Awesome Infra for AI › Workflow Orchestration for AI

bosun-ai/swiftide

⭐ 789 Rust repository created 2024-06-09

Swiftide is a Rust-based framework designed for developing advanced LLM applications, focusing on composable agents, structured task orchestration, and efficient Retrieval-Augmented Generation (RAG) pipelines. It provides an agent harness that enables the construction of autonomous agents capable of looping over LLM calls, tool execution, and lifecycle hooks, with support for human-in-the-loop approval and streaming responses. The framework introduces typed task graphs as its orchestration layer, allowing developers to define explicit workflows by composing various Swiftide components, agents, and custom functions with strong compile-time type checking. This includes capabilities for parallel branch execution, pausing, resuming, and resetting tasks. For RAG, Swiftide offers streaming, composable indexing and query pipelines, which are essential for handling large-scale data and ensuring efficient retrieval. The framework integrates with tracing tools and metrics, including Langfuse, for comprehensive observability into agent, task, and pipeline execution. It supports various integrations, such as OpenAI, and provides mechanisms for adding custom tools, managing message history, and controlling agent behavior. Swiftide's core philosophy emphasizes leveraging Rust's performance and type safety to build robust and scalable LLM applications.

https://github.com/bosun-ai/swiftide

RustLLM applicationsAI agentsRAGworkflow orchestrationprompt managementtask graphsstreamingobservabilityLLM Ops