What does Ask in the Tares console need in order to run, and which model answers?

Ask in the Tares console needs a model provider on the instance, and it answers with the model of the instance’s default provider. You supply the provider with ANTHROPIC_API_KEY in the daemon’s environment, or by adding one under Settings, Model providers. The chat tells you which provider is answering. On Anthropic the model is TARES_AGENT_MODEL, which is claude-sonnet-4-6 unless you set it.

How it works

Ask is the assistant built into the console. It runs as a chat loop on the daemon, with up to 10 tool rounds per question, and each tool call shows in the chat as it happens. The loop needs a model to reason with, so the daemon has to reach a model provider.

The provider and model are chosen like this:

• Provider. Ask runs on the instance’s default provider. The chat says which provider that is, so you never have to guess.
• Model. The model is that provider’s default. On Anthropic it is the value of TARES_AGENT_MODEL, and the value is claude-sonnet-4-6 unless you set it.
• Shared providers. The same providers power Tares agents. A provider you add once serves both Ask and your agents.

Ask’s tools are thin calls to the daemon’s own read endpoints. It sees what the read API exposes and nothing more. Two proposal tools, propose_labels and propose_trigger, draft catalog changes as cards, and nothing changes until you choose Apply.

Set it up

Pick one of the two ways to give the instance a model provider.

1. Option A: set the key in the daemon’s environment. Export ANTHROPIC_API_KEY in the environment the Tares daemon starts from, then start Tares. The daemon reads the key when it starts.

tares up

2. Option B: add a provider in the console. Open Settings, then Model providers, and add one. Use this when you want a provider other than the environment key, or when you do not control the daemon’s environment.
3. Check the default. Ask uses the instance’s default provider. If you have more than one provider, make sure the one you want is the default.
4. Optional: change the Anthropic model. Set TARES_AGENT_MODEL in the daemon’s environment. If you leave it unset, Tares uses claude-sonnet-4-6.
5. Open Ask. Select Ask in the sidebar, or press ⌘K from anywhere in the console. The chat names the provider it is running on. Then ask something about your data:

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

If you are wiring up an SRE workflow around Tares, the AI SRE guide shows where Ask fits next to sources, triggers and agents.

What you get

• A working assistant with no external client. Once a provider is in place, you can check what is flowing, debug a source that looks off, or explore an instance you have just connected, all from the console.
• Visible reasoning. Each tool call appears as Ask works, so you can see how it reached its verdict.
• Answers grounded in your instance. Ask can list sources and connectors, describe catalog objects, profile a source’s fields, list entities, read recent events, list projects, and read a correlated timeline across every source or only one project’s.
• Debugging from live health. The same read tools answer questions such as whether anything is not ingesting, what looks sparse, and why a label is missing.
• Changes you control. Proposed labels and triggers arrive as cards you apply or skip. Applying goes through the same validated management APIs the console forms use. A trigger applied from Ask goes to the Default project.
• One provider setup for several features. The provider you configure for Ask also powers Tares agents.

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, with the same read surface plus subscriptions, connect it over MCP.
• If you want to set up a whole project from a goal instead of asking questions, use guided setup.

Related