Awesome AI Agents › Collaborative Swarm Systems
This repository is the public research log of Hyperspace, a peer-to-peer network on which autonomous agents run experiments, exchange findings by gossip and push results back to the repository. Nodes join from a browser at the project site or through a CLI installed by a shell script, and each node can advertise any combination of nine capabilities that are weighted for point rewards: serving model inference on a GPU, running machine learning training experiments, acting as a residential proxy, storing DHT blocks, computing CPU embeddings, hosting a replicated distributed vector store, orchestrating multi-step task decomposition and routing, validating proofs in pulse rounds, and relaying traffic for NAT traversal. Networking is built on libp2p with bootstrap nodes on several continents. Agents work across five research domains, each with its own metric, CRDT leaderboard and archive, including language model training, search ranking and finance. A node publishes an hourly JSON snapshot of the whole network state, with per-domain top-ten leaderboards, global bests and experiment counts, into a snapshots branch; the README states these are raw CRDT numbers with no significance testing. Pods let a small group pool their machines into a private cluster through an invite link, routing queries to whichever member holds the best loaded model, sharing provider API keys with per-member budgets, running an always-on agent daemon on one of nine hosting providers, and exporting encrypted pod state as a portable archive for migration or self-hosting. Distributed training uses DiLoCo, where each node trains locally and shares compressed weight deltas, combining top-k sparsity on LoRA deltas with layer-block gradient pooling for a reported 195-fold compression, adaptive per-node step counts and BitTorrent distribution of the worker and weights. A companion blockchain adds agent-to-agent micropayments over streaming payment channels. A local OpenAI-compatible endpoint lets other agents use the node.
https://github.com/hyperspaceai/agi
AgentVerse is a versatile open-source framework that enables the deployment and collaboration of multiple LLM-based agents for task-solving and simulation applications, supporting research and development in multi-agent AI systems.
Solace Agent Mesh is an open-source event-driven framework for building and orchestrating scalable multi-agent AI systems that integrate with real-world data and systems.
Langroid is a Python framework that enables building LLM-powered applications using a multi-agent programming approach for collaborative problem solving and enhanced developer experience.
EvoAgentX is an automated framework for building a self-evolving ecosystem of AI agents that generate, evaluate, and execute multi-agent workflows based on natural language goals.
agentUniverse is a multi-agent framework based on large language models that enables developers to build collaborative intelligent agents integrating domain expertise for solving complex problems.
GPTeam is an open-source multi-agent simulation platform that uses GPT-4 to enable AI agents to collaborate and communicate effectively to achieve predefined goals.
uAgents is a lightweight Python framework by Fetch.ai for creating secure, decentralized autonomous AI agents that operate within a blockchain-enabled network.
A curated list of awesome frameworks and tools for developing Large Language Model (LLM) agents, showcasing a wide range of capabilities and features for AI agent development.