Core concepts
Understand colleagues, work, results, approvals, sandboxes and governed memory.
ColleagueOne models AI work as a set of explicit organisational objects. Understanding those objects makes the product easier to operate and govern.
Colleague
A colleague is a defined AI role with a persona, tools, rules, model and optional schedule. It is a reviewable configuration, not a human employee and not an unrestricted general-purpose agent.
Team of colleagues
A team groups specialised colleagues around a broader outcome. Each role can retain its own tools, instructions and boundaries rather than collapsing all capability into one profile.
Work and run
Work is the goal and its durable context. A run is an execution attempt for that work. Keeping these separate allows an outcome to retain history even when execution pauses, needs guidance or is tried again.
Sandbox
Each colleague runs in its own Firecracker microVM. This isolated cloud computer is where it browses, works with files and uses its available tools. Credentials remain in the gateway rather than being handed to the colleague.
Result and artifact
A result is the reviewable outcome of work. An artifact is a file produced or collected during it, such as a document, presentation, PDF or spreadsheet. A successful run means the system finished its process; it does not remove the need to verify important content.
Needs You and approval
Needs You gathers questions and decisions that require a person. An approval is an explicit choice about a sensitive proposed action. Review the action, data and destination rather than approving from the title alone.
Skill, plugin and connection
A skill provides reusable know-how. A plugin extends available capability. A connection gives controlled access to a vendor service, usually through OAuth. Organisation policy and the colleague profile determine how each may be used.
Memory
Memory contains learned facts proposed for future use. Users can review, approve and revert those facts. This makes persistence a visible decision rather than an invisible side effect.
Governance layer
Roles, identity, scoped model keys, budgets, data classification, DLP policy, approvals and audit history form the governance layer around colleagues and their work.