We're thrilled to announce our latest update, featuring new enhancements from the past 15 days.
We shipped all-new multi-agent architectures, Multi-Agent Execution Utilities, new examples, updated documentation, improvements, and much more!
Learn more ⬇️🧵
2 /
All-new: BatchedGridWorkflow
The BatchedGridWorkflow is an all-new multi-agent orchestration pattern that executes tasks in a batched grid format, where each agent processes a different task simultaneously.
This workflow is particularly useful for parallel processing scenarios where you have multiple agents and multiple tasks that can be distributed across them.
We have significantly simplified the MajorityVoting architecture by incorporating an agent as the final aggregator, enhancing prompt design, and removing unused code.
⎆ Now easier to use
⎆ Streamlined design and architecture
⎆ Enhanced aggregation/consensus step
To conclude, this update introduces a new multi-agent architecture, multi-agent runtime utilities, and improvements such as streaming callbacks essential for enterprise-grade deployment.
This update empowers app builders to create faster and more scalable multi-agent systems!
Get started building bleeding-edge agent applications with the Swarms framework:
This major Swarms framework update introduces Claude Agents Compatibility, an enhanced AgentLoader class for loading thousands of agents, along with various new improvements and bug fixes!
Learn more ⬇️🧵
2 /
All-New [Reasoning Integration]
Added advanced reasoning capabilities for improved agent decision-making and logical processing
Many people don't realize that Swarms was one of the first agent frameworks ever created more than 3 years ago.
Swarms has inspired countless frameworks, including CrewAI, OpenAI's Swarm, Autogen, Agno, and Langraph all of which are descendants of Swarms.
Learn more 🧵👇
2 /
Swarms was originally conceived after the release of the Chain of Thought paper.
Our founder, @KyeGomezB, implemented the Chain of Thought paper while working on open-source AI research.
He would then go on to implement Tree of Thoughts, Algorithm of Thoughts, and countless other AI model architectures and prompting techniques.
He discovered that, regardless of how intelligent individual LLMs were, they needed to collaborate to enhance their performance, reduce hallucination, handle multiple tasks concurrently, and increase the memory capacity.
Our latest Swarms update significantly enhances reliability and performance at scale, paving the way for large-scale multi-agent simulations and structures.
Learn more ⬇️🧵
2 /
All-New: SenatorAssembly
We've introduced the SenatorAssembly, a simulation featuring agents modeled after the 100 U.S. senators.
Use Cases:
> Predict senator votes before official tallies.
> Simulate an autonomous democratic system.
>Model committee hearings.
The Board of Directors is a an all-new multi-agent architecture that implements collective decision-making through democratic processes, voting mechanisms, and role-based leadership.
This architecture provides an alternative to single-director patterns by enabling collaborative intelligence through structured governance.
We're thrilled to announce the latest updates to our Rust framework, including all-new multi-agent architectures, enhanced logging modules, bug fixes, and various improvements!
Learn more ⬇️ 🧵
2 /
All-New: AgentRearrange is Here
AgentRearrange is a multi-agent framework that allows you to customize how agents interact based on their names and specific symbols. For example, -> denotes sequential communication, while commas (,) indicate parallel communication.
The new logging system in swarms-rs provides a feature-rich, colorful logging infrastructure specifically designed for multi-agent systems. It includes:
⎆ Environment-based configuration via SWARMS_LOG_LEVEL environment variable
⎆ Contextual logging macros for agents, tasks, tools, workflows, memory, and LLM interactions
⎆ Performance monitoring with dedicated performance logging macros
⎆ Color-coded output using the colored crate for better readability
⎆ Structured logging with timestamps, log levels, and target information
⎆ Agent-specific context in all log messages for better debugging and monitoring