Create your first agent

Create an agent when a named, persistent role should own work across multiple check-ins. The agent receives a home project and one continuing conversation, so configure it as an ongoing teammate rather than a scheduled copy of a prompt.

Start with a responsibility

When the workspace has no agents yet, Agents starts with the same responsibility-first path as New chat. Select Describe the work, explain what should come off your plate, and send it. d5s creates a named coworker and uses that responsibility as its first assignment in the coworker's continuing conversation.

The first-login welcome tour opens workspace chat after setup. To create a persistent coworker, describe its responsibility here or configure a profile manually using the steps below. Review the instructions, connectors, and schedule before enabling autonomous work.

1. Create the profile

From New chat, open the agent selector beside the composer and choose Add an agent. Quick create opens a compact form for the agent's name, persona, and standing instructions, then selects the new agent as the composer target. Agent Builder opens the guided setup workspace. Preset or template opens the reviewed catalogue and starts a draft from your selection so you can review it before activation.

You can also open Agents and select New agent. Search the reviewed catalogue or narrow it with the canonical Team categories. Opening a preset shows its deliverable, steps, recommended connections, run mode, and human approval boundary; browsing or previewing does not create anything. Select Use this template to open a reviewable draft based on that exact preset version, then review it before activation.

A preset copies its trusted profile and working instructions, not authority. Starting from a preset does not connect an account, inherit workspace credentials, configure a channel, create a schedule, or run a task. Recommended connections and timing remain guidance that you can configure later.

For a role that is not in the catalogue, choose Agent Builder and describe it in your own words. This opens a private configuration workspace. On a wide desktop, the editable configuration appears beside a Setup assistant; on a tablet or phone, use Configure, Assistant, and Preview to move between the same views. Describe the work, result, preferences, and boundaries in the assistant, or edit the seeded configuration directly. Setup replies use your workspace model and credits.

The setup assistant updates the draft but is not the future agent. It cannot run tools, execute work, select accounts, or activate itself. Review and edit Instructions, Capabilities and knowledge, Triggers, and Settings directly. Preview shows the configuration the future agent will receive. Form edits are backed up temporarily in this browser profile, so reopening the same draft there can recover fields whose saved version has not changed. This local recovery is best-effort, not a durable server save, and not available in another browser. Select Save draft, send another assistant message, or activate the agent to persist the exact visible configuration. Drafts and setup history are private to their creator in the workspace.

The setup conversation explains your communication options before asking about connectors. Where draft identity setup is enabled, your agent gets a private, inactive identity during the conversation. When managed email is available, its real inbox address appears with Enabled and Copy address. You can configure Slack, Telegram, or feature-gated Discord inside setup. A connected channel does not activate the agent or authorize sending messages. In workspaces without draft identity setup, channel preferences are saved and setup opens after creation. You can skip extra channels and set them up later in Manage agent → Channels.

In guided preset setup, Connections starts with the preset's recommended products and uses the same searchable rows as the workspace connector directory. Select a row to attach an existing workspace account or connect a new one directly; when reusable accounts exist, d5s offers those first. If several accounts are available, check the account label before choosing one. New connections require a workspace administrator, while members can attach an account that an administrator has already authorized. You can continue without a connection and add it later.

In the full Builder, Capabilities and knowledge keeps capability selection separate from authorization. Selecting a connector records the intended capability; it does not connect the app or grant access. Choose an existing workspace connection explicitly, connect a new account, or choose Set up later. After selecting an account in either setup path, describe directly on its connection card how the agent may use it before activation; the setup assistant can also fill this guidance from your conversation. You can instead choose Set up later and add the connection after activation.

Connector setup preserves the draft across sign-in redirects. Slack and mailbox authorization return to the draft and reopen the relevant channel settings. Cancellation or failure leaves the draft available to resume. A connection successfully added to the workspace remains there even if you do not finish creating the agent.

Review the visible draft, then select Activate agent (or Create agent where draft identity setup is unavailable). Activation keeps the same identity, inbox address and configured channels, applies the reviewed instructions and connections, and enables agent work. A draft cannot run tasks or reply through its channels. Setup retries return the same agent. Find your private setup conversation under Agent settings → Setup history, including when returning to the agent later.

Conversationally created agents use the connections explicitly reviewed during setup. They do not automatically receive other existing or future workspace connections. You can change their access later in Manage agent → Connectors. Instructions about how to use an app are working guidance, not additional provider permission boundaries.

This setup creates an agent that works when you ask. Timing requests and requested trial tasks are saved as deferred: no schedule is configured and no task has run. Configure those separately after creation.

2. Write standing instructions

Open the agent, select its avatar at the right end of the page header to open Manage agent. The Profile section holds Persona and Standing instructions below the name and model. Standing instructions should describe the responsibility that remains true across every turn:

The model starts on Workspace default. If a workspace administrator has enabled a shared Codex connection, that means the same Codex model and workspace-owned connection for every member; it is not a personal connection. An Owner or Admin can explicitly pin this agent to another enabled Codex or catalog model. Select Workspace default again to clear that pin and resume inheritance.

• the outcome the agent owns;
• sources it should check;
• decisions it may make;
• actions that require approval;
• the expected output or destination; and
• what to do when there is nothing meaningful to report.

Avoid placing today's task or a one-off deadline in standing instructions. Send that as a direct message, or set up a task-specific scheduled wakeup for it — either by asking the agent or by adding one yourself under Manage agent → Autonomy.

3. Add durable context and capabilities

Use the Manage sections to configure the agent's home project:

• Files for reference material that should remain available;
• Connectors to review selected or inherited workspace credentials and remove any this agent should not use; and
• Skills for reusable methods or instructions.

Workspace connectors are available by default to agents made through advanced creation. Agents made through conversational creation start with only their reviewed connections and retain that explicit access policy. Removing one affects only that agent. MCP tool policy still controls whether individual operations allow, ask, or deny; API operation controls apply only when they are available and enabled for that credential.

For independent recurring checks, open Automations below Memory on the agent page, then open a routine and select Manage team, then Add people or agents…, and add the agent. Keep informed shares short result handovers when the agent next works. Enable Act on findings only when actionable results should wake it. The automation keeps its own run context; attaching an agent does not grant access or copy its conversation into the run.

4. Test in the conversation

Keep autonomy off and send a representative request directly to the agent. Review the sources it selected, tool calls it made, approval requests it raised, and whether the result followed the standing instructions.

Test at least three conditions: a normal case, incomplete evidence, and a case where there is nothing to do. Correct the instructions or capability set and repeat until all three behave predictably.

5. Add autonomy last

Open Manage agent → Autonomy only after direct tests pass. You can create, edit, pause, and remove one-time or recurring scheduled wakeups, and configure the daily budget and agent pause state.

Start with a conservative cadence and a finite daily budget. A broad Routine check-in wakeup can ask the agent to inspect its project and decide whether work is due, while named wakeups should carry one concrete responsibility. Every wakeup still invokes the configured model. Pausing the agent blocks work without deleting its conversation or configuration.

Read Agent wakeups and schedules before enabling the first autonomous check-in.

New agents created by signed-in users with a verified email automatically get a d5s inbox where available. Incoming and outgoing permissions default to the organization's inferred business domain or the creator's exact personal-provider address, and sending requires approval. Review these rules under Manage channels → Email before relying on email for work. Existing agents can choose Create d5s address, and any agent can connect a dedicated Gmail mailbox. The optional Accept replies from people this mailbox emails exception starts off for automatic inboxes and on for manual mailbox setup. Review outgoing rule changes carefully: saving them can authorize matching sends without individual approval. Read Agent email for details.

When Telegram is enabled for your workspace, an authorized Owner or Admin can select Add Telegram in the conversation’s Channels panel or open Manage agent → Channels → Telegram. Follow the illustrated guide to create a bot in BotFather, connect its token, link your account and test a reply from Telegram. Bot connected confirms the bot connection only; see Telegram teammate DMs.

For Telegram, you can connect the bot while your agent is a draft. Activate the agent before linking your personal Telegram account or trying a message; return to Telegram in channel settings for those remaining steps.

When Discord is enabled for your workspace, an authorized Owner or Admin can begin setup while the agent is a draft or open Manage agent → Channels → Discord later. Create a dedicated application and bot in Discord, connect its Application ID, Public Key, and bot token, configure the generated interactions endpoint, and bind only the server channels this agent should hear. Teammates confirm their own d5s account before a Discord DM can enter the agent's conversation. See Discord agent channel for the complete setup and test boundary.

Where dedicated Teams identities are available, open Manage agent → Channels → Microsoft Teams after creating the agent. An organization Owner or Admin with workspace administration access connects provisioning once, then creates the named identity and continues through Microsoft installation. Activate the intended verified standard channels separately. See Teams agent setup for permissions, testing, and supported scope.

Optional local file access

To let this agent work with local files, add folders in Settings → Desktop, then grant Read only, Write only, or Both in Manage agent → Local files. Confirm access in the native dialog and ask for the operation in a Desktop conversation. See Local files.