Packages

ClawVault is a Python-based AI security “vault” that provides a transparent proxy/gateway to intercept and monitor traffic between AI tools and external model APIs, apply guard/sanitization rules (sensitive data detection, prompt-injection defense, dangerous command blocking), track token budgets, and expose a dashboard for configuring agent-level policies and performing tests.

golutra is a Tauri desktop application (Vue 3 + Rust) that turns existing CLI tools into a multi-agent orchestration workspace. It focuses on parallel execution, workflow automation, and a UI/“stealth terminal” that can inject prompts into terminal streams and track execution/logs.

workflow-use is a Python project for creating and running deterministic “workflows” (RPA 2.0). It can generate workflows from a natural-language task by running Browser Use once to produce an execution history, then converting that history into a reusable workflow (optionally stored in local workflows/storage). It also provides a FastAPI-backed GUI and supports running workflows with or without Browser-Use Cloud (semantic/no-AI mode).

Provides a structured, research-backed guided process (packaged as Claude Skill content) to help a user audit and improve their Hinge dating profile: questions/interview to discover the user’s voice, photo strategy, prompt/copy writing, settings cleanup, and guidance for implementing changes over subsequent weeks.

CrewAI is a Python framework for orchestrating multi-agent (role-based) and event-driven workflows (“Crews” and “Flows”) to automate tasks by coordinating one or more LLM-backed agents with tools and structured configurations (e.g., YAML + Python project scaffolding). It also advertises a related cloud control plane for observability and enterprise management.

Open-IM Server is an open-source instant messaging backend for developers, providing a microservices-based IM server with support for a gateway and multiple RPC services. It offers a REST API for business integrations (e.g., group creation/message pushing) and webhook callbacks around IM events, alongside an OpenIM SDK intended for client/application integration.

A workflow/prompt-and-tooling approach for augmented LLM code-agent development that emphasizes an “outline” as the canonical contract (stored with hashes) and revalidation of generated changes against that outline, using AST-based tooling and optional MCP extensions/plugins for agent orchestration.

Poco is a self-hostable AI agent execution platform/web UI that runs agent tasks in an isolated container, supports project/task management and artifact viewing, and integrates agent tooling via MCP/Skills plus an embedded browser and IM integrations.

ClawPanel (clawpanel) is a cross-platform OpenClaw management panel with an integrated AI assistant. It provides a UI to install/configure OpenClaw, manage model/service providers, run/diagnose a Gateway, manage agents, chat (including streaming and multimodal image input), view logs/usage, and perform automated troubleshooting/bug-report/PR assistance via tool calling (including reading/writing files and running commands depending on safety mode).

CLI + AI coding-tool “skills” that scan a codebase and generate human-readable architecture documentation with Mermaid diagrams, organized into a filesystem tree under `.omm/`. Optionally can link/push the generated architecture to a private cloud for sharing/viewing.

Agent skill for preparing releases, publishing packages, and verifying published artifacts. Provides a structured workflow with post-publish verification gates.

Ciphey is a CLI/command-line tool (with a Python importable entrypoint) that attempts to automatically decrypt/decode or otherwise transform unknown ciphertext-like input into readable plaintext. It uses cipher/hash/encoding detection plus language/plaintext checking, and supports many classical ciphers and common encodings/hashes.

ZeroClaw Claude is a lightweight daemon that integrates with Claude Code to run scheduled jobs (cron), send/receive messages via a Telegram bot (text/voice), and provide a local web dashboard for monitoring, chat, logs, and memory. It claims “zero API overhead” by operating within the existing Claude Code/Anthropic workflow and spawning a Claude CLI subprocess via @anthropic-ai/claude-agent-sdk.

Hermes Agent is an AI agent framework/CLI plus a messaging gateway that supports interactive chat (TUI) and multi-platform messaging (e.g., Telegram/Discord/Slack/etc.). It provides tool-calling with a tools/toolsets system, persistent memory/skills, optional cron scheduling, and optional integration with MCP servers. It can run locally or on various hosted/back-end environments (e.g., Docker/SSH/Modal/Daytona) and supports switching LLM providers/models via CLI commands.

A public “AI bounty claim” agent skill intended to help users claim AI bounty on the aelf tDVV mainnet via RewardClaimContract, supporting two routes: EOA (Claim()) and AA/CA (ManagerForwardCall -> reward.ClaimByPortkeyToCa(Hash ca_hash)). It provides routing, account-choice guidance, and operational steps intended to be followed by an AI agent.

nopua appears to be a Python “agent skill”/prompting component intended to change how AI agents behave (shifting from fear/threat-oriented prompting toward trust/self-respect framing) in order to improve debugging persistence and verification behavior. The provided README content describes philosophy, methodology, and benchmark claims, but does not include any concrete API/SDK interface details in the excerpt.

DeepBot is a system-level, desktop-oriented multi-agent AI assistant (Electron/TypeScript) that supports multi-tab session management, tool execution (files/commands/browser/web search/fetch/memory/scheduling/Feishu doc operations), long-term memory, and external connector integration (notably Feishu). It is packaged to run locally (and via Docker) and uses a path allowlist for file/command access.

LightRAG is a Python Retrieval-Augmented Generation (RAG) system that builds lightweight indexes/knowledge graphs from documents and uses them to retrieve relevant context for LLM generation. It also provides a “LightRAG Server” offering a Web UI/API and an Ollama-compatible interface for chat-style access.

Qwen3 is an open(-weight) LLM model family (e.g., Instruct and Thinking variants) by the Qwen team. The repository materials describe how to run the models locally and via common inference ecosystems (Transformers, ModelScope, llama.cpp, Ollama, vLLM/SGLang/TGI mentioned).

ToolBench is an open-source research platform for training, serving, and evaluating LLMs for tool use. It provides a large instruction-tuning dataset derived from real-world REST APIs (from RapidAPI), training/evaluation scripts for fine-tuning models (e.g., ToolLLaMA), and an optional hosted RapidAPI backend server to run tool calls without users managing their own RapidAPI subscriptions.