Agent Team

Agent Team staffs your business with specialized AI roles that handle recurring close, bookkeeping, payables, and cash analysis work. Roles can be added from the built-in role library, written from scratch, or generated automatically from an SOP. Every agent works within its assigned scope, instructions, skills, and monthly spend cap — and no agent ever writes directly to the ledger without human approval.

Key capabilities

• Five built-in library roles — Data Ops (read-only Phase 1 tie-out), Bookkeeper (the only writer on the close team; proposes categorizations, matches, and journal entries), Controller (independent read-only Phase 3/4 verification and close package), AP Desk (bill intake, approval chase, holds, and pay-list proposals), and Cash Analyst (13-week cash model and week-over-week bridge)
• Three ways to add a role — start from a Library role (pre-loaded with instructions and skills), build a full team From an SOP (extracts jobs, roles, and schedules in one step), or create a Blank role
• Strict write guardrails (Scope) — Read only roles (scope="read") can inspect the books and report findings but never propose edits; Proposes changes roles (scope="changes") queue every gated write in Proposed Changes for a person to approve or reject
• Standing instructions & modular skills — Markdown instructions define the role's standing orders on every run, paired with toggleable Skills (including run-protocol and Learned rules)
• Tool access & monthly spend caps — give a role Everything, or Only these: whole tool groups such as Bill Inbox or Accounts payable, chosen connected services such as Gusto, and exact tools under Advanced. Set a Spend cap per month (USD) that automatically refuses new runs once reached; put one on high-frequency roles so runaway loops or unexpected volume never exceed your budget
• Team-wide & per-role pause controls — pause or resume a single role, or click Pause team / Resume team to halt or restart all scheduled and manual runs across the business
• Engines & models — every role runs on the DayZero (Gemini) engine by default (google/gemini-4-argon), with preset or custom model selection; DayZero staff can also assign internal roles to Claude Code or Codex on firm subscriptions
• See who is working now — a Working now panel above the team lists every job in progress: the role, the job, what the agent is doing this moment, how long it has worked, and where the job came from (Scheduled, From chat, Assigned, or API). Jobs waiting on you say Needs you, and Watch opens the job's live transcript
• Hand off from anywhere — give a teammate a job in Team chat with @Name, give it work with Assign task, pick Message in Team chat from the ⋯ menu (opens Team chat with @Name already typed), type @ in your Regent chat composer to mention and task an agent directly, click Hand to an agent (the bot icon), ask Regent in plain English, or trigger roles automatically from Workflows

How it works

Each role on your team combines standing instructions, enabled skills, a permission scope (Read only or Proposes changes), and an optional monthly spend cap. When you start a job — manually, from an SOP schedule, or from Regent — DayZero launches an Agent Run using the requester's permissions on that business. Read-only tools execute immediately; any gated write is previewed and filed into Proposed Changes so a person stays in control of the books.

flowchart LR
  source["Role library / SOP / Blank"] --> role["Configure Role (scope, skills, cap)"]
  role --> trigger["Start job (Team / Regent / Schedule)"]
  trigger --> run["Agent Run executes steps"]
  run --> scope{"Write needed?"}
  scope -->|"Read only"| report["Verify & post run report"]
  scope -->|"Proposes changes"| prop["Queue in Proposed Changes"]
  prop --> report

How to use it

1. Open Agent Team in the Regent workspace.
2. Click New role and choose Data Ops, Bookkeeper, Controller, AP Desk, Cash Analyst, From an SOP, or Blank role.
3. In the role drawer, review or edit the Name, Title, and Markdown Instructions (the standing orders the role follows on every run).
4. Review the Skills list: toggle individual skills on or off, remove skills you do not need, or add a new skill with a lowercase-dash name (for example bank-recs), description, and Markdown body.
5. Set the Scope to Read only (never proposes changes) or Proposes changes (queues gated writes in Proposed Changes), and optionally set a Spend cap per month (USD). Under Tool access, keep Everything or pick Only these and tick the tool groups and connected services the role needs.
6. Click Add to team (or Save when editing an existing role).
7. Click Assign task on any active role's row, describe what it should do and an optional Done when check, and click Start to launch the run and open its live transcript. Or click Chat on its card to see its work in Team chat and give it a job there (see Team Chat).
8. To follow a job while it runs, look at Working now at the top of the page. Each line shows what the agent is doing right now (for example Using Gusto · get_payroll_register) and updates every few seconds. Click Watch to open the job's live transcript. The role's row in the team table shows the same line under its status. The panel disappears when no job is in progress.
9. Use Pause in a row's ⋯ menu to temporarily stop a single role, or Pause team in the header to pause every role on the business at once.

Team chat and the Team tab

Every job a teammate does, wherever it started, is a post in the business's shared # Team room on the Chat tab, with its live Activity log. Anyone who can make changes gives teammates work with @Name, Adjusts or Stops a job under its post, answers its questions, and approves or rejects what it proposes there. Your own Regent · just you chat stays private. See Team Chat.

Every teammate has a color, a vibe and an avatar with its initial. Regent sits at the top of the team as Chief of staff: it hands out work, checks it and tells you who it gave a job to. The Team tab shows a card per teammate (switch to Table with the toggle above the cards), and Regent's welcome lists the team with what each one is on; click one to see its work in Team chat. In the role editor, Look & personality sets the color, vibe (Steady, Upbeat, Witty (fun mode) or Blunt) and an optional greeting. The vibe changes how it talks in chat only: memos, ledger text, documents and proposed changes stay plain, numbers stay exact, and an Adjust never skips approval.

Pro tips

• Use Client Operations for website reports, business email, receipt requests, portal conversations and routines learned from completed work.
• Keep separation of duties intact: let Bookkeeper and AP Desk run with Proposes changes, while keeping Data Ops, Controller, and Cash Analyst on Read only so the verifier never silently edits what it checks.
• Always include a concrete Done when check when running a job (for example, "Every September transaction is matched or listed as an exception"). Every library role checks Done when first so re-running a finished job changes nothing.
• Use From an SOP when onboarding a new client or codifying a multi-role process — DayZero reads the SOP's jobs, matches owners to library templates, merges your business-specific instructions, and schedules the recurring workflows automatically.

In-depth guide

Built-in role library

Role drawer fields and guardrails

Tool access

Everything lets the role use every DayZero tool its scope and the requester's permissions allow, and every connected service. Only these narrows it to what you tick:

• Tool groups — one per area of DayZero, such as Bill Inbox, Accounts payable, Banking, or Month-end close. Each shows how many tools it holds.
• Connected services — the outside servers connected under External MCP servers. A role reaches a service only when it is ticked here. A service whose connection is limited to other roles, or has agent access off, says so under its name.
• Advanced — exact tool names, on top of the groups. Names saved before tool groups existed stay here.

Tool access only narrows what a role can do. Its Scope still decides whether it can propose changes, and every change still waits in Proposed Changes.

How every role runs (run-protocol)

Every library role includes the run-protocol skill, which enforces an eight-step discipline:

1. Work out the context — business days (BD1, BD2, …) follow the Federal Reserve bank-holiday calendar; the close period defaults to the prior month during the first 20 days of a month.
2. Load what it knows — combines standing instructions, enabled skills, and client facts (parameters) extracted from the SOP.
3. Re-read the live books — never reuses figures from an earlier run; checks close checklist steps first on close jobs.
4. Check "Done when" first — if the completion condition already holds, records the step and stops without making duplicate changes.
5. Execute steps in order — previews writes first and files gated writes as Proposed Changes in batches.
6. Verify by re-reading — re-reads the ledger or trial balance to confirm the intended state.
7. Report once — summarizes what was done, what waits on whom, and every exception with a dollar amount.
8. Record the close step — logs completed, blocked, failed, or skipped status to the Monthly Close Checklist when linked to a close item.

Permissions and identity

An agent never has elevated privileges of its own. Every run executes under the requesting DayZero user's identity and business permissions. Roles on Claude Code or Codex can use every DayZero tool, including the Bill Inbox, workflows, the Daily Brief, and close status. A few actions there need a person's approval in a way an agent can't file for you; the agent leaves those out and says in its report what you need to do. Deleting a role removes it from the active team and stops its scheduled jobs, while preserving all past runs and transcripts in Agent Runs.