engine

Configures which inference provider the agent loop uses. Without this field the runtime falls back to whatever engine is set in the environment.

import { Axon, Cerebras, Codex, Ollama, OpenRouter } from "@axon/engines"

export default defineAgent({
    engine: Codex(),
})

Managed engine

Axon() routes inference through an Axon-managed Cognos engine. Pass a model to pin a specific catalog entry, or omit it to use the account default.

import { Axon } from "@axon/engines"

export default defineAgent({
    engine: Axon({ model: "axon-astra-v1" }),
})

This is the default when you deploy with axon deploy. No key management — the platform handles it.

Local engines

Local engines are functions imported from @axon/engines. The function receives the message history and returns a string. Cognos calls it via the engine bridge — it never has direct access to the inference provider.

import { Ollama } from "@axon/engines"

export default defineAgent({
    engine: Ollama({ model: "llama3.2" }),
})

Available engines:

engine is not a string

engine: "gpt-4o" is not valid. Engine is always an adapter function call such as Axon(), Cerebras(...), Codex(), Ollama(...), OpenRouter(...), or Mock(). The string form does not exist.