Glossary

Agent

A dedicated Pneum.ai assistant environment with its own cloud computer, disk, runtime, credentials, notes, tools, and conversation state.

Cloud Console

The authenticated control plane used to launch, monitor, schedule, estimate, start, stop, and terminate hosted agents.

Agent workspace

The interface served by a ready agent. It contains chat, voice, Profiles, Notes, Vault, Tools, and Agents Studio.

Runtime profile

The launch-time compute/software class. CPU supports everyday workloads. GPU adds eligible GPU-backed capabilities such as hosted Voice Studio and compatible open-weight models.

Profile

A workspace configuration for assistant identity, system prompt, provider, model, speech, and activation. Profiles on one agent share the same underlying VM and storage.

Tool

A structured callable operation exposed to a model under a tool profile and approval policy.

Integration

A marketplace package that supplies tools, authentication, resource allowlists, and sometimes an external chat gateway.

Tool profile

A least-privilege capability set for a main or background agent, including allowed tools, guidance, skills, and approval behavior.

Subagent

A background agent launched for a bounded specialized task such as research, coding, notes, or scheduling.

Managed model

A model request authorized through an agent-specific key provisioned and metered by Pneum.ai.

BYOK

Bring your own key. A model provider credential you supply and the provider bills directly.

Approval

A decision required before a permitted tool executes a particular request. Approval can be one-time, session-bound, persistently argument-bound, or tool-wide when policy allows.

Allowlist

An explicit set of permitted IP ranges, repositories, channels, projects, chats, object types, or other resources.

Tailscale

A private mesh network option used for agent UI access and approved private-service connectivity.

MagicDNS

Tailscale DNS naming used to reach a tailnet agent by stable private hostname.

Boot disk

Persistent storage attached to an agent VM. It continues billing while the VM is stopped and is deleted on termination.

Estimate

A planning value based on selected provider pricing and configuration. It can differ from delayed, reconciled provider charges.

Reconciliation

The process that attributes provider-exported cloud or model usage to an account and agent before metering or invoice presentation.