How do I set up an AI SRE for slow database queries with guided setup instead of building it by hand?

Tares sets up an AI SRE for slow database queries from a one-sentence goal, with a plan you can edit before anything runs. The project exists as a draft from the moment planning starts. If you leave, Finish setting up on the Projects list reopens the draft where you left it. This is part of the AI SRE workflow in Tares.

How it works

Guided setup is on the Projects page under Create new. You describe the outcome the way you would to a colleague, for example that you want to know when database queries get slow and why. Tares then writes a plan covering what to watch, when to wake, who does the work and what they may use. The full guided setup page lists every control.

The flow has four steps, shown at the top of the page:

• Goal: you write what the project should achieve and pick who does the work. Tares agents run inside Tares on your model provider. Your own agent, such as Claude Code or a service of yours, joins with a project key.
• Plan: Tares reads what is already on your Tares, looks at what a source sends, writes the plan, checks it and fixes what the check found. It usually takes under a minute.
• Connect: Tares lists what only you can connect.
• Try it: one practice run shows a result before a real one arrives.

The draft starts the moment planning starts. You can close the tab while Tares plans or while you edit, and the draft waits on the Projects list. Nothing of a draft is set up yet, so deleting it with Delete draft removes nothing else.

Set it up

1. Make sure your instance has a model provider. Guided setup uses the same one as Ask. If none is configured, the Goal step links to Add a model provider. On a local machine, start Tares first:

tares up

2. Open Projects, click Create new, and answer What should this project achieve? in a sentence. You can also click one of the examples under Or start from one of these.
3. Pick Who does the work? and click Plan it. The project now exists as a draft.
4. Read the four plan cards: Watches, Wakes when, Agents and Know-how. Each item has Change and Remove. Under the cards, Tools the agents can use lists the MCP servers in the plan.
5. Edit before anything is set up. For a source, start with the kind of source, then pick one of yours or choose Connect a new one. In Wakes when, change the numbers in place, including how long to wait before waking again. To change the plan in words, type in Say what to change, for example “only for payments-api”, and click Update the plan.
6. Open Show what Tares sets up for you to see every part the plan creates or uses. When it reads right, click Looks right, set it up. Tares creates the project and everything it needs in one go.
7. In Connect, finish what only you can do. That can be the endpoint your system sends events to, with Send a test event for me, a credential for a new source, or your own agent’s key, which is shown once. Click Continue, or I will connect it later to leave it for the project page.
8. In Try it, click Run it once now. The agent that looks first runs once on the latest event its wake-up would count. The result is marked Practice and is left out of today’s totals. Then click Open the project.

If you leave midway, go to the Projects list and click Finish setting up on the draft’s row. It opens where you stopped.

What you get

• A normal project, with every part on its own page and editable afterwards. Guided setup only saves you the assembly.
• A plan written in plain words that you review before anything runs.
• Reuse of what you already have. An existing source, agent, tool or skill is used as it is and stays in the projects that already use it.
• Safe interruptions. The draft is kept, you resume with Finish setting up, and nothing is half created.
• A practice result before a real event arrives. It uses your model provider like a real run.
• Agents that follow optional skills, the notes you add or upload as SKILL.md, or pick from ones already on Tares.
• A clear note in the plan when no installed connector fits part of your goal, instead of a forced poor match.

When something else fits

• If you would rather choose the parts yourself, the first step links to build it by hand from existing objects.
• If a ready-made shape is close to what you want, the first step also links to Start from a template.
• If the plan says no connector fits part of your goal, the note tells you so. You can then connect that part yourself, or write to help@glassflow.ai with what you would like to connect next.

Related