Kanbrio AI and Agents Notice

Version 1.0 · Effective 29 September 2026

Published at: https://kanbrio.com/aiProvider: [to be confirmed] (ABN [to be confirmed]), trading as Kanbrio, of [to be confirmed]Contacts: [email protected] (data protection) · [email protected] (contract notices) · [email protected] (security)

This notice explains how AI agents and AI features work in Kanbrio. It forms part of our Terms of Service (https://kanbrio.com/terms, clause 1.4). The Terms (section 5), the Privacy Policy (section 6) and the Data Processing Agreement (section 17) say the same things in more detail. If this notice and one of those documents differ, that document applies.

Where a feature is not finished yet, this notice says so: it reads "being implemented", and says what happens in the meantime.

1. Agents your organisation connects

1.1 Your organisation can connect its own AI agents to Kanbrio, for example a coding assistant or a script, through our API or through our MCP endpoint, which is being implemented (meanwhile, agents connect through our API).

1.2 Each agent is a member of your organisation with the "agent" role. It is shown as an agent wherever it appears, never as a person, and every action it takes is recorded in the audit log as an agent action.

1.3 Kanbrio-issued agent API keys are being implemented; meanwhile each agent has its own separate credential, which only owners and admins can create and which they can revoke at any time. Keep an agent's credentials secret, and revoke them if you think they have been exposed.

1.4 The AI model behind an agent you connect is chosen and run by you, under your own terms with its provider. That provider is not one of our sub-processors. Whatever your agent reads from Kanbrio is sent to that model under your arrangements, not ours.

1.5 An agent acts with the role and permissions your organisation gives it. As between you and us, what your agents do in Kanbrio counts as done by you (Terms clause 5.1).

2. Human control

2.1 Every change an agent makes goes through the same single path as a person's change. That path checks permissions and writes an audit record.

2.2 When an agent asks to do something destructive (for example deleting or archiving work), Kanbrio does not do it straight away. It creates a suggestion that waits until a person in your organisation accepts or rejects it. Agents cannot approve suggestions, and they cannot request the erasure of personal data.

2.3 Agent actions that Kanbrio records as reversible can be undone by a person with the right permissions.

2.4 Pausing, resuming and cancelling an agent's run is being implemented. Once it is available, it is cooperative: the agent stops at its next safe point, not instantly. Meanwhile, to stop an agent, revoke its credentials.

3. AI that Kanbrio runs

3.1 Hosted coding executor. The hosted coding executor is being implemented and is not yet offered to customers. Once it is offered, a coding task works like this: a server Kanbrio rents from DigitalOcean in Sydney, Australia copies the repository you name; it sends the task title and instructions and at most 24,000 characters of relevant file contents (no more than 8,000 from any one file) to Anthropic in the United States, which proposes a change; the change is checked in an isolated environment with no network access and no credentials; and nothing is merged until a person accepts it.

3.2 Your own AI key. Using your own AI provider key is being implemented. Once it is available, hosted AI calls made with your key are governed by your agreement with that provider and do not use Credits.

3.3 Search. Semantic search is being implemented. Once it is available, it is computed on our own servers in Australia, and no text is sent to a third party to build search indexes.

3.4 Credits. Agent credits are being implemented; until they are, hosted AI is not charged, and rate limits cap how much an agent can do. Ordinary actions through the API, including actions by your agents, do not use Credits.

3.5 Usage records. The AI usage ledger is being implemented. It records the model, token counts, cost and timing of each hosted AI call, and never the prompt or the response.

4. Your data and model training

4.1 We do not use your data to train AI models, and we do not let our AI provider do so. Anthropic's commercial terms exclude training on data sent through its API, and Anthropic deletes API inputs and outputs within 30 days, except where it must keep them longer to investigate misuse or meet a legal obligation.

4.2 We recommend that you do not put personal information in coding tasks or agent instructions.

5. AI output

5.1 AI can be wrong, incomplete or insecure. Review what an agent or AI feature produces, especially code, before you rely on it, publish it or put it into production.

5.2 Do not use AI output as the only basis for a decision that has a legal or similarly significant effect on a person (for example about their employment, credit or access to services) without meaningful human review.

6. Automated decisions

Kanbrio's automation rules and agents create, change, move, assign and comment on work items, using the information in your organisation's workspace. We do not use Kanbrio to make decisions about individuals that have legal or similarly significant effects solely by automated means. If your organisation sets up its own automations or agents to act on information about its people, your organisation decides what they do and is responsible for them.

7. Questions

Email [email protected] about personal information, [email protected] about these terms, and [email protected] to report a security issue (see https://kanbrio.com/security).

Change log