AI in Orneos

Orneos is built for teams whose work is increasingly agent-assisted. That is not the same as putting a model in front of everything, and this page is about where the line is.

AI and agent-facing surfaces

Surface	What it does	Enabled by
Intent drafting	Proposes a goal, acceptance criteria and non-goals from an issue.	Your own drafting key
Insights	Answers questions and groups related work.	Server-configured provider; applicable allowance and consent
Funded Intent drafting	Proposes an Intent through the server when enabled.	Orneos credit
MCP surface — work in progress	Serves context to your agent; the MCP tool itself does not perform inference.	Personal token; release acceptance pending

Context delivery, capture and review are active work. Autonomous agent execution remains a planned capability. Insights can synthesize an answer; that is different from background automation over every task.

The rules

These hold everywhere a model is involved, and they are the reason the list above is short.

A model proposes; a person decides

Every model output is a proposal. An AI-drafted Intent is a draft until a person approves it, and the approval records who. An Insights answer is an answer, not a fact about your data.

Model output can be retained as a draft or conversation, but it does not become an authoritative approval or human decision on its own.

Model involvement is disclosed

Where a draft was AI-assisted, that is part of the record. “Who wrote this agreement” is something a reviewer may reasonably want to know, and it should not require asking.

Provider and funding are explicit

BYO Intent drafting uses your browser-held key. Insights and enabled funded drafting use a server-configured provider, with the applicable consent and allowance controls. Absence of a personal drafting key does not disable every model call. Manual authoring and context assembly do not require a model.

Agents use the same substrate as people

In the MCP implementation in progress, separate from this app snapshot, an agent is subject to the same authorisation as a person: a token identifies a human, resolves to that human’s actual memberships, and is re-checked on every request. An agent cannot see a team you are not in.

When agents eventually write, they will use the same durable path, the same permissions and the same audit trail as human changes. A separate faster path for agents would mean the one class of actor most in need of an audit trail is the one without it.

Where your data goes

The data path depends on the feature. BYO drafting sends its request from your browser to your provider account. Insights and enabled funded drafting go through the server-configured provider. The interface discloses applicable data use and credit; linked Vault page bodies are not automatically read merely because an issue links them.

• Orneos does not train on your data.
• Provider configuration, consent and funding are specific to each surface.
• A context bundle makes no model call. Loading missing discussion can use the network; where you paste the exported context is your decision.

See Privacy and your data and Drafting keys for what is stored and where.

Why not more AI

The gap in agent-assisted development is not the model. It is that agents work from context assembled on the fly and leave no durable record of what they knew or why they chose what they chose — so when something is wrong, nobody can reconstruct it.

Adding a chat widget does not touch that. Making the agreement explicit, the context inspectable and the record durable does. That is where the effort goes, and it is why there is less visible AI here than you might expect from the category.

Related

• Drafting keys — configuring a key, and what is stored.
• Insights — questions over your board.
• Agent execution — the planned direction and what it will require.