one to trust.

EU AI Act Article 50: transparency by design
Article 50 has applied since 2 August 2026. What it asks of providers and deployers, why a line in your terms is not enough, and why transparency is native to a governed platform.