---
url: /handbook/start-here/who-uses-dutygraph.md
description: Each role that touches DutyGraph, what they see and do, and how they get in.
---

# Who uses DutyGraph

DutyGraph is advisor-led. One advisor runs each company. Most client people take part through private links and need no account. People who approve agent access use their own DutyGraph sign-in.

## At a glance

| Role | Main job in DutyGraph | How they get in |
| --- | --- | --- |
| Advisor or FDE | Runs the whole engagement | Advisor account (email and password) |
| Client sponsor | Corrects the company picture, sends the team list | Private link, no account |
| Executives | Join the kickoff, answer their own asks | Private links, no account |
| Employees | Describe their own work, review an agent plan | Private links, no account |
| Managers and approvers | Start agent requests, decide in the approval chain | Their own DutyGraph sign-in |
| IAM and Internal Controls | Supply access and rules, decide their chain steps, provision in their own tools | Their own DutyGraph sign-in |
| Auditors | Trace samples of agents requested through DutyGraph | Through the advisor today |
| AI platform teams | Run approved agents in their own runtime | Exported files and a per-agent gateway credential |
| Department heads (Planned) | Sponsor discovery for one department | Not built |
| Application owners (Planned) | Validate which system permissions a task really needs | Not built |

## Advisor and forward-deployed engineer

An advisor or a forward-deployed engineer (FDE) operates DutyGraph for a company. Both use the same app.

* **What they see.** Home, the Inbox, Companies, Settings and Help. For each company: Overview, Discovery, Repo, Work Map, Opportunities, Agents, Outcomes, Strategy and Reports.
* **What they do.** They run the [9 steps](run-a-company.md). They review every proposal before it becomes a fact. Only an advisor can set up governance for a company. On a sample company, they operate the sample personas, and each decision records "operated by" the advisor.
* **How they get in.** An advisor account with an email and a password. For scripts and AI tools, a personal access token from **Settings › MCP access**, with the scopes `read`, `write` and `send_email`. An Admin can share a company with others as Admin, Assistant or Read-only from **Settings › Members & roles**.
* **Read next.** [Run engagements](../advisors/index.md), [DutyGraph Agent](../advisors/repo-copilot.md), [MCP for advisors](../connect/mcp.md).

## Client sponsor

The sponsor is the client's main contact.

* **What they see.** A private sponsor page, one screen at a time: **Is this right?**, the questions, **Who works here?**, **Anything else?** and **Here's what we heard**. Later, the advisor can send a status page and delivered reports as private links.
* **What they do.** They correct the company facts from research. They answer the questions or say **I do not know yet**. They upload the team list and tick who joins the executive kickoff.
* **How they get in.** A private link. No account is needed. To read results, the advisor can invite them with a single-use sign-in link that works for 24 hours. That view is read-only. It shows reports the advisor approved, never raw evidence or other companies.
* **Read next.** [Sponsor packet](../discovery/sponsor-packet.md), [Reports](../opportunities/reports.md), [Engagements and payment](../advisors/engagements-and-payment.md).

## Executives

* **What they see.** A private pre-read page for the kickoff: the date, the place, the agenda, a **Yes** or **Can't attend** reply and optional pre-work questions. After the kickoff, they get their own team asks first.
* **What they do.** They attend the kickoff and describe how their part of the company works.
* **How they get in.** Private links. No account is needed. The advisor can also give an executive the same read-only results view as the sponsor.
* **Read next.** [Executive kickoff](../discovery/executive-kickoff.md).

## Employees

Employees are the people who do the work.

* **What they see.** A private page built for a phone. They choose **Start with voice** or **Type instead**. Then they see one question per screen, a tap range for hours a week, **Here's what we heard** and **Send**. Later, they may get an agent plan page labelled "Proposed plan · not deployed".
* **What they do.** They describe their own work in their own words. They confirm the version DutyGraph understood. On an agent plan, they answer **This looks right** or **Something's off**. They cannot approve or change anything on the plan page.
* **How they get in.** A private link. No account or password is needed. Each link opens only that person's own questions or plan.
* **From their own AI.** An employee with a DutyGraph sign-in can request an agent from their own AI chat over the Model Context Protocol (MCP). They use a token with only the `request_agents` scope, bound to their person record. They create it in **Agents › Requests › Connect your AI**.
* **Read next.** [Team asks](../discovery/team-asks.md), [Share the agent plan](../agents/agent-plan.md), [Request from your AI](../governance/request-from-your-ai.md).

## Managers and approvers

* **What they see.** Only the requests they own or decide. On a phone, the approver gets one bar with the three decisions.
* **What they do.** On a real company, only the Task Card's owner can start an agent's access request. Each approver decides in **Agents › Review queue**: **Approve as proposed**, **Approve with changes** or **Deny**. The requester never decides. A person may decide only one step of the same request.
* **How they get in.** Their own DutyGraph sign-in, linked to their person record in the company. Managers also answer team asks through private links.
* **Read next.** [Request a governed agent](../governance/request-a-governed-agent.md), [Approval chain](../governance/approval-chain.md).

## IAM and Internal Controls

IAM (identity and access management) and Internal Controls already approve human access. DutyGraph routes agent access to the same people.

* **What they supply.** An access export as a CSV file, the segregation of duties (SoD) rules, the mitigating controls and the names in the approval chain. The advisor loads them in **Agents › Rules › Set up governance**.
* **What they do.** Internal Controls checks SoD conflicts and may tag a mitigating control. IAM issues the agent with **Record the issuance**. The gateway credential shows once. DutyGraph gives IAM a "For IAM to do" card. DutyGraph records the justification and the decision; the company's own IAM tools create the agent identity and grant the access.
* **What to know.** The evaluator keeps an agent within its owner's holdings and the rules. It does not yet limit scopes to what the Task Cards need, so reviewers should check each scope against the work. When an owner leaves, DutyGraph does not suspend their agents today. IAM or Internal Controls revokes them. See [Built and planned](built-and-planned.md).
* **How they get in.** Their own DutyGraph sign-in, as approvers.
* **Read next.** [The key join](../governance/the-key-join.md), [Holdings](../governance/holdings.md), [Rule packs and SoD](../governance/rule-packs-and-sod.md), [Set up governance](../governance/set-up-governance.md).

## Auditors

* **What they see.** **Agents › Audit**: the register of agents, the ledger with **Verify chain** ("Chain intact"), and the Auditor workspace. The workspace draws a reproducible sample and opens the trace from an agent to its owner, Task Card, rules and decisions.
* **What they do.** They trace a sample of agents requested through DutyGraph. They take the evidence as CSV or JSON. The register covers only those agents. Reconciling it with every agent identity in the company's own systems is **Planned**.
* **How they get in.** DutyGraph has no auditor role today. The advisor opens the Auditor workspace, draws the sample and exports the evidence for the auditor.
* **Read next.** [Ledger](../governance/ledger.md), [Audit evidence](../governance/audit-evidence.md), [Evidence and provenance](../trust/evidence-and-provenance.md).

## AI platform teams

* **What they see.** Agent files formatted for Claude Code, Claude Managed Agents and OpenAI Agents, with a `GAPS.md` file that lists what is missing.
* **What they do.** They load the files into the company's own runtime. They can point the agent's tools at the optional DutyGraph gateway. It checks each call routed through it against the agent's status, expiry and issued scopes. Calls made another way are not checked. The gateway has no per-action human approval; use the runtime's own approval feature where one exists.
* **How they get in.** Each issued agent has its own gateway credential. It shows once at issue and can be revoked.
* **Read next.** [Exports and runtimes](../agents/exports-and-runtimes.md), [Agent gateway](../governance/agent-gateway.md).

## Planned roles

::: warning
**Planned.** Neither role is built. Today department heads answer the same team asks as everyone else.
:::

| Role | What they would do | Read next |
| --- | --- | --- |
| Department head | Sponsor discovery for one department, set its priorities and choose who takes part, without a full company map | [Three ways in, one model](three-ways-in.md) |
| Application owner | Validate which permissions in their system a task step needs. This is evidence about the system, not approval of the access | [Task Cards](../discovery/task-cards.md) |

::: info
**Status.** Advisor accounts, private links without an account, the read-only client results view, own sign-in for owners and approvers, advisor MCP tokens, employee request tokens, exports and gateway credentials are **Built**. The business systems behind the gateway are **Simulated**. Single sign-on (SSO), OAuth sign-in for web and mobile AI connectors, hosted agents, run tracking, the department-head and application-owner roles, and suspending agents when an owner leaves are **Planned**. DutyGraph has no auditor role of its own.
:::
