cognis

Implementation layer for the Cognis LLM framework. Provides concrete chat model integrations, agent execution, chains, memory, document loaders, text splitters, embedding providers, and built-in tools.

Chat Model Providers

Each provider is behind a feature flag to keep compile times and dependencies minimal.

Key Modules

• agents -- Agent executor with middleware pipeline (retry, PII redaction, summarization, human-in-the-loop, tool selection, and more)
• chains -- LLM chain, conversation chain, sequential chain
• memory -- Buffer, window, and summary memory strategies
• document_loaders -- Text, CSV, JSON, and directory loaders
• text_splitter -- Character, recursive, markdown, HTML, JSON, code, and token splitters
• embeddings -- OpenAI and Ollama embedding providers
• tools -- Calculator, shell command, and JSON query tools

Usage

[dependencies]
cognis = { path = "../cognis", features = ["anthropic"] }

use cognis::chat_models::anthropic::ChatAnthropic;
use cognis_core::runnables::Runnable;
use serde_json::json;

let model = ChatAnthropic::new("claude-sonnet-4-20250514");
let result = model.invoke(json!({"messages": []}), None).await.unwrap();

Feature Flags