Skip to Content
Tares

Can an AI SRE agent use OpenAI models instead of Claude?

Yes, a Tares AI SRE agent can run on OpenAI GPT models instead of Claude. The instance holds the model providers, and you choose a provider for each agent. One agent can run on GPT while another stays on Claude, and one provider is the default for everything else.

How it works

Which model an agent reasons with is a provider that the instance holds under Settings → Model providers. An instance can hold several providers at once. For the wider picture of how an AI SRE agent works in Tares, see AI SRE.

  • Provider kinds. Anthropic gives you Claude models. OpenAI gives you GPT models through the OpenAI API. An OpenAI-compatible endpoint covers anything that speaks OpenAI chat completions, such as LiteLLM, OpenRouter, Ollama, vLLM and Azure.
  • One default. One provider is the default. Ask and the guided setup run on it, and so does any agent that has no choice of its own.
  • Per agent choice. The agent form has a provider picker next to the model picker. Leave the provider on the cell default and the agent follows the default. Pick a provider and the agent is pinned to it.
  • Model choice. Leave the model empty and the agent runs on that provider’s default model. On Anthropic that is TARES_AGENT_MODEL, otherwise the first model the provider lists.
  • Model list. When you save a provider, Tares asks the endpoint which models it serves and stores the list, so the agent form offers what your key can use. If a provider lists nothing, the model field is free text.

Full details are on the model providers page.

Set it up

  1. Open Settings → Model providers and add an OpenAI provider. Supply an API key. Add a base URL only if you reach OpenAI through Azure or a proxy.
  2. If you prefer an OpenAI-compatible endpoint, add that kind instead. Give it a name, the base URL ending in /v1, and a key if the endpoint wants one. The form has presets for LiteLLM, OpenRouter and Ollama that fill in the URL.
  3. Check the model list on the provider’s row. Use Refresh models after an admin adds an alias or changes the key’s allowance.
  4. Open the agent you want to move to GPT. Set its provider to OpenAI and pick a model, or leave the model empty to use the provider’s default.
  5. Leave your other agents on the cell default, or pin them to Anthropic, so they stay on Claude.
  6. Optionally make OpenAI the instance default, so agents without a choice of their own follow it.

A deployment can also seed providers from the environment. The console values win over the environment.

ANTHROPIC_API_KEY=...            # the Anthropic entry (ANTHROPIC_AUTH_TOKEN, ANTHROPIC_BASE_URL as before)
OPENAI_API_KEY=...               # an OpenAI entry
OPENAI_BASE_URL=...              # its base URL, for Azure or a proxy
TARES_MODEL_PROVIDER=openai      # the default, when none is chosen in the console
TARES_AGENT_MODEL=...            # the default model on the default provider

In catalog YAML, an agent names its choice with the fields provider and model. The provider is the id from Settings: anthropic, openai, or the slug of an endpoint’s name. If an agent from a template or catalog import names a provider your instance does not hold, it runs on the default, and the agent page tells you until you pick a provider that exists.

What you get

  • Mixed models on one instance. One agent runs on GPT while another stays on Claude, and a third can run on a model your own router serves.
  • A readable history. Every run records the provider and model it ran on, on the agent page and in the runs table. The history stays clear after a provider is removed.
  • Cost per model. The spend meter on the Overview prices each run from the model that answered. Anthropic and OpenAI models are priced from Tares’s own price table. A router that reports its own cost, such as OpenRouter or LiteLLM with cost tracking on, wins over the table. A model with no price records its tokens with no dollar figure, and the meter counts the unpriced calls instead of showing zero.
  • Spend by provider. Once more than one provider has run, the Overview splits spend by provider.
  • No rework for existing Claude setups. An instance configured through ANTHROPIC_API_KEY or the older Model access tab shows up as a configured Anthropic provider with nothing to redo.

When something else fits

  • Small or local models. Connecting a small open model is one provider entry, but the agent loop hands the model tools over several rounds. A small model may write a tool call as JSON text, mis-case a tool name, or answer with nothing. Tares runs a call written as text, matches a mis-cased name, and marks a run that ends with no conclusion and no tool call. An 8B-class model on a laptop is good for trying things out, and it can invent events on an empty result. For root-cause work on real data, use a hosted model or a 30B-class one and above.
  • A LiteLLM in front of Tares. You can put it on the Anthropic entry as an LLM gateway, where Tares sends Claude model ids and the gateway maps them. Once an instance mixes models, adding it as an OpenAI-compatible provider is the natural choice. The picker then offers exactly the aliases the key may use, and LiteLLM’s cost figure lands on each run.
  • Staying on Claude. If every agent should keep using Claude, leave the Anthropic provider as the default. You do not need to change anything.

Still stuck? Get help

Last updated on
Quickstart