Your idea. A working change.

Start your first session

1. Create an account when sign-up is open, or use your organization’s invitation. Already a member? Sign in.
2. Open New session in the sidebar. Describe what should change and how the agent should verify it.
3. Choose a repository if you want the agent to change an existing codebase. Without one, it starts a project in its workspace.
4. Leave the mode on Auto or choose an effort level. Check the estimate and start the session.

If the app warns that an AI provider is unavailable, ask an organization admin to check Settings → Model providers. A session needs a working provider and available credits or configured billing allowances.

Turn a system design into a build

Open System design in the sidebar when you want to think through an architecture before writing code.

1. Describe the system: who uses it, the main jobs it must do, and any constraints. Add expected scale if you know it, or use a starting point.
2. Click Create design. The page updates as the design is generated. This uses Dev Units.
3. Review the architecture diagram and component list. Open Full document for requirements, data model, APIs, reliability, security, trade-offs and milestones.
4. Use Edit document for direct Markdown and Mermaid edits, or describe a change under Refine the blueprint. Revisions also use Dev Units.
5. Save your changes, select a repository and build mode, then click Build it. A new agent session receives your saved design and starts the first milestone with tests and a pull request as its goal.

A diagram that fails to render still shows its Mermaid source so you can correct it. Designs are shared within your organization; editing is limited to the author and organization administrators. Revising a built design lets you start a new build from the updated blueprint.

Designs are proposals, not guarantees. Check capacity, security and cost assumptions against your requirements before implementation.

Follow, steer and review

The session shows what the agent is doing. Read its plan, inspect the terminal and browser, and review the files and diff. Send a message to adjust the task while it works, or stop it before giving new instructions.

For work that needs closer supervision, use step mode and command approvals. Checkpoints can restore workspace files; review their scope before restoring. A passing test is evidence about the tested behavior, not a substitute for code review.

With a connected repository, the agent can commit changes and open a pull request. Review that pull request in your git provider before merging. SaaSWish does not need a customer command-line application.

Understand Dev Units

Dev Units (DU) cover model usage and running sandbox time. Sleeping sandboxes do not accumulate running-time charges. Choose an appropriate mode and start with a focused milestone to control work.

Admins manage prepaid credits, budgets and limits in Settings → Billing. Check the session estimate before starting and its itemized receipt afterward. Credits or configured limits can pause new agent turns.

See the pricing and rate card for standard rates and the service terms for refund and price-change provisions.

Connect your tools safely

• Repositories: connect your git provider in Settings → Repositories and enable the repositories your team needs.
• Secrets: use Settings → Secrets instead of pasting credentials into a task. Saved values are encrypted and supplied to the workspace as environment variables.
• Knowledge & playbooks: save team conventions and repeatable instructions in Settings, then reuse them across sessions.
• Access: invite members and manage organization roles in Settings. See security controls and privacy information before connecting sensitive resources.

More detailed feature documentation is available under Docs in the signed-in app. Open SaaSWish