Concepts

A short mental model for how CruzClaw works. Every page in these docs assumes these terms.

Employee

An AI worker you hire. An employee has a display name, a role or job title, a tagline, and editable instructions that define how it behaves: what it owns, how it communicates, and what it must escalate. You create one from a ready-made template (for example a CFO or an executive assistant) or build a custom one from scratch. Employees go through a lifecycle: provisioning right after creation, then active while working, and paused or archived when you stop them.

Example: you hire “Atlas” as a sales development representative, with instructions to draft outreach every morning and escalate any reply that mentions pricing.

Persona

The identity half of an employee: the name, role, avatar, and instruction files that make two employees behave differently even on the same task. A template comes with a persona pre-loaded; you can edit the persona later from the Employees page without reprovisioning the runtime. In the dashboard and most of these docs, “employee” and “persona” refer to the same thing.

Example: two employees hired from the same research-analyst template can diverge completely after you rewrite one persona’s instructions to prioritize competitive intelligence over market sizing.

Deployment

The dedicated runtime that an employee runs on. One deployment backs exactly one active employee; the product enforces this at the database level, so an employee can never silently share a runtime (or its workspace, memory, and channel credentials) with another. Deployments are what you pay for: each employee is a separate line on your bill, and each one can be restarted, reset, or deleted independently from the Infrastructure page.

Example: your CFO and your EA each run on their own deployment; restarting the EA’s runtime for a config change does not interrupt the CFO.

Owner

You, the human account holder. The owner directs employees, sets their instructions and permissions, and is the only one whose approval unblocks consequential actions. Every approval records who decided and when, and rejections require a reason so the employee learns what to do differently.

Example: your EA prepares a reply to an unhappy customer and asks to send it. Nothing goes out until you, the owner, approve the exact message.

Leader

On a TEAM plan, one employee you designate as the team leader from the Organization page. The leader sits at the top of the reporting tree: you can delegate through it, and it coordinates work that spans multiple employees. The leader must be an active employee, and an executive-assistant employee cannot hold the role.

Example: instead of telling three employees separately to prepare the board deck, you tell the leader once; it breaks the work down and reports back.

SOLO and TEAM plans

Plans differ in how many employees you can run and how they collaborate:

• SOLO and SOLO Pro include exactly one digital employee each, with monthly AI credits and human approval workflows. SOLO Pro adds more credits and higher execution limits.
• TEAM includes up to ten digital employees, shared monthly credits, and multi-employee features: the Organization page, the leader, delegation and reporting lines, discussions and debates, and the Decisions page.

The plan surfaces that require multiple employees (Decisions, Organization, Orchestration, Notebook, Communications) are hidden from SOLO accounts rather than shown as locked teasers.

Example: a founder running one EA stays on SOLO Pro; a three-person startup that wants a leader with two reports upgrades to TEAM.

Decisions

A recorded proposal your employee raises when it believes a judgment call is yours to make: a topic, the question, the context, its recommended answer with rationale and a confidence score. Decisions land on the Decisions page, where you approve or reject. Approving can be a plain sign-off, or you can route follow-up work through the Action Box. Rejecting requires a reason, which is recorded on the decision. The Decisions page unlocks once a second employee joins your team.

Example: your analyst proposes “Pause outreach to segment B for two weeks” with rationale and a 0.8 confidence score; you approve it, and the plan is recorded for later review.

Action Box

A free-text input on a decision, also shown when you approve, that turns your verdict into routed work. Type an instruction, @ to tag a specific employee, and # to attach an integration with one of its available actions. The box confirms coverage when you submit, so you can see who was instructed and which integrations were triggered.

Example: approving “Follow up with the vendor” and typing in the Action Box: “Draft the renewal email @EA and #email, and get my approval before sending.”

Approvals

The gate in front of consequential external actions. When an employee wants to send an email, message a contact, or execute a gated tool, the action pauses as a pending approval showing the tool, the action, and a secret-free summary of the payload. You can approve, reject, edit the payload and approve, or answer with an instruction instead. Approved actions execute exactly once. Pending approvals surface in the dashboard’s top-bar notifications, so you don’t have to watch any particular page.

Example: your EA drafts a meeting request; you open the notification, tweak the time in the payload, approve, and the single edited version is sent.

Work loops

A durable unit of ongoing work that an employee drives: a goal, a plan of steps, its current status, and any open questions for you. Work loops appear on the Work Loops page, grouped into sections like Needs you (waiting on your input or approval), Stuck, Interrupted, Waiting on others, and Queued, so you can see at a glance what every employee is doing and where each one is blocked.

Example: “Prepare the Q3 board update” becomes a work loop with a five-step plan; it sits in Needs you until you answer the employee’s question about which revenue figure to use.

Scheduled routines and briefings

Recurring work you set on a schedule (a cron expression, a timezone, and a task) from the Scheduled Jobs page. A routine can run as a plain operation or be executed by a specific employee, and daily briefing routines deliver a regular summary (one per employee on SOLO, a consolidated briefing on TEAM). Overdue runs either catch up or skip, according to the job’s catch-up policy.

Example: every weekday at 8:00 your timezone, your EA compiles an inbox-and-calendar briefing before you start work.

Last verified: 2026-10-24 against the shipped codebase.