Expand description
LLM driver abstraction.
Defines the LlmDriver trait — the interface between the agent
loop and LLM inference backends. The default implementation is
RealizarDriver (sovereign, local GGUF/APR inference).
Modules§
- apr_
serve - AprServeDriver — first-class inference via
apr servesubprocess. - chat_
template - Chat template formatting for local LLM inference.
- mock
- Mock LLM driver for deterministic testing.
- realizar
- RealizarDriver — sovereign local inference via GGUF/APR models.
- remote
- Remote HTTP driver for external LLM APIs.
- router
- Routing driver for local-first, remote fallback inference.
- validate
- Model file validation (CONTRACT:
apr_model_validity).
Structs§
- Completion
Request - Request to the LLM driver.
- Completion
Response - Response from the LLM driver.
- Tool
Call - A tool call request from the model.
- Tool
Definition - Tool definition for the LLM (JSON Schema).
- Tool
Result Msg - A tool result message.
Enums§
- Message
- Message in the agent conversation.
- Stream
Event - Streaming event from the LLM driver.
Traits§
- LlmDriver
- Abstraction over LLM inference backends.