Agent infrastructure

Turns local machines into a self-hosted personal cloud OS for reclaiming data ownership while running AI agents and models for individual users.

An open-source AI gateway that lets developers call more than 100 language models through familiar SDKs, with built-in failover, load balancing, cost controls, and end-to-end tracing.

Enterprise-oriented self-hosted registry for AI agent skills, supporting skill package publishing, version management, RBAC governance, audit logging, and on-premise deployment via Docker or Kubernetes.

Provides an official multimodal gateway and model catalog with OpenAI-compatible interfaces for text, image, video, and agent workflows, plus examples and documentation for developers integrating Agnes AI models.

Observability and enforcement layer for AI agent harnesses capturing every run, auditing behavior, enforcing 40 built-in policies, blocking risky tool calls, and offering a local dashboard plus cloud option.

A CLI that hooks into Git workflows to capture and index AI agent coding sessions alongside commits. Creates a searchable history for developers tracing how code was produced and restoring prior work.

Gives cross-functional AI teams a workbench for building and improving AI systems with evals, prompt optimization, RAG, fine-tuning, and synthetic data management.

A decentralized AI node runtime that installs locally with one command, serving local models with vector retrieval and exposing a chat-accessible distributed agent service.

Operates as an open-source proxy gateway for agent traffic, unifying governance, security, and observability across LLM, MCP, and A2A protocols.

Lightweight desktop app for developers using multiple coding assistants, centralizing management, syncing, and organization of AI agent skills across more than fifty tools including Claude Code and Cursor.

A command-line tool for teams to centrally manage shared skills, rules, knowledge, and MCP configurations, enabling experience synchronization and collaborative improvement across multiple AI agents.

Consolidates New-API and Sub2API relay accounts into a browser extension dashboard, helping users monitor balances, compare prices, automate check-ins, and manage channel health.

Local inference server optimized for consumer hardware that profiles available resources, recommends suitable models, and connects to existing coding agents such as OpenCode and Claude Code.

Open-source AI gateway and registry unifying MCP, A2A, and REST or gRPC APIs behind one endpoint, providing centralized discovery, guardrails, observability, and plugin-extensible agent tool management.

Inference setup for running large language models locally on microcontrollers, storing models in flash memory to demonstrate offline low-resource text generation demos.

Provides a secure MCP-to-OpenAPI proxy for developers exposing Model Context Protocol tools as documented HTTP endpoints, with authentication and unified hosting for multiple tool servers.

Enterprise AI platform centralizing MCP management with integrated gateway, skill registry, orchestration, sandbox execution, and policy guardrails for controlled deployment.

Provides an open-source observability and APM platform for engineers, collecting OpenTelemetry traces, metrics, and logs with dashboards, alerts, and fast query analysis.

VS Code extension for developers that analyzes local AI coding session logs and presents progress trends, anti-patterns, and skill recommendations in a dashboard.

Zero-setup sandbox for AI agents providing secure, multiplexed execution paths with least-privilege isolation and near-zero added latency across reusable terminals.

Free open-source visual debugging interface for Claude Code, helping developers inspect session logs, tool calls, token usage, subagents, and context window consumption.

Dependency management infrastructure for coding agents that standardizes declaring, resolving, and installing skills, plugins, and services across different agent platforms.

Cross-platform visual manager for OpenAI Codex desktop and CLI, unifying provider switching, session sync, prompt injection, Skills and MCP management, plus visual TOML configuration.

Local router connecting Codex to external models including guided Kimi OAuth/API and DeepSeek, providing assisted setup, safe migration, and rollback management for developers.

Self-hosted management panel and observability dashboard for CPA / CLIProxyAPI gateways, giving operators unified visibility into requests, usage, cost, quotas, failures, and account health with automation.

Companion utility that synchronizes Codex session files with provider metadata in the index, offering desktop, web, and CLI management for developers maintaining provider configurations.

Open-source sandboxed agent harness for teams, providing each employee with a secured personal agent under centralized gateway credential injection and enforced access policies.

Distributed platform runs isolated agent environments at scale for reinforcement learning research, offering millisecond snapshot forking, scaling, and lifecycle management for ML infrastructure teams.

Distributed LLM grid that pools crowdsourced compute and VRAM into unified serving, automatically routing or splitting very large model inference to power agents and chat.

Acts as an identity and access-control plane for AI agents, centralizing authentication, permission assignment, and audit logging across more than 1200 MCP, API, and CLI integrations.

A native macOS menu-bar application that routes existing Claude Code and ChatGPT subscriptions to AI coding tools, handling authentication, token management, and model routing without separate API keys.

TUI and web manager for parallel AI coding-agent sessions, using tmux persistence and workspace isolation to support Claude Code, OpenCode, Codex, Gemini, and others.

A lightweight local-first desktop client that bridges OpenClaw agents to WeChat, Feishu, Slack, and Discord, supporting Claude Code, Codex, other LLMs, BYOK, OAuth, and phone-based remote chat.

One-click installer and environment manager for developers using multiple coding agents, simplifying setup, deployment, and configuration of supported AI programming assistants.

A general verification framework that scores agent trajectories for best-of-N selection, progress tracking, and test-time scaling.

Fork of llama.cpp focused on local LLM inference, adding state-of-the-art quantization formats and CPU performance optimizations for faster execution on consumer hardware.

A containerized toolkit that spins up a complete local LLM stack with one command.

Provides a multi-account load-balancing proxy for Codex and ChatGPT with usage tracking, key rate limiting, dashboards, and OpenCode-compatible gateway endpoints for teams managing shared quotas.

LLM traffic router directing requests across models and providers while preserving OpenAI- and Anthropic-compatible APIs for benchmarking and cost-performance optimization.

Offers a local-first task board and CLI for Codex workflows that embeds agent task handoff and lifecycle management into development tools for automation builders.

An enterprise-grade MCP and agent gateway connecting AI copilots and autonomous agents to company tools through unified secure access, permission controls, and full observability.

A security scanner for local AI agent components, MCP servers, and agent skills that detects misconfigurations, risky permissions, secrets exposure, and prompt-injection vulnerabilities before deployment.

A secure, high-density runtime for running millions of AI agent sandboxes with fast suspend, resume, and isolation.

A local gateway that routes Cursor requests through user-supplied model API keys while preserving agentic tool-call workflows, designed for developers wanting flexible model selection without losing native IDE integration.

Local registry and analytics platform for AI components, managing and sharing Skills, MCPs, and Agents within defined scopes. Provides observability and governance for running AI assets.

Local registry and analytics platform for AI components, letting developers define, share, and observe Skills, MCP servers, and agents within a scoped environment.

Open-source library routing each query to the most suitable language model, bundled with evaluation benchmarks and a unified interface for flexible selection.

All-in-one pure C++ inference engine powered by ggml for audio models, covering TTS, STT, VAD, voice conversion, and music generation with optimized performance and no Python requirement.

A cross-platform desktop management panel for OpenClaw and Hermes agents, built with Tauri v2, combining one-click installation, a multimodal built-in assistant, and support for eleven languages.

Fast LLM inference server for consumer-grade GPUs that uses speculative decoding and custom kernels to raise throughput while lowering memory use.

An AI gateway that bridges MCP stdio servers to SSE and SSE to stdio. It enables remote access to local MCP tools with a single command.

Lightweight Python tool that cleans AI refusal messages from Codex CLI session files and repairs context, helping developers maintain usable conversation history for coding sessions.

Management tool for switching between multiple Claude Code accounts, providing automatic rate-limit rotation, centralized usage dashboards, and support for parallel sessions.

Efficient long-context LLM serving framework using head-aware KV reuse and SegPagedAttention to accelerate inference for developers deploying memory-intensive language models.

A proxy bridging Streamable HTTP and stdio transports for MCP servers and clients, enabling bidirectional protocol conversion with straightforward configuration and containerized deployment for mixed environments.

A Docker-packaged MCP aggregator, orchestrator, middleware, and gateway. It combines multiple MCP servers under one endpoint with namespacing, routing, and administrative observability.

System of record for team-based agent-assisted software development, consolidating sessions across 20+ CLIs with replayable traces for auditing, review, and collaboration.

Containerized AI coding workstation bundling Claude Code, a web interface, multiple AI command-line tools, a headless browser, and dozens of development utilities for ready-to-use environments.

A unified hub for managing multiple MCP servers and APIs centrally. It offers dynamic orchestration, separate endpoints, flexible routing, authentication, and observability for gateway control.

Dependency-free, embeddable C inference engine for ultra-large mixture-of-experts models, streaming activated Kimi K3 weights from NVMe with paged caching to run beyond available RAM.

MacBook notch status panel showing real-time activity for thirteen AI coding tools. It supports approval handling, replies, and quick jumps, with companion apps for iPhone and Apple Watch.

A fast Apple Silicon inference engine that runs local LLMs with exact speculative decoding behind an OpenAI-compatible API.

A desktop console for managing the CLIProxyAPI core, handling providers, auth, routing, usage analytics, and wiring popular coding agents to local endpoints.

Lightweight hardware-isolated micro-VM substrate designed for AI agents, embeddable on laptops and elastically scalable for agentic clouds. Provides isolated execution for agent-generated code.

A high-performance C++ and CUDA inference engine for single-GPU execution, optimized for Qwen checkpoints and RTX 5090 hardware across text, image, and video with a compatible API.

Operates a unified gateway and directory for agent tools, allowing agent developers to access thousands of APIs through metered proxying with centralized authentication and team key management.

Local-first desktop widget tracking token usage, costs and limits across more than 37 AI coding tools, including Claude Code, Codex, Cursor and OpenCode, with multi-device synchronization.

Centralized registry and gateway for AI-agent tools, hosting shared skills and secrets server-side with metered upstream relay so agents consume tools without managing keys.

Open-source HTTP credential proxy and vault for AI agents that intercepts outgoing requests, injects stored secrets, and enforces access controls and audit logging to prevent key leakage.

Lightweight MCP gateway service and management UI that converts existing MCP servers and APIs into standard MCP endpoints without code changes. Supports Docker deployment for agent infrastructure.

A testing and debugging platform for MCP servers, MCP apps, and ChatGPT apps. It lets developers chat with servers, inspect traces across clients, and evaluate behavior to improve release quality.

Single-file distribution format and packaging specification for MCP servers, with CLI tooling and desktop loading validation. Enables one-click local installation in desktop applications.

Comprehensive document-parsing benchmark with 1,651 pages of annotated real-world data plus end-to-end evaluation code for researchers evaluating layout analysis, OCR, and structured extraction systems.

Open-source end-to-end platform for evaluating, tracing, and improving LLM and agent applications with evals, simulations, datasets, gateway, guardrails, and self-hosting under Apache 2.0.

An open-source menu-bar companion for heavy coding-agent users, monitoring multiple Claude Code, Codex, and OpenCode sessions with approval alerts and one-click terminal switching.

Local-first engine for managing multiple coding agents under unified control. It handles session storage, background synchronization, and cross-device follow-along for developers working across machines.

Population-scale persona simulation infrastructure for evaluating AI systems and products. Heterogeneous simulated users let product teams test behavior and gather feedback before real-world release.

C-based speech-to-text inference library built on GGML supporting more than sixteen model families, designed for efficient streaming and batch transcription workloads.

Adds safety-first guardrails for AI-driven cloud and Kubernetes operations, scoring and blocking risky configurations while constraining autonomous agents from unauthorized privileged actions.

Gateway service for developers that converts Grok web capabilities into an OpenAI-compatible API, handling proxying and verification so existing clients and tools can connect easily.

Management console for the Null agent ecosystem, centralizing installation, configuration, monitoring, workflow orchestration, task pipelines, system health dashboards, and updates.

An ultra-low-latency LLM inference runtime using tile-level scheduling and speculative decoding for giant models.

Pooled worktree manager for agent coding sessions that provides instant isolated environments while reusing dependency builds and caches without requiring manual worktree handling.

Porting solution for running OpenClaw agents on Android without Linux or proot, requiring only linker installation. Includes an all-in-one terminal app package for mobile deployment.

Benchmark for frontier coding agents built around original long-horizon engineering tasks, using isolated environments and programmatic verifiers for objective scoring.

Local-first, zero-configuration dashboard that tracks token usage and costs across more than twenty-five AI coding tools, including Claude Code, Cursor, Copilot, and Gemini. Offers a macOS menu-bar app, desktop widgets, and detailed analytics for individual developers.

A local-first observability dashboard that tracks token usage, cost, and rate limits across dozens of AI coding tools, with native desktop apps and widgets.

A real-time observability interface for Claude Code orchestration that visualizes agent branches and coordination through node graphs, timelines, and full execution transcripts for developers debugging multi-agent runs.

A tuned vLLM serving stack that runs Qwen 3.8 27B fast on a single RTX 3090 with long context and an OpenAI-compatible API.

Enterprise security system for AI agents deployed at Uber, providing asset discovery, behavioral auditing, security benchmarking, observability, and threat detection with response.

Local gateway turning Codex into OpenAI- and Ollama-compatible APIs so existing chat and coding tools can call its models directly.

Manages session resume and context handoff across sixteen AI coding tools, including Claude Code, Copilot, Gemini, Codex, and Cursor, for developers switching environments.

A bridging daemon linking instant-messaging platforms to AI coding CLIs, mapping each chat topic to a dedicated CLI session with live streamed messages.

Optimized quantization and inference library for running large language models locally on consumer GPUs, supporting parallel execution, offloading, and open APIs.

An integration exposing DeepSeek V4 models inside the Copilot Chat model picker while preserving agent mode and tool calling, with API keys stored in the system keychain.

Lossless compression gateway for AI coding agents that compresses tool output and history to reduce token usage, while keeping original text retrievable and supporting drop-in BASE_URL integration.

Open-source observability platform aggregating traces, logs, and metrics with grouped alerting, using AI agents to help diagnose issues and support self-healing workflows.

Deployment recipe for serving DeepSeek-V4-Flash multimodal inference on two DGX Spark machines, including container images, weights, and networking optimizations for practitioners.

Provides a benchmark for evaluating and improving AI agents on legal work, combining realistic task datasets with execution harnesses and scoring scaffolds for professional workflows.

Native iPhone chat client for remotely operating a self-hosted Hermes agent, providing conversation access and session management for users away from their desktops.