Can I ask Tares what happened during error spikes straight from the console?

Yes, Tares has Ask, an assistant built into the console where you type a question such as which service had the most errors in the last 24 hours and what happened to it. Ask investigates with the daemon’s read tools and shows each tool call as it works. It needs a model provider on the instance and runs on the instance’s default provider, and the chat tells you which one it is using.

How it works

Ask runs as a chat loop on the Tares daemon, with up to 10 tool rounds per question. Its tools are thin calls to the daemon’s own read endpoints, so it sees what the API exposes and nothing more.

When you ask about an error spike, Ask can use these read tools:

• List sources and connectors, and describe any catalog object, including its schema, entities, freshness and samples.
• Profile a source’s fields.
• List entities and read recent events.
• List projects.
• Use read to get a correlated timeline, across every source or only a project’s.

You see each call in the chat as it happens, then the verdict Ask reaches. That lets you check the evidence behind an answer instead of taking it on trust.

The same read tools also answer debugging questions from live health and field coverage, such as whether anything is not ingesting, what looks sparse, or why a label is missing. If a spike looks odd because a source stopped sending data, you can ask that in the same chat.

Ask also has two proposal tools, propose_labels and propose_trigger. They draft catalog changes as cards that you apply or skip. A proposed trigger names its sources and, when you apply it from Ask, goes to the Default project. The proposals themselves change nothing. Apply goes through the same validated management APIs the console forms use. After you understand a spike, you can ask Ask to propose a trigger for it.

Set it up

1. Make sure the instance has a model provider. Either set ANTHROPIC_API_KEY in the daemon’s environment, or add a provider in the console under Settings → Model providers. If you run Tares locally, start it from the environment that holds the key.

tares up

2. Open the console and choose Ask in the sidebar. You can also press ⌘K from anywhere in the console.
3. Look at the chat to see which provider Ask is running on. Ask uses the instance’s default provider, and the chat names it.
4. Type your question about the spike. Name a time range and what you want to know, for example:

Which service had the most errors in the last 24 hours, and what happened to it?

5. Watch the tool calls as Ask works. It reads sources, fields, events and a correlated timeline, then gives its verdict.
6. Follow up in the same chat. Ask which source a burst came from, what else happened around the same time, or whether a source stopped ingesting.

On Anthropic, the model is that provider’s default. TARES_AGENT_MODEL sets it, and it is claude-sonnet-4-6 unless you change it. The same providers also power Tares agents, so one setup covers both.

The empty state of Ask offers starter questions for exploring data, debugging a source and organizing it. They are a quick way to see what Ask can do before you write your own question.

What you get

• A answer to what happened during a spike without leaving the console and without wiring up an external client first.
• A visible trail of tool calls, so you can see which sources, fields and events the verdict rests on.
• A correlated timeline view across all sources or only one project’s, so errors can be read next to other events around the same time.
• Debugging help in the same place: ingestion gaps, sparse fields and missing labels come from live health and field coverage.
• Draft labels and triggers as cards. You decide whether to apply or skip each one, and nothing changes until you apply it.
• Findings that Tares agents leave on the timeline sit alongside the events Ask reads. For the wider picture of how Tares helps you investigate incidents, see AI SRE.

For more on the console assistant itself, see Ask.

When something else fits

• Ask lives inside the console and reads this one instance. 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.
• If you want to set up a whole project from a goal rather than ask questions, use the guided setup in the console instead of Ask.

Related