cognis-core

The foundation traits and types for the Cognis LLM framework.

Workspace | API Docs

cognis-core defines the trait interfaces and types that every crate in the Cognis workspace builds on. It has zero workspace dependencies — everything starts here.

Core Abstractions

BaseChatModel  ─  Chat model providers (Anthropic, OpenAI, ...)
BaseTool       ─  Tools that agents can call
Runnable       ─  Composable async computation unit (the LCEL equivalent)
Message        ─  Human | AI | System | Tool message types
Embeddings     ─  Vector embedding providers
VectorStore    ─  Similarity search interface
Document       ─  Text + metadata, used across loaders and retrievers

Composable Chains with chain!

The Runnable trait and its combinators let you compose pipelines:

use std::sync::Arc;
use cognis_core::chain;
use cognis_core::language_models::{ChatModelRunnable, FakeListChatModel};
use cognis_core::output_parsers::StrOutputParser;
use cognis_core::prompts::ChatPromptTemplate;
use cognis_core::runnables::Runnable;
use serde_json::json;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let prompt = ChatPromptTemplate::from_messages(vec![
        ("system", "You are a helpful assistant."),
        ("human", "Explain {topic} in one sentence."),
    ])?;

    let model = FakeListChatModel::new(vec![
        "Rust ensures memory safety without garbage collection.".into(),
    ]);

    let chain = chain!(
        prompt,
        ChatModelRunnable::new(Arc::new(model)),
        StrOutputParser
    )?;

    let result = chain.invoke(json!({"topic": "Rust"}), None).await?;
    println!("{}", result.as_str().unwrap());
    Ok(())
}

Runnable Combinators

Testing

Fake model implementations are included for testing without API keys or network:

• FakeListChatModel — cycles through predefined string responses
• FakeMessagesListChatModel — cycles through predefined Message responses
• GenericFakeChatModel — word-level streaming from predefined messages
• FakeListLLM — completion-style fake model
• DeterministicFakeEmbedding — produces reproducible embeddings

Part of the Cognis Workspace