Today we’re diving deep into 25 open-source AI agent projects on GitHub, From intelligent memory layers to multi-agent orchestration platforms.
00:00 - Introduction
00:12 - Julep https://github.com/julep-ai/julep
00:38 - Self-hosted AI starter kit https://github.com/n8n-io/self-hosted-ai-starter-kit
01:03 - Big-AGI https://github.com/enricoros/big-AGI
01:28 - AnyTool https://github.com/HKUDS/AnyTool
01:56 - playwriter https://github.com/remorses/playwriter
02:23 - vibe-kanban https://github.com/BloopAI/vibe-kanban
02:49 - Claude Code PM https://github.com/automazeio/ccpm
03:17 - Agent S https://github.com/simular-ai/Agent-S
03:41 - Praison AI https://github.com/MervinPraison/PraisonAI
04:06 - potpie https://github.com/potpie-ai/potpie
04:31 - liveblocks https://github.com/liveblocks/liveblocks
04:57 - SuperMCP https://github.com/dhanababum/supermcp
05:22 - oh-my-opencode https://github.com/code-yeongyu/oh-my-opencode
05:48 - Kortix https://github.com/kortix-ai/suna
06:15 - MCPProxy https://github.com/smart-mcp-proxy/mcpproxy-go
06:42 - Agno https://github.com/agno-agi/agno
07:07 - BubbleLab https://github.com/bubblelabai/BubbleLab
07:30 - Claude-Mem https://github.com/thedotmack/claude-mem
07:55- Bolna https://github.com/bolna-ai/bolna
08:20 - MemMachine https://github.com/MemMachine/MemMachine
08:46- InsForge https://github.com/InsForge/InsForge
09:12- klavis https://github.com/Klavis-AI/klavis
09:35- Neo https://github.com/neomjs/neo
09:59 - browser-debugger-cli https://github.com/szymdzum/browser-debugger-cli
10:23 - GHOSTCREW https://github.com/GH05TCREW/ghostcrew
No.1
Julep is basically Firebase for AI agents. It’s an open-source platform for deploying serverless, agent-based AI workflows—without touching infrastructure. Instead of chaining prompts together, Julep lets you orchestrate complex, multi-step agent logic with memory, and parallel execution. You can build AI agents that remember past interactions, call external APIs, and scale automatically as usage grows.
No.2
This project is a one-command, self-hosted AI lab. The Self-Hosted AI Starter Kit is an open-source Docker Compose setup that instantly spins up a full local AI stack—perfect for building and testing AI agents without relying on cloud services. It bundles n8n for low-code workflows, Ollama for running local LLMs, Qdrant for vector search, and PostgreSQL for reliable data storage.
No.3
Big-AGI is not another AI chat clone—it’s an expert-grade AI workspace. a multi-model AI environment built for engineers and researchers who need serious control and serious performance. What really sets it apart is Beam and Merge, which let you combine multiple models to reduce hallucinations and validate outputs, plus native search and support for models—all in a web app.
No.4
AnyTool is the missing layer between AI agents and real-world tools. It’s a universal tool-use system designed to make agent automation actually reliable in production—solving problems like tool overload, unreliable integrations, and limited capabilities. AnyTool uses smart context filtering to surface only the right tools in milliseconds, keeping agent prompts clean and fast, even when you’re working with thousands of tools.
No.5
Playwriter fixes browser automation for AI agents—by not overengineering it. Instead of spawning a separate Chrome instance like Playwright MCP, Playwriter runs as a browser extension, letting your AI agent work directly inside your existing Chrome session. That means 90% less context window usage, full access to the entire Playwright API, and the ability to collaborate with your agent in real time.
No.6
Vibe Kanban is a control panel for your AI coding agents. As AI agents write more of the code, developers are shifting into a role of planning, reviewing, and orchestrating—and that’s exactly what Vibe Kanban is built for. It gives you a Kanban-style board to manage multiple coding agents like Claude Code, Gemini CLI, Codex, and Amp—letting you run them in parallel or in sequence, all from one place.
No.7
Claude Code PM turns AI coding from a solo experiment into a real team workflow. It’s a project management system built on GitHub Issues and Git worktrees, designed to let multiple Claude Code agents work in parallel—without losing context or stepping on each other. Instead of vibe-coding from memory, this system converts PRDs into epics, epics into issues, and issues into production code, with full traceability at every step.
No.8
Agent S is teaching AI to use a computer like a human. It’s an open-source agentic framework that lets autonomous agents interact with real computer interfaces—clicking, typing, navigating apps—through a true Agent-Computer Interface. What makes Agent S stand out is performance. On benchmarks like OSWorld, Agent S3 already beats previous state of the art.
No.9
PraisonAI is a production-ready framework for building serious multi-agent systems. It’s a low-code, multi-AI-agent platform that brings together PraisonAI Agents, AG2—formerly AutoGen—and CrewAI into a single, unified workflow. The focus here is simplicity, customization, and real human-agent collaboration. You get agents with reasoning, self-reflection, short- and long-term memory.
No.10
Potpie turns prompts into code-aware engineering agents. It’s an open-source platform that builds AI agents specialized in your own codebase, using a knowledge graph to deeply understand how everything fits together. Instead of generic code assistants, Potpie’s agents understand relationships between files, APIs, and components—making them far more effective at real engineering tasks.
No.11
Liveblocks is the collaboration layer for human and AI workflows. In the AI era, the best products aren’t solo experiences—and Liveblocks makes it easy to add real-time collaboration without rewriting your app or blowing up your roadmap. It offers ready-made, customizable features like AI copilots, multiplayer editing, comments, presence, and notifications—components you can drop straight into your product.
No.12
SuperMCP is the platform for managing multiple AI assistants and their data connections. It lets you create multiple isolated MCP servers from a single connector, bridging AI assistants with databases, APIs, and other data sources. Each server runs independently with its own credentials, and settings. That means multi-tenant apps, and enterprise deployments become far simpler to manage.
No.13
Oh-My-OpenCode is coding on steroids for AI agents. This open-source project turns your LLM agents into a full development powerhouse, orchestrating multiple specialized agents —all working together asynchronously. It comes battery-included with curated MCPs and a Claude Code compatibility layer, so your agents can handle everything from debugging to feature development without choking on context.
No.14
Kortix is a full platform for building autonomous AI agents that work for you. At the heart of Kortix is the Kortix Super Worker, a flagship generalist agent that demonstrates what’s possible—handling research, data analysis, and complex workflows through natural conversation. Beyond that, Kortix lets you build custom agents tailored to any domain or workflow, from customer service to content creation.
No.15
MCPProxy is a smart proxy that supercharges your AI agents. This open-source desktop app lets agents federate hundreds of MCP servers, bypassing API limits like Cursor’s 40-tool cap or OpenAI’s 128-function limit. It also saves tokens and accelerates responses—agents only load a single retrieve_tools function instead of hundreds of schemas, reducing token usage by ~99% while improving accuracy.
No.16
Agno is a fast framework for building, running, and managing multi-agent systems. Designed for companies that want secure, containerized AI products, Agno goes far beyond simple API calls—it provides a full stack for agent engineering. The Framework lets you build agents, multi-agent teams, and workflows with memory, knowledge, context compression, and over 100 built-in toolkits.
No.17
Bubble Lab is an open-source, agentic workflow automation builder with full code ownership. Unlike traditional workflow platforms that trap you in JSON nodes, Bubble Lab is TypeScript-native, giving you full control, type safety, and transparency. Everything you build compiles into clean, production-ready code that you can deploy, debug, and integrate anywhere.
No.18
Claude-Mem is the memory backbone for your Claude Code sessions. This plugin automatically captures everything Claude does, compresses it using AI, and injects relevant context into future sessions—so your agent never forgets what it learned. You can explore memory in real time via a web UI, search it directly from Claude Desktop, and even reference past decisions with claude-mem.
No.19
Bolna is an end-to-end open-source framework for building voice-driven AI applications. It provides a full orchestration platform to create LLM-powered voice agents, combining ASR, LLM, and TTS models over WebSockets for seamless conversations. With Bolna, you can build AI voice agents capable of handling complex tasks through natural speech, all on a production-ready open-source backbone.
No.20
MemMachine is a universal memory layer for AI agents. It lets your agents store, recall, and learn from past interactions, creating context-aware, personalized AI assistants that remember preferences, session history, and knowledge across multiple agents and LLMs. MemMachine supports working , persistent , and personalized memory types, giving your agents the ability to reason and adapt over time.
No.21
InsForge is the backend platform built for AI-assisted development. It lets you connect any AI agent—Claude, GPT, or others—and instantly add authentication, databases, storage, serverless functions, and AI integrations to your app. With InsForge, developers can build complete AI-driven apps using natural language, empowering agents to manage the backend, handle data, and interact with users seamlessly.
No.22
Klavis AI is an open-source MCP integration platform that lets AI agents use tools reliably at any scale. With Klavis API, your agents can connect to hundreds of SaaS tools, databases, and APIs in minutes—no need to build integrations from scratch. Authentication is fully managed, with OAuth flows, refresh tokens, and credential security handled for you.
No.23
Neo.mjs is the AI-native platform for multi-window web applications. Designed for desktop-class performance in the browser, Neo.mjs uses a pioneering Off-Main-Thread architecture to keep the UI buttery smooth, no matter how many real-time updates your app handles. Neo.mjs empowers agents to act as autonomous developers with the AI SDK, and local script execution at machine speed.
No.24
Browser Debugger CLI brings Chrome DevTools Protocol straight into your terminal. Designed for AI agents and developers who want direct, low-overhead browser control, bdg opens a persistent connection to Chrome, letting you execute commands sequentially via Unix pipes. Use bdg when you need precise, low-level browser automation for AI agents or quick debugging.
No.25
GHOSTCREW is an AI agent framework built for modern security teams. Designed for bug bounty hunters, red-teamers, penetration testers, and security educators, GhostCrew combines LLM autonomy with multi-agent coordination to model how real-world offensive security workflows operate—without bloated tooling. At its core, GHOSTCREW focuses on structured, reproducible agent workflows.