Privacy notice

This notice describes how Agent Labora handles data in the course of running an agent platform. It covers the marketing website you are reading and the product at https://app.agentlabora.com.

The marketing website sets no analytics or advertising cookies, runs no third-party tags and holds no account data. It is static content served from a CDN.

For the data your agents read and act on — your customers, their invoices, their conversations — you are the controller and we are the processor. We hold and process it on your behalf, under the configuration you set.

For the developers and administrators on your team who hold a login, we are the controller of their account data.

If one of your customers asks what is held about them, that request goes to you. We cannot disclose it to them on your behalf.

Agent configuration: purposes, capabilities, grants, limits, approval rules, owners and versions.

Run data: what triggered a run, what the agent read, what it decided, what it did, what it asked about, what it cost.

Business data reached through granted capabilities: customer records, invoices, conversations, cases, content and analytics — much of it read from systems you connect rather than stored here.

Audit records: the actor, inputs, policies applied and result of every consequential action.

It does not use your agents' work, your customer records or your approved knowledge to train models made available outside your account. That is a contractual commitment, not a preference, and it binds the model providers we use.

It does not share data between customer accounts. Cross-organisation reads are not a permission anybody can be granted, including us.

It does not monitor your team. No keystroke tracking, no time-online, no idle detection and no per-person scoring exist in the product. Agents read roles, responsibilities and approval authority — never salary, performance or assessment data.

It does not characterise the people your agents contact. No product output describes an individual as difficult or unreasonable, and none produces a score to that effect.

It does not make you compliant with anything. No configuration of any software does that.

Message bodies, attachments, imported pages and event payloads are treated as data. An instruction inside them cannot change how an agent behaves, what it may access or what it may decide.

This is stated here as well as in the security documentation because it is a processing commitment rather than only a technical one: content reaching an agent does not acquire authority by arriving.

Where a capability depends on a person's consent — individual-level behavioural data in particular — that state is read from your consent system per use rather than assumed or cached here. An absent or unreadable state is treated as a refusal, and failing open is not a configuration option.

Retention is set per category: run data, audit records, agent memory scopes and business data reached through capabilities can each carry their own period, because the obligations genuinely differ.

Audit records of consequential actions are append-only for their retention period and cannot be edited, including by an administrator — an audit trail somebody can quietly tidy is a record of what they wanted you to see. Records under legal hold are retained regardless, and the hold is visible and attributed.

Write to [email protected] with any question about this notice.

If an agent belonging to a company that uses Agent Labora contacted you and you want to know what they hold about you, ask them. They are the controller of that data and we cannot disclose it to you on their behalf.