Models & agents

Model versus agent

A model performs inference: it interprets input and generates output. An agent is a configured actor that uses a model with session context, available tools and a workflow. A tool call is not a property of the model alone; Aleph decides which capability is exposed and where that operation executes.

The Workshop composes these pieces. Specialized workspaces provide a domain-specific starting environment. Humans remain responsible for reviewing consequential outputs and authorizing connected or privileged actions where appropriate.

Provider layer

Aleph is designed to remain useful when the selected model changes. The product direction includes a bundled local Qwen 8B path and bring-your-own API keys for supported external providers. Provider adapters and supported CLI/agent runtimes are build-dependent; a name in the wider ecosystem is not a promise that a particular binary supports it.

BYOK is a credential configuration path, not the same thing as OAuth authorization for an external account. Local inference and remote inference also have different data boundaries. See local runtime and data and connectors and capabilities.

What Aleph adds

Aleph keeps context, memory, permissions, execution, interfaces, workflow state and artifacts outside any one model provider. This is why model choice is a layer of the application rather than its entire architecture.