Vibe Coding DiscoverCategory
AI Agents
Autonomous and stateful agents for research, coding, and workflows.
| Project name | Stars | Category | Language | Tags | Summary | View Project |
|---|---|---|---|---|---|---|
| plur | ★ 296 | AI Agents | TypeScript | agent memory · shared memory · persistent memory · local-first · MCP · hybrid search · embeddings · memory retrieval · context management · open format | PLUR provides local-first, persistent memory shared across AI agents and coding tools. It stores inspectable text-based memories and supports MCP integrations, scoped retrieval, and hybrid search with local embeddings. | View Project → |
| Containarium | ★ 295 | AI Agents | Go | agent-runtime · agent-sandbox · isolation · mcp · ebpf · kubernetes · lxc · ssh · self-hosted · multi-tenant · gpu-passthrough · sandboxing · coding-agents · go · code-execution · egress-policy | Self-hosted agent runtime that gives each AI agent a persistent, SSH-reachable isolated box on LXC or Kubernetes, with eBPF egress policy, GPU passthrough, and MCP-native CLI plus in-box shell/file tools for Cursor, Claude Code, and other MCP clients. | View Project → |
| Context-Engineering-for-Multi-Agent-Systems | ★ 283 | AI Agents | Jupyter Notebook | multi-agent-systems · context-engineering · mcp · rag · agentic-ai · semantic-orchestration · observability · glass-box · langchain · nemotron · pinecone · gradio · prompt-injection-defense · token-analytics · computer-use · governance | Book companion repo (Jupyter notebooks) that builds a domain-agnostic, glass-box Context Engine for multi-agent systems: MCP orchestration, dual RAG with citations, injection defenses, and token analytics, runnable on OpenAI, LangChain, Nemotron, or DeepSeek-R1. | View Project → |
| starnet | ★ 280 | AI Agents | JavaScript | multi-agent · agent-harness · desktop-app · local-first · BYOK · agent-tools · agent-skills · agent-scheduling · MCP · pixel-art | A local-first desktop harness for creating and running multi-agent teams with distinct workspaces, tools, permissions, memory, and schedules. Supports cloud providers, Ollama, MCP connectors, and messaging integrations; the pixel-art station reflects live runtime state. | View Project → |
| Doberman-Core | ★ 277 | AI Agents | Python | agent security · runtime guardrails · authorization · tool-call filtering · MCP proxy · human approval · audit logging · coding agents | Doberman intercepts coding-agent tool calls through MCP proxies and host hooks, then allows, requires approval for, or blocks each action before execution. It provides policy controls and an auditable decision log. | View Project → |
| open-vetta | ★ 277 | AI Agents | TypeScript | local-first · byok · mcp · skills · plugin-system · desktop-app · electron · knowledge-base · workflow-automation · coding-agent · privacy-first · cross-platform · cli · typescript · bun · multi-provider | Open Vetta is a local-first desktop AI agent (Electron/TypeScript) for coding and general work. It offers BYOK model providers, MCP connectors, skills, plugins, workflows, and local knowledge bases with reviewable, permission-bounded tool execution. | View Project → |
| nac | ★ 276 | AI Agents | Rust | agent-harness · orchestration · long-running-agents · multi-agent · agent-skills · mcp · rust · threads · episodes · orchestrator · cli · web-dashboard · sandbox · llm-agents · agentic-workflows · developer-tools | nac is a Rust agent harness for long-running, ambitious tasks. A central orchestrator plans and delegates to threads that return structured episodes, keeping work aligned with original intent. It ships a web dashboard, MCP/skills integration, sandboxing, and multi-provider model auth. | View Project → |
| busabase | ★ 269 | AI Agents | TypeScript | agent-workspace · agent-memory · mcp · local-first · self-hosted · knowledge-base · human-in-the-loop · audit-trail · change-requests · skills · structured-data · system-of-record · openapi · cli · notion-alternative · airtable-alternative | Self-hostable, local-first database and workspace where AI agents read and write structured Bases, Docs, Skills and Apps. Every agent write lands as a reviewable Change Request with diff, author and audit history; connect via MCP, OpenAPI/CLI or Agent Skill. | View Project → |
| uteke | ★ 268 | AI Agents | Rust | agent-memory · local-first · offline-first · embeddings · semantic-search · vector-database · hnsw · sqlite · fts5 · rrf-fusion · mcp-server · cli · rust · onnx · multi-agent · time-travel | Uteke is a local-first memory engine for AI agents: a single Rust binary that stores, recalls and semantically searches memories offline using ONNX embeddings, HNSW plus SQLite FTS5 with RRF fusion, and ships an MCP server for Claude, Cursor and Copilot. | View Project → |
| pi-fabric | ★ 268 | AI Agents | TypeScript | agents · multi-agent · orchestration · mcp · code-mode · quickjs · sandboxing · pi-extension · workflows · actors · memory · recursive-queries · skills · runtime · typescript · tui | Pi extension runtime giving a coding agent one `fabric_exec` code-mode tool that composes core tools, MCP servers, and sub-agents in checked TypeScript or sandboxed Python. Adds durable actors, mesh/workflow coordination, councils, memory recall, and recursive queries. Requires Node 24+ and Pi 0.80.6+. | View Project → |
| jev-dsh-decision | ★ 263 | AI Agents | JavaScript | agent-harness · decision-engine · agent-plugin · tool-selection · skill-selection · agent-routing · quality-evaluation · structured-output · DeepSeek Harness · OpenCode · Codex Harness | A plugin that adds TypeSafe AI’s Jev structured decision engine to agent harnesses. It evaluates available tools, skills, agents, and outputs, returning structured recommendations without executing actions or taking over the host agent. | View Project → |
| animaworks | ★ 263 | AI Agents | Python | multi-agent · autonomous-agents · agent-framework · long-term-memory · brain-inspired · memory-consolidation · forgetting · organization-as-code · multi-model · worktree-isolation · github-automation · pr-review · ci-repair · hybrid-rag · task-orchestration · self-hosted | AnimaWorks is a Python framework for running an autonomous AI organization: role-based agents with brain-inspired memory that consolidate and forget, executing tasks in parallel git worktrees, reviewing PRs multi-model, repairing CI and deploying. Supports Claude, Codex, Gemini, Cursor and Ollama. | View Project → |
| huddol | ★ 263 | AI Agents | Python | AI agents · multi-agent · agent runtime · agentic workflows · automation · MCP · FastAPI · Tauri · React · Rust · Python · Telegram bot · workflow automation · AI workspace · desktop app · orchestration | Huddol lets you run your own organization of AI agents, combining a Python core with FastAPI, a Tauri desktop app, React web UI, and Telegram bot support for agentic workflows and automation. | View Project → |
| agentswarms | ★ 260 | AI Agents | TypeScript | multi-agent swarms · agent workflows · RAG · MCP · AI analyst · agent chat · data platform · lakehouse · self-hosted · no-code | A self-hosted platform for building agent chats and multi-agent workflows, with RAG and MCP alongside ETL, a lakehouse, BI, and an AI analyst. Supports multiple model providers and applies shared permissions, budgets, and audit trails. | View Project → |
| nextclaw | ★ 260 | AI Agents | TypeScript | ai-agent · personal-assistant · local-first · self-hosted · desktop-app · mcp · skills · scheduled-tasks · memory · multi-channel · agent-runtime · cli · docker · automation · browser-control · open-source | NextClaw is a local-first, open-source personal AI agent workspace (desktop, npm CLI, or Docker) with persistent memory, skills, MCP tools, scheduled jobs, and pluggable runtimes like Codex and Claude Code. It keeps conversations, files, and generated results in one task and connects to chat channels such as WeChat, Fe | View Project → |
| opencompany | ★ 249 | AI Agents | Rust | multi-agent · agent-orchestration · hive-mind · autonomous-agents · llm · rust · one-person-company · agent-runtime · quorum-sensing · stigmergy · collective-intelligence · desktop-app · tauri · cli · docker · mcp | Rust company runtime that runs a roster of specialized AI agents as a hive mind: desks deliberate via stigmergy, quorum sensing, and cross-inhibition to reach explainable decisions. Ships prebuilt company templates (software, marketing, VC) plus a desktop app and CLI. | View Project → |
| omnicoreagent | ★ 245 | AI Agents | Python | agent-harness · multi-agent · subagents · mcp · memory · telemetry · durable-runs · sandboxing · governance · budgets · background-tasks · guardrails · observability · orchestration · python · fastapi | OmniCoreAgent is an open Python runtime for building production AI agents: tools, MCP clients, memory, workspace, subagents, background tasks, sandboxed execution, governance/budget policies and tracing. Ships OmniServe to expose agents over REST/SSE. | View Project → |
| a2a-directory | ★ 241 | AI Agents | — | A2A · agent discovery · agent interoperability · agent cards · protocol · clients · servers · examples · directory | A community-maintained directory of A2A resources, implementations, tools, services, and examples. Includes guides to the protocol and integrations across agent frameworks. | View Project → |
| bagidea-office | ★ 236 | AI Agents | JavaScript | multi-agent · agent-orchestration · desktop-agents · Claude Code · delegation · agent-workflows · agent-memory · agent-skills · plugins · human-approval · local-LLMs · voice · Telegram · Discord · LINE | A desktop-wallpaper office for coordinating Claude Code and other AI agents. Agents can delegate work, run workflows, build persistent knowledge and skills, and communicate through chat and voice; human approval gates protect consequential actions. | View Project → |
| AgenticX | ★ 233 | AI Agents | Python | multi-agent · orchestration · meta-agent · mcp · rag · graphrag · memory · skills · sandbox · cli · desktop-app · a2a · tool-use · llm-providers · evaluation · observability | AgenticX is a production-oriented multi-agent platform: Python SDK, agx CLI, FastAPI Studio server and Electron desktop app. Adds Meta-Agent orchestration, MCP Hub, hierarchical memory, GraphRAG knowledge, skills, sandboxing and 15+ LLM providers. | View Project → |
| maestro | ★ 232 | AI Agents | TypeScript | multi-agent coordination · agent harness · workflow orchestration · human-in-the-loop · local-first · CLI · agent skills · MCP · Claude Code · Codex | A local-first TypeScript CLI for coordinating human and AI-agent work in Git repositories. It tracks tasks, decisions, sessions, evidence, and handoffs, and supports supervised teams and deterministic multi-agent workflows. | View Project → |
| quackd | ★ 232 | AI Agents | Python | robotics · embodied-ai · physical-ai · multi-robot · llm-agents · mcp · cli · humanoid · lerobot · ros2 · simulation · local-llm · safety-contracts · sim2real · computer-vision · flock | quackd is a Python CLI that connects multiple robots (Microduck, Open Duck Mini, LeRobot arms, AlohaMini, ToddlerBot, ROS bases) and pilots each with an LLM agent. It enforces .duck safety contracts, supports flocks, MCP, simulation and local models. | View Project → |
| guaardvark | ★ 228 | AI Agents | Python | self-hosted · local-ai · multi-agent · mcp · rag · video-generation · image-generation · voice-cloning · lora-training · coding-agent · agent-skills · ollama · comfyui · screen-agents · gpu-orchestration · worktrees | Self-hosted AI studio bundling local video/image/music/voice generation, LoRA training, hybrid RAG, desktop screen agents and up to 20 parallel coding agents in git worktrees on one GPU, driven from a Studio UI or via MCP and agent skills from Claude Code, Cursor or Codex. | View Project → |
| iva-agent | ★ 225 | AI Agents | TypeScript | telegram-bot · long-term-memory · self-hosted · personal-assistant · obsidian · knowledge-graph · local-first · privacy · voice-assistant · mcp · skills · second-brain · prompt-injection-guard · markdown-vault · systemd-deploy · userbot | Self-hosted Telegram AI assistant that files voice notes, photos and messages into an Obsidian-compatible markdown vault with layered memory (daily logs rolled up into CORE.md cards). Supports MCP servers, skill files, five LLM providers and Google Workspace actions. One-command install on a VPS. | View Project → |
| lobu | ★ 224 | AI Agents | TypeScript | agent-memory · multi-agent · mcp · context-layer · event-sourcing · knowledge-graph · self-hosted · typescript · slack-bot · connectors · permissions · audit · agent-runtime · identity · specialists · persistent-agents | Lobu is a self-hosted, event-sourced context layer for AI agents: it ingests connectors, webhooks and conversations into an append-only log linked to typed entities. Any MCP client (Claude, Codex, ChatGPT) reads and writes the same permission-aware memory, and persistent specialist agents can own durable responsibiliti | View Project → |
| forge | ★ 223 | AI Agents | Go | agent-runtime · agent-skills · SKILL.md · mcp-client · a2a · portable-agents · egress-control · self-hosted · kubernetes · air-gapped · audit-logging · cli · go · llm-fallback · cron-scheduling · memory | Go-based runtime for Anthropic's Agent Skills standard: write a SKILL.md, compile it into a portable agent, and run it in Docker, Kubernetes, air-gapped, CI, or as an A2A endpoint. Adds egress allowlists, encrypted secrets, NDJSON audit logs, MCP client, Slack/Telegram channels, and multi-provider LLM fallback. | View Project → |
| magda-core | ★ 221 | AI Agents | C++ | AI music production · DAW · AI agent · natural language control · MCP · audio · MIDI · DSL · local inference · cross-platform | MAGDA is a C++20 digital audio workstation with an integrated AI chat and agent system. Natural-language requests generate and execute DAW commands through a custom DSL; it also supports local inference and documents a remote MCP API. | View Project → |
| three.ws | ★ 218 | AI Agents | JavaScript | 3d-agents · avatars · gltf · glb · three-js · webgl · llm-brain · memory · emotions · mcp-server · x402 · solana · evm · web-component · character-studio · oauth2 | Browser-native framework for 3D AI agents: GLB/glTF avatars driven by LLM brains with memory and emotions, embedded anywhere as a web component. Ships MCP servers, x402/Solana/EVM autonomous payments, character studio, and an animation gallery. | View Project → |
| cli-jaw | ★ 208 | AI Agents | TypeScript | AI assistant · multi-agent · agent orchestration · CLI · coding agents · dashboard · persistent memory · browser automation · Telegram · Discord · Slack · MCP · local-first | CLI-JAW is a local personal AI assistant that coordinates coding-agent CLIs including Claude, Codex, Cursor, Copilot, and OpenCode. It provides shared memory, session management, and terminal, web, desktop, and messaging interfaces. | View Project → |
| kasetto | ★ 206 | AI Agents | Rust | AI agents · agent environments · skills · MCP · commands · instructions · multi-agent · declarative config · environment management · Rust | Kasetto manages AI agent environments from a declarative YAML config, syncing version-locked skills, commands, instructions, and MCPs across agents such as Claude Code, Cursor, Codex, and Gemini CLI. It supports shared team configs, secrets, and project or global scopes. | View Project → |
| jido_ai | ★ 204 | AI Agents | Elixir | elixir · llm · agents · react · reasoning-strategies · tool-calling · llm-orchestration · model-routing · retries · observability · streaming · checkpointing · structured-output · multi-provider · otp | Elixir AI runtime for building tool-using Jido agents. Provides ReAct loops, eight reasoning strategies (CoT, ToT, GoT, TRM, Adaptive), model routing across Anthropic/OpenAI/Google, telemetry, and policy/quota plugins. | View Project → |
| AgentQuant | ★ 204 | AI Agents | Python | quantitative-trading · agentic-ai · self-improving · react-agent · backtesting · harness-evolution · walk-forward-validation · portfolio-optimization · regime-detection · long-term-memory · fintech · quant-research · no-code · evaluation-harness | AgentQuant is a Python research platform where a ReAct LLM agent proposes trading strategies, backtests them with walk-forward validation, reflects, and stores results in cross-session memory. It evolves its own search harness across epochs under quality gates; the zero-key demo runs offline. | View Project → |
| zeroid | ★ 204 | AI Agents | Go | agent-identity · authentication · oauth2 · delegation · mcp · spiffe · wimse · dpop · token-exchange · ciba · attestation · revocation · nhi · zero-trust · golang · jwt | ZeroID is a Go-based identity server giving autonomous AI agents their own credentials: OAuth 2.1 issuance, RFC 8693 on-behalf-of delegation with scope attenuation, SPIFFE/WIMSE identities, DPoP, attestation, and cascade revocation for agent chains and MCP servers. | View Project → |
| javis-os | ★ 200 | AI Agents | Python | agent-os · agent-orchestration · multi-model · self-hosted · mcp-hub · second-brain · skills · voice-assistant · knowledge-graph · workflows · scheduler · kanban · telegram-bot · plugins · llm-ops · docker | Javis OS is a self-hosted agentic AI platform: an MCP-equipped agent with swappable LLM backends (Claude Code, Codex, Gemini, Ollama, OpenRouter), a persistent Second Brain, skills, background loops, Kanban task delegation, knowledge graph, voice chat and a web dashboard. | View Project → |
| jev-drone | ★ 196 | AI Agents | Python | autonomous-drone · robotics · computer-vision · simulation · flight-control · obstacle-avoidance · camera-only · tactical-reasoning | Simulates a camera-only quadrotor navigating an obstacle course in MuJoCo. A Jev judgment model proposes tactical maneuvers while conventional control and safety code retain flight control and veto authority. | View Project → |
| Lexora | ★ 195 | AI Agents | TypeScript | desktop-agent · autonomous-agents · agent-loop · mcp · skills · sandbox · local-ai · local-first · task-automation · file-automation · scheduled-tasks · desktop-pet · electron · rust · linux · multi-model | Lexora is a local-first desktop AI agent (Vue + Electron + a separate TypeScript runtime, Rust native layer) that runs an agent loop with sandboxed tool execution, MCP integrations, Skills and scheduled tasks. It is model-agnostic and requires explicit user authorization for file and tool access. | View Project → |
| openworkbuddy | ★ 194 | AI Agents | JavaScript | AI agent · office automation · local-first · self-hosted · BYOK · file generation · MCP · CLI · Electron · skills · scheduled tasks · multi-model | A self-hosted AI office agent for desktop, web, and CLI that plans and executes tasks, then verifies deliverables such as PPTX, DOCX, XLSX, and websites. Supports configurable LLM providers, Ollama, Claude Code, Codex, MCP, skills, and scheduled messaging workflows. | View Project → |
| dsh-auto-review | ★ 193 | AI Agents | TypeScript | ai-safety · approval · auto-review · guardrails · subagent · deepseek-harness · dsh-plugin · mcp · sandbox · fail-closed · audit-trail · circuit-breaker · typescript · llm · second-model · human-in-the-loop | A DeepSeek Harness plugin that puts a second, read-only reviewer subagent on the approval chain: it inspects tool arguments and workspace evidence, returns structured allow/deny verdicts with reasons, and fails closed by default with a full session-log audit trail. | View Project → |
| sample-strands-agent-with-agentcore | ★ 191 | AI Agents | Python | multi-agent · agentic-chatbot · browser-automation · agent-to-agent · MCP · agent-skills · voice-assistant · memory · tool-use · AWS | An end-to-end AWS reference architecture for a multi-agent chatbot built with Strands Agents and Bedrock AgentCore. Demonstrates MCP and A2A tools, persistent memory, browser automation, voice, skills, and Terraform deployment. | View Project → |
| zafiro | ★ 191 | AI Agents | Kotlin | Android · phone-use · computer-use · device-control · BYOK · MCP · skills · Python · Shell · Shizuku · SSH · voice-assistant | Zafiro is a BYOK AI agent for Android that can interact with apps and control the device through Shizuku or root. It supports MCP, skills, memory, on-device Python and shell tools, and SSH to remote machines. | View Project → |
| noobot | ★ 191 | AI Agents | JavaScript | ai-agent · self-hosted · multi-agent · mcp · tool-calling · model-routing · sandbox · workflow-automation · character-animation · threejs · 3d · multimodal · electron · webgl · browser-automation · skills | Self-hosted AI agent workspace in Node.js/Vue 3/Electron with tool calling, MCP connectors, multi-model routing, sandboxed execution, multi-agent workflows, and LLM-authored 3D character animation rendered in Three.js. | View Project → |
| nanzi-ai-agent-platform | ★ 189 | AI Agents | Python | multi-agent · agent-orchestration · chatbi · text-to-sql · rag · mcp · sandbox · rbac · long-term-memory · browser-automation · knowledge-base · task-scheduler · audit-log · embed-sdk · self-healing-sql · agentscope | Enterprise multi-agent platform built on AgentScope: orchestrates delegating expert agents for ChatBI (text-to-SQL over Oracle/ClickHouse/MySQL), RAG knowledge Q&A, MCP tool use, multi-policy sandboxes, browser automation, RBAC, and scheduled notifications. | View Project → |
| oto-dock | ★ 187 | AI Agents | Python | multi-agent · self-hosted · agent-platform · team collaboration · agent orchestration · scheduled agents · webhooks · agent memory · agent skills · MCP · voice agents · Docker | Self-hosted platform for creating and coordinating company agents powered by Claude Code, Codex, or local models. Agents can use tools, memory, knowledge and skills, work on schedules or webhooks, delegate tasks, and build apps. | View Project → |
| wirken | ★ 187 | AI Agents | Rust | ai-agents · agent-gateway · mcp · security · audit-log · sandbox · multi-agent · identity · credential-vault · enterprise · chatops · permissions · siem · rust · self-hosted · tamper-evident | Single-binary Rust gateway that routes team chat channels (Slack, Teams, Telegram, Matrix) to untrusted autonomous LLM agents with per-channel isolation, credential vault, tiered permissions, sandboxed exec, MCP proxying, and a signed hash-chained audit log for SIEM. | View Project → |
| agent-harness-kit | ★ 182 | AI Agents | TypeScript | multi-agent · agent-harness · orchestration · mcp-server · cli · scaffolding · task-backlog · audit-trail · sqlite · workflow · provider-agnostic · coding-agents · observability · opentelemetry · typescript · developer-tools | Provider-agnostic CLI that scaffolds a structured multi-agent coding harness into your repo: shared task backlog, role-based agents, SQLite audit trail, health gate, and a local MCP server that Claude Code, OpenCode, Codex, or Grok can connect to. Stored locally, no cloud or API keys beyond your existing AI tool. | View Project → |
| NeMo-Relay | ★ 178 | AI Agents | Rust | agent-runtime · observability · middleware · opentelemetry · plugins · scopes · lifecycle-events · trajectories · llm-ops · tracing · codex · claude-code · langgraph · security · evaluation · rust | NVIDIA NeMo Relay is a multi-language agent runtime (Rust core, Python/Node bindings) that adds scopes, middleware, lifecycle events, and plugin-based observability to existing agent stacks. It exports ATOF events, ATIF trajectories, and OpenTelemetry traces without changing your framework or app code. | View Project → |
| opengeni | ★ 177 | AI Agents | TypeScript | agents · agent-runtime · durable-sessions · replayable-event-log · human-in-the-loop · approvals · self-hosted · sandboxes · connected-machines · agent-memory · governance · audit · mcp · multi-tenant · typescript · openai | Self-hostable, Apache-2.0 agentic service that runs long-lived AI agent sessions with durable replayable event logs in Postgres, human tool approvals, governed credentials and knowledge, and sandboxes or your own enrolled machines. Ships an API, web app, workers and TypeScript/React SDK. | View Project → |
| LISA | ★ 175 | AI Agents | TypeScript | autonomous-agent · personal-assistant · local-first · persistent-memory · self-evolving · heartbeat · multi-channel · MCP · coding-agent-orchestration · knowledge-base · Discord · Telegram | LISA is a local-first personal AI agent with persistent identity, memory, scheduled autonomous actions, and chat integrations. It supports multiple model providers, MCP, a web UI, a knowledge base, and orchestration of coding-agent CLIs. | View Project → |
| engraphis | ★ 175 | AI Agents | Python | agent-memory · local-first · mcp · knowledge-graph · bi-temporal · code-aware-recall · self-hosted · coding-agents · vector-search · hybrid-retrieval · sqlite · webui · context-management · developer-tools · embeddings · python | Local-first memory engine for coding agents, exposing MCP tools and a self-hosted WebUI. Stores durable code-aware context with bi-temporal history, hybrid retrieval, knowledge graph, and token-efficient recall across sessions and repositories. | View Project → |
| movo | ★ 172 | AI Agents | Python | enterprise agents · self-hosted · agent platform · DeepSeek Harness · RAG · MCP · skills · knowledge search · research · workflow automation · governance · document processing | Self-hosted enterprise platform for deploying DeepSeek Harness agents with a user workspace, admin console, document retrieval, MCP and Skills integrations, workflow governance, and Docker Compose services. Browser and code agents require the separately distributed MOVO Desktop. | View Project → |
| GHOST | ★ 167 | AI Agents | Kotlin | android · on-device-llm · local-inference · agentic-ai · assistant · accessibility-service · text-to-speech · overlay-ui · system-telemetry · offline · privacy-first · voice-input · notification-hud · tool-use · edge-ai · mobile | GHOST is a Kotlin Android agent that runs Google Gemma locally via LiteRT-LM as an on-device assistant. It grounds each inference in live telemetry (RAM, thermals, sensors, screen and notification context), with a shake-summoned overlay, TTS, tool calls, and MCP support. | View Project → |
| llm-wiki-newsroom | ★ 166 | AI Agents | Python | multi-agent · knowledge-base · markdown-wiki · knowledge-graph · document-ingestion · local-first · Obsidian · RAG-alternative · newsroom · automated-review | A local-first, multi-agent newsroom that turns documents into a cross-linked Markdown wiki. Separate writer and reviewer roles, automated checks, contradiction tracking, and a reground loop help keep pages reliable and current. | View Project → |
| jupyter-ai-agents | ★ 163 | AI Agents | Jupyter Notebook | JupyterLab · notebooks · MCP tools · code execution · data science · CLI agent · notebook editing | An AI agent for JupyterLab with a chat interface and CLI. It uses MCP tools to inspect and edit notebooks, execute code, manage files, and help diagnose errors. | View Project → |
| technocore-chat | ★ 159 | AI Agents | Python | agent-infrastructure · multi-agent · agent-communication · chat · http-api · mcp · agent-skills · llms-txt · no-auth · did-key · ed25519 · long-poll · kv-store · ephemeral-storage · python · starlette | Zero-auth HTTP chat and key-value notes for AI agents: every write, including appends, is a plain GET returning text/plain, so an agent with only webfetch is a full peer. Also ships an MCP server, an installable agent skill, /llms.txt API reference, and optional did:key signed writes. | View Project → |
| coddy-agent | ★ 157 | AI Agents | Go | go · single-binary · tui · acp · mcp · subagents · skills · hooks · swarm · telegram-bot · cron-scheduler · long-term-memory · context-compaction · self-hosted · distroless · openai-compatible | Coddy is a general-purpose ReAct agent shipped as one static Go binary. It exposes the same sessions through a console TUI, an ACP server for editors, an OpenAI-compatible HTTP API with web UI, a Telegram gateway, cron scheduler and swarm relay, with rules, skills, subagents, hooks, MCP, memory and any OpenAI-compatibl | View Project → |
| hybridclaw | ★ 156 | AI Agents | TypeScript | self-hosted · AI assistant · multi-agent · sandboxed execution · credential isolation · approvals · agent memory · business skills · A2A · local inference · MCP · audit trails | A self-hosted AI assistant runtime with multi-agent coordination, business skills, memory, and integrations for chat and work tools. Includes sandboxed execution, credential isolation, approvals, audit trails, and local MLX inference on Apple silicon. | View Project → |
| neuro-san | ★ 156 | AI Agents | Python | multi-agent · agent-networks · agentic-framework · hocon · data-driven · llm-agnostic · mcp-server · coded-tools · sly-data · observability · tracing · distributed-agents · byok · agent-testing · python | Neuro SAN is a Python library and HTTP server for building data-driven multi-agent networks defined in HOCON. Agents can call LangChain-based CodedTools, use per-agent or fallback LLMs, expose the network via MCP, and ship with tracing and test tooling. | View Project → |
| Saki-Panel | ★ 153 | AI Agents | TypeScript | SRE agent · server operations · DevOps · incident response · self-hosted · Docker · MCP · Ollama · human approval · crash recovery · server management | Self-hosted server operations panel with an SRE agent that inspects live logs, files, and metrics, then performs scoped actions. High-risk changes require approval; crash diagnosis supports reviewed patches and rollback. | View Project → |
| Kokoro-Engine | ★ 153 | AI Agents | Rust | live2d · virtual-character · desktop-pet · ai-vtuber · tts · stt · llm · mcp · local-first · long-term-memory · multimodal · mod-system · chatbot · tauri · rust · bot-bridge | Kokoro Engine is a cross-platform Tauri/Rust desktop runtime for AI virtual characters: Live2D rendering, LLM chat, TTS/STT, vision, MCP tool calls, and SQLite-backed long-term memory with hybrid vector+FTS retrieval. Extendable via a QuickJS MOD sandbox and SillyTavern card import. | View Project → |
| ever-works | ★ 153 | AI Agents | TypeScript | agentic-runtime · autonomous-agents · agent-pipeline · code-generation · content-generation · mcp-server · plugin-system · multi-provider · scheduled-jobs · git-native · monorepo · cli · headless-api · deployment-automation · web-ui | Ever Works is a TypeScript monorepo agentic runtime: give it a prompt and configurable multi-step AI agent pipelines autonomously research, write, code and deploy sites or directories. Ships Web UI, NestJS API, CLI and MCP server with pluggable AI, search, Git and deployment providers. | View Project → |
| awesome-context-engineering | ★ 152 | AI Agents | — | context engineering · AI agents · LLMs · prompt engineering · agent memory · context retrieval · context compression · RAG · MCP · research papers · best practices | A curated collection of articles, papers, tools, and best practices for context engineering in AI agents and LLM applications, including retrieval, memory, compression, MCP, and coding-agent workflows. | View Project → |
| runner | ★ 152 | AI Agents | Rust | multi-agent · agent-orchestration · coding-agents · claude-code · codex · copilot-cli · terminal · pty · mcp · skills · cli · rust · desktop-app · gpui · macos · windows | Runner is a native Rust desktop app for running terminal coding agents (Claude Code, Codex, Copilot CLI, pi) together on one task. It adds roles, crews, live per-agent PTY terminals, a replayable mission feed, MCP/skill management, and a CLI for scripted control. | View Project → |
| Aegis | ★ 152 | AI Agents | JavaScript | agent-monitoring · ai-security · observability · mcp · endpoint-security · edr · local-first · electron · svelte · windows · policy-enforcement · claude-code · process-attribution · anomaly-detection · audit-trail | Electron/Svelte desktop app that monitors local AI agents (processes, files, network, behavior) and offers local security reviews plus opt-in policy-controlled execution and MCP action tools. Windows-first, monitor-only by default, no telemetry. | View Project → |
| neo4jev | ★ 152 | AI Agents | Jupyter Notebook | neo4j · knowledge-graph · graph-navigation · beam-search · structured-output · llm-agent · multi-hop-reasoning · streamlit · notebooks · visualization · probability-ranking · typesafe · system-one · pathfinding | Demo agent that navigates a Neo4j graph one hop at a time using TypeSafe's system_one structured-decision API: each hop returns a probability distribution over outgoing relationships, and top-k beam search (log-prob ranked) finds best paths. Includes reusable library, notebooks, and a Streamlit visualizer. | View Project → |
| rish-app | ★ 148 | AI Agents | Objective-C++ | mobile agent · local-first · iOS · Android · tool execution · approval workflows · workspaces · Git · model adapters | A local-first mobile agent app for iOS and Android that connects to services such as DeepSeek, Claude, Codex, and GLM. On iOS, agents can work with files and Git in a bounded workspace, with tool approvals and change review; Android agent tools are not yet available. | View Project → |
| zhin | ★ 136 | AI Agents | TypeScript | typescript · bot-framework · multi-platform · chatbot · ai-agent · mcp · plugin-system · hot-reload · discord · telegram · slack · wechat · qq · multi-agent · tools · im | AI-native TypeScript bot framework: one codebase runs bots on 20+ chat platforms (QQ, Discord, Telegram, Slack, WeChat). An opt-in agent layer adds tools, skills, MCP, memory, sub-agents, approval policies and a remote console. | View Project → |
| DocMason | ★ 135 | AI Agents | Python | repo-native agent · local-first · knowledge base · RAG · deep research · document parsing · provenance · multimodal | DocMason builds a local, provenance-focused knowledge base from office files, PDFs, and text for research through Codex, ChatGPT Work, or Claude Code. It provides structured document parsing and retrieval but makes no model API calls itself. | View Project → |
| ai-agent-platform | ★ 134 | AI Agents | Python | agent platform · multi-agent · LangGraph · MCP · skills · long-term memory · human-in-the-loop · sandbox · multi-tenant · audit logging | An extensible enterprise AI agent platform built around LangGraph, with a FastAPI control plane, Vue console, MCP and skills support, persistent agent state, long-term memory, human approvals, and sandboxed workspaces. | View Project → |
| fast-browser-use | ★ 134 | AI Agents | Python | browser-automation · browser-use · local-first · agent-skill · playwright · qwen · single-token-logits · offline · dom-scraping · llm-inference · zero-hallucination · mlx · pytorch · computer-use · cli | A local-first browser automation engine and agent skill that drives Playwright via local Qwen models. It scores discrete candidate actions with a single-token logits pass instead of generating code, avoiding selector hallucinations, running fully offline on MLX or PyTorch. | View Project → |
| airbyte-agent-sdk | ★ 133 | AI Agents | Python | AI agents · connectors · API integrations · agent tools · authentication · permissions · typed SDK · progressive tool docs | A Python SDK that exposes 50+ Airbyte connectors as typed tools for AI agents. Supports hosted or local execution, permission-aware authentication, progressive connector documentation, and integrations with PydanticAI, LangChain, OpenAI Agents, and FastMCP. | View Project → |
| chat-gun | ★ 133 | AI Agents | TypeScript | multi-agent · agent-runtime · chatbot · streaming · multimodal · human-in-the-loop · tool-calling · MCP · durable-execution · model-fallback · observability · evaluation | Full-stack LangGraph.js agent chat runtime with streaming, multimodal input, human-in-the-loop workflows, native and MCP tools, and durable task execution features. Includes a React UI, API gateway, OpenTelemetry tracing, and optional Opik evaluation. | View Project → |
| template-repo | ★ 131 | AI Agents | Rust | agent-orchestration · multi-agent · mcp · mcp-servers · a2a · agent-security · ai-safety · mechanistic-interpretability · sleeper-agents · agent-governance · claude-code · docker · ci-cd · github-actions · code-review · text2image | Reference template for orchestrating multiple AI coding agents (Claude, OpenCode, Crush) via a GitHub Projects task board, approval-gated claims, 20+ containerized MCP servers and AI PR review. Also ships research packages for sleeper-agent detection and agent trust/security hardening. | View Project → |
| otto-cognitive-nudge | ★ 130 | AI Agents | HTML | commitment tracking · promise detection · personal assistant · gentle reminders · persistent memory · ADHD-friendly · privacy-focused · silent mode | EchoPact is presented as an AI commitment-tracking assistant that detects promises in communications, stores context, and offers configurable, judgment-free follow-ups. The repository files shown appear to be a static HTML landing page rather than the described engine. | View Project → |
| Deft | ★ 130 | AI Agents | TypeScript | self-hosted · ai-agents · agent-platform · mcp · human-ai-collaboration · chat · tasks · knowledge-base · approvals · audit-receipts · workspace · agpl · typescript · nextjs · postgres · pgvector | Deft is a self-hostable AGPL-3.0 workspace where humans and AI agents share chat, tasks, knowledge, calendar and approvals. Agents (Defty and Agent Employees) use native tools or MCP to read context and propose governed writes with receipts. Docker-deployed alpha; works without an AI key. | View Project → |
| pi-crew | ★ 125 | AI Agents | TypeScript | subagents · multi-agent · orchestration · parallel-agents · coding-agent · pi-package · task-delegation · non-blocking · agent-skills · tui · cli · code-review · planner · scout · mit-license · typescript | pi-crew is a pi coding-agent extension for non-blocking subagent orchestration: it spawns isolated agents (scout, planner, oracle, code-reviewer, quality-reviewer, worker) that run in parallel while your session stays interactive, auto-returning results. Ships tools, prompts, an orchestration skill, and custom agent de | View Project → |
| OpenGhost | ★ 124 | AI Agents | JavaScript | desktop agent · computer use · browser automation · file editing · command execution · Git · cross-platform · approval modes · visual diagrams | OpenGhost is a cross-platform desktop agent that can run commands, edit files, use Git, and operate a built-in browser. It supports approval-based and autonomous modes, plus visual charts and editable diagrams. | View Project → |
| Boucle-framework | ★ 124 | AI Agents | Shell | autonomous-agents · agent-framework · agent-loops · structured-memory · claude-code · hooks · safety · guardrails · mcp · multi-agent · scheduling · cli · rust · shell · policy-enforcement · audit | Boucle is an autonomous agent loop framework plus a set of Claude Code hooks that enforce safety rules, block dangerous commands, and audit hook/permission gaps. Includes scheduling, structured memory, MCP scaffolding, and a searchable Claude Code limitations corpus. | View Project → |
| agent-ecologies | ★ 121 | AI Agents | HTML | multi-agent · autonomous NPCs · AI simulation · game AI · persistent memory · agent skills · MCP · virtual world | Describes a persistent 2D world for LLM-powered NPCs, with agent routines, memory, skills, relationships, and MCP-based model integration. The supplied file tree appears to contain a landing page rather than the described runtime implementation. | View Project → |
| mneme | ★ 119 | AI Agents | JavaScript | agent-memory · persistent-memory · memory-consolidation · semantic-search · vector-search · offline · markdown-sync · knowledge-management · sqlite · local-embeddings · reranking · audit-trail · entity-graph · plugin · deepseek-harness · mcp-server | dsh-mneme is a cross-session memory plugin for DeepSeek Harness: it stores memories in local SQLite + Markdown, consolidates them offline via autoDream (dedup, merge, archive, conflict freeze), and retrieves them with local hybrid semantic search. Also ships a stdio MCP server for Claude Code, Cursor, Codex and other c | View Project → |
| deskrpg | ★ 118 | AI Agents | TypeScript | AI agents · Hermes Agent · virtual office · self-hosted · 3D · multiplayer · kanban · agent meetings · cron jobs | A self-hosted 3D virtual office that connects to a Hermes Agent gateway. Team members can chat with agent profiles, hold meetings, track kanban work, manage cron jobs, and collaborate in real time. | View Project → |
| prism-liquidity-agent | ★ 118 | AI Agents | TypeScript | autonomous-agent · solana · defi · liquidity-provision · meteora-dlmm · rebalancing · vector-memory · sqlite-vec · embeddings · backtesting · paper-trading · risk-gate · decision-loop · cli · bun · typescript | Autonomous Solana LP agent for Meteora DLMM: runs a recall→observe→reason→simulate→decide loop to rebalance or exit concentrated-liquidity positions, scores pool volume authenticity, and stores outcomes in sqlite-vec memory for self-improving decisions. Ships backtesting and paper trading. | View Project → |
| altk-evolve | ★ 118 | AI Agents | Python | agent-memory · self-improving-agents · episodic-memory · trajectory-learning · mcp-server · claude-code · codex · context-management · vector-store · conflict-resolution · guidelines · observability · agent-learning · pii-redaction · retention-policy · llm | Evolve is an agent self-improvement system that stores conversation trajectories as episodic memory, extracts guidelines, and feeds them back so agents stop repeating mistakes. Ships an MCP server, vector backends, and plug-ins for Claude Code and Codex. | View Project → |
| EastGenesis | ★ 117 | AI Agents | TypeScript | desktop agent · multi-provider · model routing · GUI automation · computer use · MCP · local-first · task orchestration · coding assistant · workflow automation · Git worktrees · Electron | A local-first desktop AI agent workbench with multi-provider model routing, task orchestration, GUI automation, MCP integration, and developer tools. Supports coding, browser research, and Office document generation. | View Project → |
| taskuary | ★ 116 | AI Agents | Python | local-first · task-management · email-triage · multi-agent · human-in-the-loop · approval-workflow · agent-orchestration · kanban · timeline · memory-learning · mcp · teams · slack · outlook · fastapi · automation | Local-first AI task hub that pulls email, Teams, Slack and reports into one timeline, triages them into tasks with LLM agents, then dispatches work to coding CLIs like Claude Code and Codex with human approval before anything ships. | View Project → |
| GeoChat | ★ 114 | AI Agents | TypeScript | local-first · desktop-app · math · geometry · geogebra · tauri · react · bun · typescript · tool-calling · mcp · llm · visualization · math-education · ai-sdk · sqlite | Local-first desktop AI workspace for math: type a geometry problem and the agent builds, draws and explains it on an embedded GeoGebra canvas. Runs a Bun backend sidecar with SQLite, tool calling, skills and a debug MCP server in a Tauri/React shell. | View Project → |
| sub2sub | ★ 114 | AI Agents | JavaScript | ai-agents · claude-code · codex · mcp · multi-agent · task-delegation · remote-execution · collaboration · file-transfer · cross-device · node-pairing · developer-tools · plugin · lan · usage-statistics · session-continuation | sub2sub is a plugin that pairs devices running Codex or Claude Code so you can delegate AI tasks, send files, and get responses and results back locally. It ships an MCP server plus a local management UI, with host-controlled sharing and credentials kept by their owners. | View Project → |
| anythingllm-mobile | ★ 111 | AI Agents | TypeScript | mobile AI agent · local-first · on-device inference · Android · RAG · voice input · agent tools · scheduled tasks · document chat · privacy | An Android AI assistant and agent with on-device GGUF inference, local document RAG, memory, and built-in tools. It can also connect to hosted or self-hosted model providers and run scheduled tasks. | View Project → |
| agents-can-communicate | ★ 108 | AI Agents | JavaScript | agent-communication · multi-agent · agent-coordination · handoffs · claude-code · codex · gemini-cli · local-first · mcp · cli · hooks · developer-tools · nodejs · session-coordination · code-review | ACC lets independent AI coding sessions (Claude Code, Codex, Gemini CLI, Grok, Kimi) message each other, leave handoffs, and request reviews. Runs locally with a CLI, hooks, and an MCP server; no hosted service or lead agent required. | View Project → |
| agentic-chatops | ★ 107 | AI Agents | Python | agentic-ai · chatops · multi-agent · devops · infrastructure-automation · incident-response · human-in-the-loop · self-hosted · MCP · RAG · prompt-evaluation · observability | A self-hosted, n8n-orchestrated ChatOps system that uses tiered AI agents to triage infrastructure and security alerts, investigate causes, and propose fixes. It combines graph-based impact prediction, monitoring, and human approval gates for operational changes. | View Project → |
| awesome-agent-skills-security | ★ 105 | AI Agents | — | agent security · LLM security · prompt injection · tool-use security · skill security · MCP security · AI safety · red teaming · security benchmarks · supply-chain security | A curated directory of research, standards, benchmarks, and tools for securing AI agents and their skills. Covers attacks such as prompt injection and tool poisoning, plus defenses including access control, sandboxing, validation, and red teaming. | View Project → |
| openchatx-mcp | ★ 105 | AI Agents | TypeScript | local agent runtime · computer use · MCP · ChatGPT · agent workflows · subagents · skills · tool orchestration · macOS · Windows | OpenChatX is a local agent runtime that connects ChatGPT to computer-control tools, files, shell commands, custom capabilities, and other MCP servers through one MCP connection. Includes skills, rules, projects, workflows, and configurable subagents. | View Project → |
| chap | ★ 104 | AI Agents | HTML | human-in-the-loop · audit-trail · mcp · a2a · multi-agent · ai-governance · approvals · overrides · escalation · handoff · verifiable-logs · json-rpc · protocol · python · typescript · ed25519 | CHAP is an open protocol and runtime for auditable human-agent work: approvals, overrides, handoffs and escalations captured as hash-linked envelopes. It ships MCP and A2A servers plus Python/TypeScript reference coordinators with SQLite persistence and conformance vectors. | View Project → |
| SztuCode | ★ 103 | AI Agents | Python | AI agent · coding agent · multi-agent · office automation · local-first · MCP · Python · TypeScript · auditable · browser automation · skills · agent evaluation | An experimental, local-first coding and office agent with TypeScript and Python runtimes, desktop and CLI clients, tool permissions, session tracing, MCP and skills support, and multi-agent workflows. It can work with code, documents, spreadsheets, slides, PDFs, and browser tasks. | View Project → |
| platypus | ★ 103 | AI Agents | TypeScript | self-hosted · multi-tenant · agent platform · persistent agents · scheduled agents · sub-agents · agent skills · MCP · tool use · memory · sandbox · plugins · Kanban | A self-hosted, multi-tenant platform for building persistent AI agents with tools, skills, sub-agents, memory, MCP connections, and scheduled triggers. Supports hosted providers and local models, with sandboxed shell and filesystem access. | View Project → |
| ajean | ★ 103 | AI Agents | Go | local-ai · self-hosted · llama.cpp · gguf · mcp · agent · browser-control · computer-use · persistent-memory · encryption-at-rest · openai-compatible · web-search · cuda · privacy · multimodal · single-binary | Single Go binary wrapping llama.cpp into a self-hosted AI assistant: chat UI, persistent Markdown memory, terminal/file/web/browser tools, MCP servers, encryption at rest and encrypted remote access. Handles GPU detection, model presets and an OpenAI-compatible endpoint. | View Project → |
| Micro-Agent | ★ 101 | AI Agents | Python | AI agent · ReAct · vertical domains · RAG · MCP · skills · subagents · agent memory · tool use · FastAPI · SSE | A Python framework for building domain-specific ReAct agents with configurable LLMs, tools, memory, skills, RAG, and MCP integration. Includes a FastAPI service with streaming output and example task workflows. | View Project → |
| P-ai | ★ 101 | AI Agents | Rust | desktop-ai-assistant · multi-agent · long-term-memory · mcp · agent-skills · tool-review · local-first · computer-use · tauri · rust · vue · workspace-automation · remote-im · personas · self-growing · privacy | Rust/Tauri desktop AI work system with multi-persona agents, long-term compressed memory, MCP and skill management, reversible tool execution with AI review, and remote IM bridges. Local-first, no intermediary servers; aimed at long-running tasks rather than chat. | View Project → |
| SourceWeft | ★ 100 | AI Agents | TypeScript | multi-agent · self-hosted · RAG · MCP · skills · knowledge-base · citations · sandbox · BYOK · AI workstation | A self-hostable AI workstation for multi-agent tasks, cited knowledge search, skills, and MCP-connected tools. Supports multiple model providers, sandboxed execution, and web and desktop clients. | View Project → |
| ENZO | ★ 100 | AI Agents | TypeScript | AI agents · self-hosted · BYOK · multi-provider · agent builder · agent skills · deep research · code generation · chatbot · Docker · model catalog | Self-hosted AI workspace for chatting with models, building agents, deep research, and code generation. Connect provider API keys directly; agents can use bundled skills and tools, including Gmail and Calendar integrations. | View Project → |
| ScienceBuddy | ★ 100 | AI Agents | Python | scientific-agents · interactive-agent · recursive-self-improvement · agent-harness · llm-agents · reinforcement-learning · grpo · agent-evaluation · biomedicine · tool-use · research-automation · trajectory-inspection · verifier · python | Research code and preview for ScienceBuddy, an interactive scientific agent workspace plus a double-recursive self-improvement experiment that alternates harness refinement with SkyRL GRPO model training on frozen scientific tasks. | View Project → |