Key terms

These are the terms used across the app and these pages. A longer list is in the Glossary.

People and agents

Workspace: One company’s isolated space in Polo: its members, agents, connections, tasks and records. Members have a role: viewer, operator or administrator. See Team and security.

Agent (a Polo): A Polo teammate with its own name, model, instructions, tools and limits. Its card on Agents shows what it can do, which changes need approval and what it never does. See Agents.

Supervisor: The one agent named Polo that takes every request that names no agent. It gives each part to an agent on its team and reviews the result. It has no tools of its own.

Specialist: A helper created during a task for one narrow job, such as an investigator or a verifier. It receives only the tools, budget and deadline that job needs, and never more than its parent holds.

Systems and access

Connection: A system Polo works with: a cloud account, a cluster, a repository, monitoring, an alert source or an MCP server. Its status is Connected once Polo has reached it, Registered while it is saved but not yet reached, and Reach failed with the reason when the latest request failed. See Connections.

Runner: A small Polo program on a machine you choose. It connects out to Polo, so it needs no public address and can reach systems inside your network or VPN. Where it is offered, Polo’s default runner serves cloud accounts without any setup. See Runners and vault.

Vault: Where passwords, keys and tokens are kept, encrypted. A stored value is not shown again. Agents and models never receive the values; the runner or connector uses them when an operation runs.

Model route: A model connection an agent thinks with, added under Settings → AI models. An agent has a default model, an optional fallback model that a task continues on when the default cannot answer, and optional specialist models. See Models and spend.

Work and control

Task: One piece of work, written the way you would ask a colleague. It has a budget taken from its agent.

Operation: One exact action on a connected system, such as “apply this saved Terraform plan” or “restart this machine”. Reads run directly. Changes are proposed first.

Approval: A person’s decision on one proposed operation. It is bound to that exact operation and expires. A standing approval is an administrator’s rule that approves one exact change in advance. See Approvals and guardrails.

Guardrail: A limit on an agent. Polo enforces three kinds whatever the model decides: read-only, the environments and services the agent may work in, and the approval every change needs. Forbidden actions are different: they are told to the agent but not enforced. A workspace can also set rules that replace, stop or record text before it is sent to a model.

Evidence: What Polo read to reach a conclusion or to verify a change: a plan, a machine’s state, job timings, an uptime read. It is recorded on the task.

Artifact: A durable output of a task, such as a report, a source change, a Terraform plan or an incident report. See Artifacts.

Skill: A reusable, versioned procedure an agent can follow. A skill does not grant permissions by itself. Terraform is a skill you add to an agent: it plans from a repository and applies only the exact plan a person approved. See Add Terraform to an agent.

Incident: An outage or alert Polo watches, shown on Incidents with what was found, what was changed and how it ended. See Incidents and uptime.

Capability maturity levels

Each area of work on Settings → What Polo can do carries one level. The levels are applied strictly.

Level	Meaning
Planned	Not built yet.
Implemented	Code exists, tested in parts only.
Verified locally	The real path was driven end to end against local tools or fixtures, and the outcome was checked.
Verified with a connected provider	A recorded run against a real external account.
Operated in a pilot	Used for real work in a pilot.