Quickstart

The fastest way to understand d5s is to complete one real task in a chat. You do not need to configure a project, agent, or automation first.

A new Free organization can run GPT-5.6 Luna through the Vercel AI Gateway immediately. Its Models page keeps the rest of the catalog visible but unavailable. Paid organizations can enable the other Gateway models, and a paid EU cloud organization also gets the Claude models served through Amazon Bedrock in the EU by default. There is no clearance step and no route selection. A workspace administrator enables models that the organization's plan allows, and d5s resolves each enabled model to one exact route before dispatch. If the composer reports that no model route is available, ask a workspace Owner or Admin to check the model policy, then contact d5s if an expected model is missing.

1. Start from the workspace home

New personal accounts begin with a short welcome tour at Onboarding. The five steps introduce AI coworkers through an animated room tour, show artifacts, communication channels, and computer work, and let you confirm your workspace name and available region. In the room tour you can explore different departments or skip ahead to setup. The final slide lets you continue on Free or view plans. Completing setup opens your workspace chat; it does not create an agent automatically.

If you joined an existing workspace by invitation, open that workspace and start in chat. You do not need to create a personal workspace to join your team.

Otherwise, open your workspace and start a new chat. Describe the result you want, the relevant background, and any constraints. A useful opening request names the deliverable and the decision it should support, for example:

Review the attached customer notes, group the recurring objections, and create a short brief with the five issues we should address first. Show your evidence for each group.

If the work already belongs to a listed project, choose that project before sending. Otherwise, start directly: d5s creates an unlisted project behind the chat so files, memory, and artifacts still have a durable owner.

2. Add the right context

Attach files that are specific to this conversation. Put instructions or reference material that should affect future work in the project instead. This distinction prevents a one-off attachment from silently becoming standing context.

The composer also lets you choose the model, reasoning level, available tools, and approval mode. Connectors already configured in the workspace are available through the chat's project by default; turn off an unneeded connector from that project's connector settings. Keep the default Ask for approval mode when you want connector actions configured to Ask to pause. Choose Allow all only when you are comfortable letting those actions continue for the whole session without individual prompts. Explicit Deny rules and connector or network boundaries still apply.

3. Review the result

Follow progress in the conversation. A run can show tool calls, generated files, sub-agent work, research progress, and connector approval requests. In Ask mode, approve a requested connector action once (valid for 15 minutes and consumed by one call) or for the session (valid for up to 24 hours) only when the requested action is appropriate. Switching the conversation to Allow all does not create one of these durable tool approvals; it applies only while that session remains in Allow all mode.

Continue in the same chat while the goal is changing. If a second message arrives while work is active, d5s records it as another turn and either joins it to the current work cycle or queues it for the session; the UI shows the applicable state.

4. Keep what should last

Use Save as project when the chat has accumulated context you want to find and reuse. This promotes the existing project; it does not copy the chat or move its files into a new container.

Then choose a work mode: keep the conversation as a chat, staff an agent for an ongoing responsibility, create an automation for a defined job and trigger, or create a dashboard when the maintained output matters more than the transcript.