cognis-core 0.1.0

Core traits and types for the Cognis LLM framework
Documentation

cognis-core

Foundation layer for the Cognis LLM framework. Defines the base traits, types, and abstractions that all other workspace crates depend on. This crate has zero dependencies on other workspace crates.

Key Traits and Types

Trait / Type Module Description
BaseChatModel language_models Interface for chat model providers
BaseLLM language_models Interface for completion-style LLMs
Runnable runnables Composable unit of async computation
BaseTool tools Interface for agent-callable tools
Embeddings embeddings Interface for vector embedding providers
Message messages Enum: Human, AI, System, Tool, Function
Document documents Text document with metadata
VectorStore vectorstores Interface for similarity search stores

Runnables

The Runnable trait and its combinators form the LCEL (LangChain Expression Language) equivalent:

  • RunnableSequence -- chain runnables with the chain! macro
  • RunnableParallel -- fan-out to multiple runnables
  • RunnableBranch -- conditional routing
  • RunnableLambda -- wrap a closure as a runnable
  • RunnableWithFallbacks -- automatic fallback on error
  • RunnableRetry -- retry with configurable policy

Usage

[dependencies]
cognis-core = { path = "../cognis-core" }

use cognis_core::messages::Message;
use cognis_core::language_models::FakeListChatModel;

let model = FakeListChatModel::new(vec!["Paris".into()]);
let msg = Message::human("What is the capital of France?");

Testing

Fake model implementations (FakeListChatModel, FakeListLLM, ParrotFakeChatModel, etc.) are provided for testing without network calls.