Can I ask Tares what happened during failed deploys straight from the console?

Yes, Tares lets you ask what happened during failed deploys from the console with Ask, as long as the instance has a model provider. Ask needs either ANTHROPIC_API_KEY in the daemon’s environment or a provider added under Settings, Model providers. It runs on the instance’s default provider, and the chat tells you which one that is. Open it from Ask in the sidebar or press ⌘K anywhere.

How it works

Ask is the assistant built into the Tares console. It runs as a chat loop on the daemon, with up to 10 tool rounds per question, and shows every tool call as it works, so you can follow how it reaches its answer.

Its tools are thin calls to the daemon’s own read endpoints. For a failed deploy, that means it can:

• list your sources and connectors, and check what is flowing
• describe a catalog object, with its schema, entities, freshness and samples
• profile a source’s fields and list its entities
• read recent events
• read a correlated timeline, across every source or only one project’s

Ask sees what the read API exposes and nothing more. If your deploy events already reach Tares, for example through a CI webhook, a GitHub connection or a Vercel log drain, you can ask about them in plain language. If nothing records the deploy yet, Ask can tell you which source looks quiet or missing, because the same tools answer questions like “is anything not ingesting?” and “why is this label missing?” from live health and field coverage.

The model is the default provider’s model. On Anthropic that is the value of TARES_AGENT_MODEL, which is claude-sonnet-4-6 unless you set it.

Set it up

1. Start Tares if it is not running yet.

tares up

2. Give the instance a model provider. Either set ANTHROPIC_API_KEY in the daemon’s environment, or open the console and add one under Settings, Model providers.
3. Open Ask in the sidebar, or press ⌘K from any page. The chat names the default provider it runs on.
4. Type your question. Name the service, the environment or the time window you care about. For example:

Which deploys failed in the last 24 hours, and what else changed around each one?

5. Watch the tool calls as Ask works. Open a call to see what it read, and follow up in the same chat to narrow the window or look at one service.
6. If Ask proposes a label or a trigger, it appears as a card. Nothing changes until you choose Apply, and skipping it leaves your catalog as it was.

For the full list of provider kinds, see the Ask page, which also covers the tools and what Ask needs.

What you get

• A written answer from inside the console, with no external client to wire up first.
• A visible trail of tool calls, so you can check each claim against the events it read.
• A correlated timeline view when you ask what happened around a failed deploy, spanning every source or only one project.
• A quick check on the data itself: if the answer looks thin, Ask can show which source is stale or which label is missing.
• Optional proposal cards for labels and triggers, so a deploy failure you just diagnosed can become a trigger you apply with one click. A trigger applied from Ask goes to the Default project.

The same read surface also backs the rest of the AI SRE workflow.

When something else fits

• If you want an agent in your own workflow, such as Claude Code, Cursor or your own service, connect it over MCP. It gets the same read surface plus subscriptions, instead of living inside the console.
• If you want to set up a whole project from a goal rather than ask questions, use the guided setup.
• If the instance has no model provider yet, add one first. Ask does not run without it.

Related