Set up Automode with your own models

Automode routes every request to the model it needs, from models you choose. It's the routing behind Athena, aixplain's built-in router, with your own model on each tier. Use it to put your own models to work, keep requests on machines you control or balance cost and speed your way. This article explains what Automode is, how it routes, how to set it up in aixplain Desktop and in the aixplain OS, and what happens when something fails.

What is Automode?

Automode is the same routing as Athena, with the models you choose. You put a model behind each tier, and Automode sends every request to the tier it needs. The models can run on aixplain, on your own servers, on your laptop or on any machine that serves an OpenAI-compatible model.

Automode first shipped at aixplain in 2022, routing each speech recognition and translation input to the best system for it. Today it routes LLM requests.

Athena or Automode?

How Automode routes

For each request, an AI judge on aixplain reads an excerpt and sorts it into a tier. The request then goes to the model you set for that tier. There's no training on your data and nothing to tune.

Automode fails safe:

• Not sure: when the judge's answer isn't confident enough, the request goes to the most capable tier.
• Judge unavailable: when the judge can't be reached in time, the request goes to the most capable tier.
• Needs file changes: in aixplain Desktop, a request the judge calls conversation, but that needs files changed, goes to Routine work.

The model you picked always generates the answer. Only the excerpt the judge reads goes to aixplain.

Set up Automode in aixplain Desktop

Connect your models

Any model you can reach is a candidate for a tier: an aixplain model, a provider you connected or a machine that serves an OpenAI-compatible model, from a laptop to your own GPU servers.

To add your own endpoint:

1. Open Settings → Models & providers.
2. Under Providers, choose Custom provider.
3. Enter a Provider ID, a Display name and the endpoint's Base URL, plus its API key if it needs one.
4. Under Models, select Add model for each model it serves, with its ID and a Name.
5. Save. Its models now appear in the model lists.

Choose a model for each tier

1. Open Settings → Models & providers.
2. Find the Automode (Custom) section.
3. For Hard problems, Routine work and Conversation, choose a model.

Leave a tier unset to let aixplain choose one from the catalog. The tier then shows the model it chose.

A common setup puts a frontier model on Hard problems, a model on your own servers on Routine work and a fast, low-cost model on Conversation.

Turn on Automode

1. Open the model picker in Koder or Omni: click the model in the composer.
2. Choose Automode (Custom), the row under Auto.

Every request in the session now goes to the tier it needs. To stop, choose any other model.

Create your own router in the aixplain OS

In the aixplain OS, the Automode integration (aixplain/automode/aixplain) creates your own router as a model in your workspace, with its own ID. You can give it to an agent, or call it from your own code like any other model.

1. Open the Automode integration in the Marketplace and connect it. The form starts with three levels, Fast, Slim and Fat, from cheapest to most capable.
2. For each level, choose an LLM from the Marketplace and keep or edit its prompt, which describes the requests that belong on that level. You can add examples too.
3. Leave a level's model empty to skip it. A router needs at least two levels.
4. Save the connection. Your router now appears among your workspace's models.

The level prompts are used only to choose a level: the model it picks gets your agent's own instructions. When a model fails, for example with a timeout or an overload, the router tries the next level up.

Like other models you create, a new router is Personal: only you and your workspace's Owners and Admins can use it until you share it with your team. See Manage your workspace.

Pricing and usage

Automode has no price of its own. You pay for the models behind your tiers, from your aixplain credits, and models you run yourself cost aixplain nothing. For plans and credits, see Pricing.

Privacy and data

aixplain classifies each request before it goes to the model you picked, so Automode needs an aixplain sign-in. The judge sees an excerpt of the request, and that excerpt goes only to aixplain.

Limitations

• Automode in aixplain Desktop has no ID, so you can't call it from your own code, agents or MCP clients. For that, create a router in the aixplain OS, or use Athena.
• Automode needs an aixplain sign-in, because aixplain classifies each request, even when every tier runs on your own machines.
• Savings depend on your mix of requests. Work that is mostly hard reasoning saves the least.

Troubleshooting

My custom provider's models don't appear

Check the Base URL: it must serve the OpenAI chat completions format. Then check that the machine is reachable from your computer.

Every request goes to Hard problems

The judge may be unreachable or unsure. Check that you're signed in to aixplain and online, then try again.

Frequently asked questions

Get help

• Bring your own models: see Get started with Koder for providers and local models.
• Report a problem: email care@aixplain.com with your tier models and what you asked.