How can we help?

Start here

• What Agent isAgent gives you a team of AI workers led by your AI CEO. You give the CEO an outcome. It plans the work, hands parts to its team, and asks you before risky steps.
• Quick start: 10 minutes to your first taskSix short steps take you from sign-up to your CEO's first plan.
• Your first taskWhat to ask your CEO first, and what happens after you send it.
• How your team growsYou start with one worker. When work needs a specialist, the CEO proposes a hire with the role, the reason, the evidence and a daily budget. You approve or decline.
• Ava, your assistantAva is the assistant on every page. She answers questions about Agent and your organization, takes you to the right page, and proposes changes that run only when you confirm them.

Browse by topic

Open a topic to see every guide, or use search above.

• Getting startedWhat Agent is and your first hour with it.5 articles

  View all getting started guides

  • What Agent isAgent gives you a team of AI workers led by your AI CEO. You give the CEO an outcome. It plans the work, hands parts to its team, and asks you before risky steps.
  • Quick start: 10 minutes to your first taskSix short steps take you from sign-up to your CEO's first plan.
  • Your first taskWhat to ask your CEO first, and what happens after you send it.
  • How your team growsYou start with one worker. When work needs a specialist, the CEO proposes a hire with the role, the reason, the evidence and a daily budget. You approve or decline.
  • Ava, your assistantAva is the assistant on every page. She answers questions about Agent and your organization, takes you to the right page, and proposes changes that run only when you confirm them.
• SetupThe setup wizard, your business, machine and models.12 articles

  View all setup guides

  • The setup wizardThe setup wizard has seven pages. It prepares your CEO to start work. You can stop at any time and come back later.
  • Set up for me"Set up for me" does the first part of setup in one click. It reads your website, saves your brief, picks where work runs and applies the recommended settings. It stops where it needs you.
  • Setup: Your businessTell your CEO about the business. Start with your website, check each value, then save the brief and your brand.
  • The website importAgent reads your public website to suggest your brief and brand. It respects robots.txt, reads public pages only and saves nothing until you continue.
  • Branding and white-labelAgent can show your logo, colours and fonts to everyone in your organization instead of its own brand.
  • Setup: Your computerChoose where the work runs: on your own computer with your Claude and ChatGPT plans, or only in Agent's cloud.
  • Setup: AI accessYour CEO needs at least one model to think with. Use your own subscriptions, add an API key, or start on the included allowance.
  • Setup: ChannelsChoose how you talk to your CEO. Web chat is always on. WhatsApp is optional.
  • Setup: RepositoriesYour CEO needs at least one ready repository to work on. Connect GitHub or GitLab, or a folder on your computer. You choose each repository Agent may use.
  • Setup: Your CEOYou start with one worker, your CEO. Check how much it may do alone and how it may grow the team.
  • Recommended setupAgent recommends eight settings from what it knows about your business. Each item says why. You can change any value before you apply it.
  • Setup: First taskGive your CEO its first task. Ask for an outcome, not steps. The CEO replies with a plan, what it needs from you, and whether it wants to hire.
• Workers and the organizationHire, configure and level up your AI workers.10 articles

  View all workers and the organization guides

  • WorkersA worker is an AI employee with a name, a role, a level, a brain and a budget. The Workers page lists them and opens each one.
  • The CEOThe CEO is your chief executive worker. It is the front door to your organization: it takes your requests, plans them, hands parts to its team and reports back.
  • RolesA role gives a worker its instructions, capabilities and evaluation criteria. Built-in roles cover common jobs; an admin can add roles that extend them.
  • Levels and evaluationsEach worker has a level from L0 to L5, earned from its results. The evaluator proposes level changes and an owner or admin decides.
  • Hiring and org changesWorkers can propose hires and changes to the chart. The setting org.changes decides whether a person approves them, they apply at once, or only people edit the chart.
  • DelegationManagers split work into parts and hand each part to a worker below them, with acceptance criteria. Settings control how far and how wide delegation goes.
  • Worker settingsAn owner or admin edits a worker in sections: profile, responsibilities, instructions, brain, governance and identities. Each save can carry a reason.
  • OversightThe chief of staff watches spend, loops, failures, unusual behaviour and catastrophic attempts. It records findings with evidence and can pause a worker or department.
  • ResourcesResources are the repositories, services, integrations, credentials and budgets your workers use. Each one can have an owner who approves changes to it.
  • The virtual officeThe Office shows your whole organization as a 3D building with your brand. Employees sit in rooms, walk, talk and meet. You can fly around, drop into any room and talk to anyone.
• WorkGive work, follow it and review the results.12 articles

  View all work guides

  • Work itemsA work item is one piece of work with an owner, a state, a plan, checks and evidence. Each one has a key such as AGT-12.
  • The boardThe Board shows every work item by state, owner and priority. Filter it, save views and change many items at once.
  • Work templatesA template fills in the objective, what done means and who owns it. Agent has ten built-in templates for common office work.
  • Specs and the decision logA spec is a plan that people and agents edit together. It has typed blocks, notes on each block, an audit before building, tickets for each requirement and a trace to proof.
  • PR trialsA PR trial replays a pull request you already merged. Agent builds it again blind on your computer, then compares the result, checks, time and estimated cost with the original.
  • Evidence and checksWorkers attach proof to their work: tests and results, live calls, screenshots and waivers. Agent never shows a waived check as passed.
  • First-value trackingAgent follows the first piece of work you create, so setup can show whether it ran and how it ended.
  • EscalationsAn escalation is a decision a worker could not make alone. It comes with the options and the evidence the worker gathered. You answer, and the work goes on.
  • InboxThe Inbox holds threads, requests and mentions between people and workers.
  • IntakeIntake brings work in without a person: tracker polls, schedules, event subscriptions and dependencies between items.
  • CommandCommand is mission control: what your organization does now, what waits on you and what it costs, with controls to pause and resume workers.
  • ObserveObserve lets you zoom from the whole organization to one worker, its tools and delegations, and inside a delegation, live.
• Approvals and autonomyWhat workers may do alone and what waits for you.8 articles

  View all approvals and autonomy guides

  • ApprovalsActions that need a person wait in Approvals. An approver, admin or owner approves or denies each one, with an optional comment.
  • Autonomy modesThe autonomy mode decides how much workers do before they ask you. Auto is the default: internal work runs at once, and outside or risky actions wait for you.
  • The Always ask listSome actions always wait for a person, whatever the mode or risk tier. By default the list has seven items.
  • Risk tiersEach action has a risk tier: none, low, medium, high or critical. The tier and your autonomy mode decide whether a worker acts, acts and tells you, asks, or is stopped.
  • PoliciesA policy says what workers may do with a tool: allow, notify, require approval or deny. The most specific policy wins.
  • Typed decisionsSome small, repeated choices are typed decisions. The Decisions page shows which method answers each one and how its answers compare.
  • Budgets and spend limitsBudgets cap model spend per day for the organization and each worker, and per work item. Agent reserves spend before each call, so it never goes over by more than one call.
  • Kill switch and pauseThe kill switch stops every worker in your organization at once. You can also pause one worker. Release resumes the workers the switch paused.
• Models and runnersModel accounts, providers and the local runner.8 articles

  View all models and runners guides

  • Model accounts and API keysConnect your own API keys or cloud model accounts. Work you ask for uses your accounts first, then the organization's, then the platform's.
  • Brain and modelsChoose model defaults, connect providers, set fallback order, inspect activity and choose where work runs.
  • Who pays for workThe payer is the source that pays for a piece of work: automatic, the requester's own plan or keys, the organization's keys, or Agent.
  • Run modesA run mode is a preset for how work runs: Speed, Balanced, Quality, Cost or Custom. Hard gates hold in every mode.
  • The local runnerThe runner is a small program on your own computer. It starts your own Claude Code or Codex, under your own sign-in, for work that you asked for.
  • The runner's sandboxOn macOS, work on your runner runs in a sandbox. It can write only its own task folder and reach only the model vendor and allowed package registries.
  • Subscriptions and vendor termsYour Claude and ChatGPT plans work through the runner on your own computer. Your plan's terms still apply, so an admin accepts a caveat before runners are allowed.
  • Background runs on your own planAgent can start work on your own Claude routine or ChatGPT Workspace Agent while you are away. You save its trigger in Agent.
• Code and integrationsRepositories, GitHub and connected tools.6 articles

  View all code and integrations guides

  • GitHub and GitLabConnect GitHub or GitLab and choose the repositories Agent may use. Workers then read the code, change it and open pull requests or merge requests.
  • Local repositoriesConnect a Git folder on your computer through the runner. Work on it runs on your computer with your own git access. A folder needs a GitHub or GitLab remote before workers can change its code.
  • Jira and LinearConnect Jira or Linear so the CEO can read your tickets and plan from them. Intake can also bring new tickets in as work.
  • IntegrationsIntegrations connect workers to other tools. Each integration has a status and a risk label, and you choose which workers may use it.
  • Use Agent from Claude, ChatGPT or CodexAdd Agent to Claude, ChatGPT, Claude Code or Codex. The model runs there on your plan and uses Agent's tools to read and assign work.
  • Repository AtlasThe Atlas shows the stack, infrastructure, rules and architecture of each repository, measured from its files and Git history. No model writes these facts.
• ChannelsWhatsApp, email and other ways to talk to your CEO.3 articles

  View all channels guides

  • Web chat with the CEOThe CEO conversation is your main thread with your CEO. Give work, ask for status and make decisions. It is the same thread as WhatsApp.
  • WhatsAppConnect WhatsApp to give work, get short reports and approve with one tap from your phone. Setup runs through Meta, and Meta's messaging rules apply.
  • Reports and quiet hoursChoose what you hear when work completes, when the daily digest comes, and when the CEO may message you.
• KnowledgeMemory, documents, lessons and business data.8 articles

  View all knowledge guides

  • MemoryWorkers remember what they learn, with the source, confidence and freshness of each memory. You can tell the workforce a fact, verify it, archive it or forget it.
  • DocumentsDocuments are sources that workers search and cite. Paste a document, or connect a source such as Notion or Confluence.
  • Business dataBusiness data holds shared, typed tables for customers, clients, ideas and other records. Tables are the source of truth for structured facts.
  • LessonsWhen work finishes and people give feedback, workers propose lessons. An approver accepts or rejects each one before it is used.
  • Operating modelThe operating model holds the conventions Agent learned from your history, such as how you review and release. Nothing is used until a person confirms it.
  • Examples and proceduresApproved examples are outputs people approved, used as style references. Procedures are repeated tool sequences that a person approves before workers reuse them.
  • Rules and preferencesRules and preferences say how workers do the work. Each value shows where it comes from: you, your team, a project, the organization or the platform.
  • Your personaYour persona is how you work with agents and write code, learned from sources you choose. It is private to you and applies only after you accept it.
• Organization and accountMembers, roles, the org chart and settings.10 articles

  View all organization and account guides

  • Organization settingsSettings holds the organization name, the daily budget, the time zone, autonomy, the default brain, run mode, the kill switch, WhatsApp and every organization setting.
  • Members and rolesPeople in your organization have one of five roles: owner, admin, approver, member or viewer. Admins invite people by email.
  • API tokensAPI tokens are machine credentials for CI, scripts and MCP clients. Each token has one role and expires.
  • Sign in and sign upSign in with Google, GitHub, a code sent to your email, or a password. Sign-up asks for your name, your organization and your role.
  • Switch or add organizationsUse the organization menu in the top bar to switch between your organizations or create a new one.
  • Audit logThe audit log is the append-only, hash-chained record of everything that happened in your organization. An admin can verify that nobody changed it.
  • Public status pageA public view is a read-only status page that anyone with the link can open. It is off until an admin turns it on, and it shows totals only.
  • Enterprise identityEnterprise identity connects your identity provider with OIDC or SAML, provisions people with SCIM, and can require single sign-on for your email domain.
  • The Admin areaThe Admin area collects every administrator tool in one place, including provider benchmarks and harness evaluation.
  • Platform operator pagesPlatform setup, Product feedback and the Operator console are for the person who runs the Agent server, not for organization admins.
• Billing and limitsBudgets, spend, trial allowance and plans.2 articles

  View all billing and limits guides

  • The included allowanceThe included allowance is an amount of Agent's own model credits that a new organization may use. Your Agent server's operator sets it. It can be zero.
  • What costs whatModel calls cost money. Who pays depends on the payer: your own plan, your keys, the organization's keys or Agent. Costs in Agent are estimates at list prices.
• Security and privacySign-in, tokens, data handling and audit.4 articles

  View all security and privacy guides

  • Security overviewAgent limits what workers can do with roles, policies, approvals, budgets and the kill switch, checks every outgoing call for secrets, and records everything in a tamper-evident log.
  • What Agent storesAgent stores your organization's work, settings and knowledge. It never reads or stores your Claude or ChatGPT logins, and the runner sends facts about your folders, not their contents.
  • Secrets and credentialsAgent encrypts keys and tokens in a vault. It shows only a short hint, never the full value, and never sends a secret out.
  • Export your dataAn owner or admin can export the workspace: work, memory, documents, workflows and workers. Logins and credentials are never exported.
• TroubleshootingWhen something does not work as expected.7 articles

  View all troubleshooting guides

  • The runner is offlineYour computer shows Offline, or setup waits for your computer. Start the runner, check its sign-in, and check that runners are allowed.
  • The CEO does not answerThe CEO needs a model and budget to answer. When it does not answer, check its model, the allowance, the budget and whether it is active.
  • The website import failedAgent could not read your website. Nothing was saved. Try again, use another address, or write the brief yourself.
  • A repository is not readyA repository shows Not ready, or workers can plan but not change code. The repository says why. Here is what each reason means.
  • WhatsApp messages do not arriveThe CEO's messages do not reach your phone, or your messages get no answer. Check the setup checklist, your allowed number and the 24-hour window.
  • Out of budgetWhen a budget runs out, Agent refuses new spend. Work pauses with an escalation. You can raise the budget or stop the work, and you lose no progress.
  • Send feedbackTell the people who build Agent what you need or what went wrong. Use Send feedback, or ask Ava to send it for you.
• Not yet availableWhat Agent cannot do yet, and what to do instead.10 articles

  View all not yet available guides

  • Things Agent cannot do yetAn honest list of things you may expect from Agent that it cannot do yet, with what to do today.
  • ChatGPT plan use in Agent's cloudYou cannot sign in with ChatGPT inside Agent so that hosted work uses your ChatGPT plan. It waits on OpenAI's approval.
  • Claude plan use in Agent's cloudHosted work in Agent's cloud cannot use your Claude Pro or Max plan. Your Claude plan works through the runner on your own computer.
  • Code changes in a folder without a remoteA folder with no GitHub or GitLab remote is planning-only. Workers can read about it, but they cannot change its code. Add a remote to fix it.
  • Runner platformsThe runner works on macOS and Linux. Windows needs WSL. On Linux, work that needs shell commands is not taken, because the runner has no sandbox there.
  • Other chat channelsYou can talk to your CEO in web chat and on WhatsApp. iMessage and other chat apps are not available as a channel to the CEO yet.
  • Delegates on cloud model accountsClaude Code and Codex work can use a plain Anthropic or OpenAI API key, but not an Amazon Bedrock, Google Vertex or Azure OpenAI account.
  • Paid plans and checkoutThe Billing page shows plans and invoices only when the operator of your Agent server has set up billing. Until then, prices and checkout stay off.
  • Automatic model upgradesAgent does not test and switch to each new model by itself. You choose the models in Brain & Models and on each worker.
  • Importing chat historyAgent cannot move your past Claude or ChatGPT chats and projects in as work. Your persona can learn your style from exports you choose.