Leo is building an intelligence workspace for how work is changing

Cognexa Limited will publicly launch Leo on 1 September 2026. Leo is an intelligence workspace designed around getting work done with AI rather than treating intelligence as an isolated chat experience. This announcement is being provided to press ahead of that date; Leo is available by invitation until then.

Most AI products begin and end with a question box. Leo begins with the work. A person arrives with something they need to finish: a subject to research, a document to understand, a report to draft, images to produce, a project to organise, a decision to think through.

In practice that means researching a topic and analysing what comes back, attaching a document and working on it directly, drafting and restructuring written work, generating and editing images and video, organising work into projects with owners and briefs, and turning that work into tasks someone has to do. The intent is to carry a request towards a finished outcome rather than return one reply and forget it.

Leo is not a chat product with features attached. The conversation is one way into a workspace that also holds projects, files, tasks, generated outputs and the settings that govern them.

The most familiar limit of assistant tools is not the quality of any single answer. It is that the next conversation begins with nothing. The document is gone, the decision is gone, and the person explains their situation again.

Leo is built around working context that persists. A conversation has one durable identity: its messages, attachments and saved outputs are stored, listed and reopenable, and a thread started in Studio can be continued in the main workspace or the other way round. Around it sits the wider context of the workspace: projects and their briefs, decisions, references, files, tasks and the outputs already produced.

The engineering position is deliberately modest. For each turn Leo assembles a bounded, relevant frame of that context rather than pushing an entire history at a model, and it records what was retrieved and how much. Context use is a workspace preference a person can switch off. This is a foundation being extended, not a claim of total recall.

People will not always want to open a browser tab to reach their Leo. The platform roadmap therefore includes three access routes that are in development and are not available today: a Model Context Protocol interface, a versioned external API, and a supported command line.

Your Leo should remain your Leo regardless of where you choose to interact with it.

The intent is that a Leo workspace becomes reachable from environments people already work in, including assistants such as Claude and ChatGPT, coding agents such as Claude Code and Codex, compatible editors and terminal workflows. Someone could begin a piece of work in a coding agent, call a Leo capability that agent does not have, receive the result where they are working, and still find the output filed in the right Leo workspace afterwards.

Those products are named to describe the compatibility Leo is being built towards. They are the trademarks of their respective owners, and their mention does not indicate any partnership with or endorsement by those companies.

Leo is initially opening around individual workspaces, while the underlying platform is designed with team and organisational requirements in mind. The structure already reflects that: an account holds separate workspaces, each with its own data, brand and settings, so a person running several clients or brands keeps them apart. Roles govern who can act on an account, and usage and connected tools are visible in settings.

The direction from there is richer team structures, permission scopes, governance and organisational context on top of that model. Those are being built and are described here as direction, not as available product.

On certification, Leo claims nothing it has not earned. The platform is being built towards organisational security requirements, and any external certification will be announced when it is held.

Work is changing, but responsibility for it has not moved. Leo is built to widen what a person or an organisation can do, not to remove them from their own work. Consequential actions are proposed and approved rather than taken quietly: tasks carry owners, dependencies and approval gates, and the record of what was done stays attributable to people.

Leo is built by Cognexa Limited, a technology company based in Kenya. The product has been approached as a global platform from the outset. The work people bring to Leo, the research, the documents, the projects and the deadlines, is not particular to one market, and neither is the product. Where Leo is made is a fact about the company, not a limit on who it is for.