The vocabulary
of hybrid teams.

A named, persistent AI teammate with a role, specialty, and rate. Unlike a model call, an agent remembers prior work, can be coached, and has scoped permissions across your tools.

An append-only record of every action an agent takes — tool calls, file edits, messages sent. Streams to SIEM. Required for SOC 2, HIPAA, and most enterprise deployments.

A spending cap set per-agent, per-team, or per-project. Agents stop executing (and page a human) when they approach the cap. Prevents runaway costs.

A free-text correction or preference you attach to an agent after a review. Coaching notes are injected into the agent’s context on every future task, permanently.

The maximum amount of information an agent can consider at once. TwoPlus manages this invisibly by summarizing, retrieving, and splitting work — you rarely think about it.

The act of handing a task to an agent with a spec, a deadline, and a success condition. Delegation is the atomic unit of work in TwoPlus.

A structured test of agent performance against a fixed rubric. TwoPlus runs evals nightly on your agents and alerts you if quality drifts.

The action of permanently removing an agent from your org. All memory is deleted within 24 hours. See also: Pause (reversible).

A hard rule an agent cannot violate — for example, "never contact customers outside the US" or "never push to main." Guardrails are enforced at the tool layer, not the prompt.

The action of adding an agent to your org and paying the seat fee. Hiring requires a role, a reporting line, and an initial budget.

When an agent passes a task to another agent or to a human. Handoffs carry full context — no re-briefing.

A single run of an agent on a task. Billed per invocation (seat + usage). Most workflows use 1–3 invocations per task.

The persistent store of everything an agent has done, been told, and learned. Semantic-indexed, encrypted, deletable. Distinct from context window.

The visual representation of your team in TwoPlus — humans and agents, reporting lines, teams. Drives permissions and notification routing.

An org-wide rule about what agents can and can’t do. Policies compile down to guardrails at the tool layer.

Short for "review queue" — the unified inbox of agent actions awaiting human approval. Replaces reviewing work across GitHub, Linear, Slack, etc.

The action a human takes on an agent’s proposed output: approve, request changes, or reject. Every review produces a coaching signal.

What an agent does — "staff engineer," "account executive," "brand designer." Role determines the agent’s tools, permissions, and defaults.

A sub-focus within a role — "Rust systems," "outbound cold email," "iOS motion design." Specialties are labels agents advertise on the hire page.

The written brief you give an agent when delegating. Good specs include the goal, success conditions, constraints, and context.

The umbrella term for both humans and agents in your org. TwoPlus is designed around the idea that they should be managed the same way.

Any external system an agent can act on — GitHub, Figma, Salesforce, a SQL database, your phone system. 40+ native, plus bring-your-own via webhook.

The metered component of your TwoPlus bill — compute, API calls, third-party costs passed through. Usage is transparent, line-itemed, and capped by your budget.

An agent that can answer or make phone calls with sub-second latency and human-interruptible speech. Built on our voice infrastructure.

A TwoPlus account for a company. Workspaces can have sub-teams, sub-budgets, and separate SSO configurations.