Can I ask Tares what happened during 5xx errors on an API straight from the console?

Yes, Tares has Ask, an assistant built into the console, and you can type a question about the 5xx errors on your API and get an investigation back. Ask reads your sources, entities, events and correlated timelines and shows each tool call as it works. It needs a model provider on the instance, either ANTHROPIC_API_KEY in the daemon’s environment or one added under Settings, Model providers. It runs on the instance’s default provider, and the chat tells you which one.

How it works

Ask runs as a chat loop on the 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 read API exposes and nothing more.

For a question about 5xx errors on an API, 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 and list its entities.
• Read recent events.
• Read a correlated timeline, across every source or only one project’s.

Each tool call appears in the chat as it happens, so you can follow how Ask reaches its verdict. The same read tools also answer debugging questions such as “is anything not ingesting?” and “why is this label missing?” from live health and field coverage. That is useful when the API’s data looks thin during the incident.

Ask runs on the instance’s default model provider. On Anthropic the model is that provider’s default, which is TARES_AGENT_MODEL, set to claude-sonnet-4-6 unless you change it. The same providers power Tares agents.

Set it up

1. Start Tares if it is not running yet. This brings up the daemon and the console.

tares up

2. Give the instance a model provider. Use one of these two routes:
   • Put ANTHROPIC_API_KEY in the daemon’s environment before the daemon starts.
   • Open the console, go to Settings, then Model providers, and add a provider there.
3. Make sure the data about your API is flowing into Tares from the sources you have connected. Ask can only read what the instance holds. If you are not sure, ask Ask whether anything is not ingesting.
4. Open Ask in the sidebar, or press ⌘K from anywhere in the console.
5. Check the chat, which says which provider it runs on. If you have several providers, Ask uses the instance’s default one.
6. Type your question, for example:

What happened to the API during the 5xx errors, and which service was involved?

7. Read the tool calls as they stream in, then read the verdict. Follow up in the same chat to narrow the time window or look at a single service.

For a broader view of how Tares watches systems and helps with incidents, see AI SRE.

What you get

• An answer from the console, with no external client to wire up first.
• A visible investigation: every tool call is shown, so you see which sources, entities and events Ask looked at.
• A correlated timeline read across every source, or only a project’s, so the API’s errors are seen next to what else happened around them.
• Debugging help in the same chat when a source looks off, a label is missing or a field is sparse.
• Optional proposals. Ask can draft label changes or a trigger as cards you apply or skip. A proposed trigger names its sources and, when you apply it from Ask, goes to the Default project. The proposals change nothing until you apply them, and Apply goes through the same validated management APIs the console forms use. This is handy if you want the next 5xx spike to raise an alert.

When something else fits

• If you want the same read surface inside your own workflow, in Claude Code, Cursor or your own service, connect that agent over MCP instead of using the console chat. That route also adds subscriptions.
• If you want to set up a whole project from a goal rather than ask questions, use guided setup.
• Ask reads one instance. If the instance has no model provider, Ask has nothing to run on, so add one first.

Related