Your agent’s data is your playbook.

What’s in a trace

Every time an agent works, it leaves a trace: what it was asked, what it decided, which tools it called, what came back, and whether the job got done. Across thousands of runs, those traces add up to a detailed record of how your company operates.

• Your workflows. The exact steps, tools and systems your agent uses to get work done.
• Your judgment. How edge cases are handled, and what a good result looks like in your business.
• Your customers. The requests, context and data behind every run.

Traces can teach a model your job

Models learn from examples. Give a model enough traces of a task being done well, and it can learn to do that task. That is exactly how you train an agent to get better. It is also how someone else could train a model to do what your agent does, without the years it took you to learn how. Training one model on another’s outputs, known as distillation, is common enough that AI labs now write it into their terms and watch for it.¹

Where your traces go today

Most agents send every step to a model behind someone else’s API, and often to other services that log, monitor or store the runs. Each of them handles your data under its own terms. Retention rules differ, terms change, and once the data has left your environment, you can’t take it back. In 2023, Samsung restricted generative AI tools after employees pasted internal source code into a chatbot, citing how hard such data is to retrieve and delete.² In 2025, a US court order required one major provider to preserve output logs, including from its API, that would otherwise have been deleted.³

Most of this happens without a plan. In IBM’s 2025 breach study, one in five organizations reported a breach involving unsanctioned AI tools, and 97% of those with AI-related incidents lacked proper AI access controls.⁴

None of this requires bad intent. It only requires your playbook to sit somewhere you don’t control.

What it costs if they leave

For a company built on agents, the moat isn’t the model. It’s knowing how to do the work well: the workflows you refined, the edge cases you learned to handle, the customers you understand. Your traces hold all of that. In the wrong hands, they become the fastest way to build a product that competes with yours.

Depending on a model you don’t control carries its own risk. In 2025, one AI coding company said it lost most of its direct access to a major model family with less than five days’ notice.⁵ When your workflows and your model live with you, decisions like that are yours.

How to keep them

• Keep traces in your environment. Capture every run in storage you control, in our cloud, your VPC or your own data center.
• Limit who can see them. Access only for the people you choose.
• Train on them yourself. Turn your traces into a model of your own, so the advantage they hold works for you.
• Own the result. The model and its weights are yours, and run wherever you decide.

That is what Voltspan is built for. Your data stays yours, it trains only your model, and it never leaves the environment you choose.

See how the platform works →