Which models does Eidolon support and how do you get started?

Eidolon is an open-source, enterprise-grade AI Agent Server that lets you build and deploy genAI applications. Based on its documentation, you can define agents that reference multiple models — the sample configuration lists GPT4-turbo, MistralLarge, Llama3-8b, ClaudeOpus, and ClaudeSonnet — and mark one as the default. You get started by building agents (using declarative YAML or your own code), deploying them on the agent server, and consuming them. The source code is available on GitHub.

Which models Eidolon supports

Eidolon's own example configuration shows a multi-model chatbot with these model entries:

A few things to keep in mind about this list:

• It comes from a single sample agent definition, so it reflects what that example wires up rather than a definitive, exhaustive compatibility list.
• The models are declared as apus entries inside an agent's spec, each with a title and an optional default flag. That means model selection is a configuration choice you make per agent, not a fixed platform setting.
• Because agents are defined declaratively, you can add or swap model entries in the YAML to match what you want that agent to offer.

How to get started

The site describes bringing agents into production in three steps: Build, Deploy, Consume.

1. Build

You have two paths:

• Use a pre-built agent. Eidolon provides a variety of pre-built agents to get you started.
• Define your own. You can write agents in vanilla code or use another agent framework, or define them with Eidolon's declarative YAML.

A minimal agent can be as short as a single line; more complex behavior is added as needed. The YAML uses an apiVersion of server.eidolonai.com/v1alpha1 and a kind of Agent, with metadata.name and a spec.

For example, a multi-model chatbot agent looks like this in the documentation:

apiVersion: server.eidolonai.com/v1alpha1
kind: Agent
metadata:
  name: multi-model-chatbot
spec:
  system_prompt: You are a helpful assistant. Output all multi-media content as embedded content including image, video, and audio. Only generate images or speech if directly asked by the user.
  apus:
    - { apu: GPT4-turbo, title: GPT4 Turbo, default: true }
    - { apu: MistralLarge, title: Mistral Large }
    - { apu: Llama3-8b, title: Llama3 8b }
    - { apu: ClaudeOpus, title: Claude Opus }
    - { apu: ClaudeSonnet, title: Claude Sonnet }

Agents can also work together. The docs show an agent that references another agent via agent_refs, and a retriever agent that searches a GitHub repository:

apiVersion: server.eidolonai.com/v1alpha1
kind: Agent
metadata:
  name: agent
spec:
  description: An agent that searches the eidolon code and documentation
  agent_refs: ["doc-search"]
  system_prompt: You are a software expert responsible for answering questions about a code base.
---
apiVersion: server.eidolonai.com/v1alpha1
kind: Agent
metadata:
  name: doc-search
spec:
  implementation: RetrieverAgent
  description: "Search the Eidolon github repository for code, examples, and documentation"
  document_manager:
    loader:
      implementation: GitHubLoader
      owner: "eidolon-ai"
      repo: "eidolon"
      pattern: ["**/*.md", "**/*.py"]
      exclude: "**/test/**/*"

This shows two capabilities worth noting for getting started:

• Agent-to-agent communication through agent_refs.
• RAG storage added to any application using simple services, with the flexibility to customize RAG to your needs.

2. Deploy

Eidolon is described as a secure, enterprise-ready deployment server for agentic applications — the "Agent Server" half of the product. The SDK is the pluggable framework developers use to build the agents that run on it.

3. Consume

Once deployed, agents are consumed as services. The documentation frames the whole flow as Build → Deploy → Consume, so the consumption step is where your application or users interact with the running agents.

Practical starting checklist

1. Go to the GitHub repository (eidolon-ai/eidolon) to get the open-source code — the site links to "Get Started" and "Star on GitHub."
2. Pick a starting point: use a pre-built agent or write your own YAML/code.
3. Define your agent's system_prompt and, if you want multiple models, list them under apus with one set as default: true.
4. Add capabilities as needed — agent_refs for agent-to-agent communication, or a retriever agent with a GitHubLoader for RAG over a codebase or docs.
5. Deploy to the agent server and consume the agent.

What the source does not specify

The available page evidence does not state pricing, licensing terms beyond "open source," supported model versions beyond the names listed, or exact installation commands. For those details, check the GitHub repository and the project's own getting-started documentation rather than assuming defaults.