With 34,000+ GitHub stars and a v1.5 release driven by 36 merged community pull requests, DeepTutor from Hong Kong University's Data Science Lab delivers a full agent-native learning workspace that goes far beyond chatbot wrappers. Eight integrated surfaces — Chat, Deep Solve, Quiz Generation, Deep Research, Math Animator, Co-Writer, Book generation, and Mastery Practice — share a unified context so the objective follows the learner, not the tool. The platform's three-layer memory architecture (L1 working, L2 session, L3 long-term) makes personalization inspectable rather than opaque, letting users see exactly what the system remembers and why. Knowledge retrieval operates across five pluggable engines — LlamaIndex with FAISS vectors, PageIndex for page-level citations, GraphRAG for knowledge-graph traversal, LightRAG for local or server-offloaded retrieval, and linked Obsidian vaults — with document parsing via MinerU, Docling, markitdown, or PyMuPDF4LLM. Partners extend the tutoring brain to 15+ messaging platforms including Slack, Discord, Telegram, Matrix with E2EE, and Mattermost, each carrying private memory with branch, resume, and replay capabilities. Subagent integration brings Claude Code, Codex, Gemini, and Kimi directly into learning sessions. The system supports 30+ LLM providers from OpenAI and Anthropic to Ollama for fully local operation, with multi-user isolation, admin controls, and a full CLI interface. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache 2.0 licensed.

DeepSeek Harness gained over 60,000 GitHub stars within hours of its August 2026 launch, establishing itself as the first fully modular open-source agent runtime where literally every component is a swappable plugin. Built on the Cordis framework—a programming paradigm for spatiotemporal composability—dsh decomposes the entire agent stack into independently replaceable pieces: model adapters for DeepSeek, Anthropic, OpenAI, AWS Bedrock, Azure, and Google Gemini; tool registries covering bash execution, file system operations, web search, subagent delegation, and todo management; plus session stores, sandboxes, approval policies, orchestration loops, and the user interface itself. Four operating modes serve different workflows: Standard provides the full toolset, Code mode uses model-generated code to compose multi-round tool calls, Minimal strips down to a shell and editor for benchmarking, and Creator mode lets developers inspect the running runtime and test Cordis plugins in memory. The kernel handles plugin mounting, unmounting, and dependency resolution while typed events and services coordinate between components. Profiles and bundles allow the same codebase to produce entirely different products—a terminal coding agent, a browser-based workspace, a headless automation service, or an ACP/JSON-RPC endpoint—by swapping YAML configuration layers. Session history is stored as an append-only event stream for full trajectory replay, and project-level hooks on agent lifecycle events enable fine-grained behavioral customization. MCP client integration connects to external tool servers, while Agent Client Protocol enables programmatic orchestration. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

FreeLLMAPI collapses the chaos of 29 free LLM providers — Google AI, Cerebras, Groq, Mistral, OpenRouter, GitHub Models, Cohere, Cloudflare Workers AI, NVIDIA NIM, HuggingFace, SiliconFlow, Reka, Z.ai, and more — into a single /v1 endpoint that speaks both OpenAI and Anthropic protocols. The smart router selects the best available model for each request, automatically fails over to the next provider when rate limits hit, and tracks per-key token consumption so you never exceed a free-tier cap. Keys are stored with AES-256-GCM encryption and clients authenticate using a single unified bearer token, never exposing upstream provider credentials to downstream applications. The catalog tracks 251 model families across 358 provider/model endpoints with approximately 4 billion tokens per month of aggregate free-tier capacity, auto-refreshing from a signed manifest at freellmapi.co twice daily without requiring git pulls. Beyond chat completions, the proxy handles embedding, image generation, and audio/TTS endpoints, plus structured outputs with JSON schema forwarding, JSON healing, and format-ignore failover. An integrated MCP server at /mcp provides gateway introspection for coding agents, while the self-hosted OpenAPI reference at /v1/docs documents every route. Compatible with OpenAI SDKs, LangChain, LlamaIndex, Continue, Claude Code, and Hermes — just change base_url. Deploy via Docker, npm, or build from source. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Paste in a job posting and Dossier, a self-hosted job search app, shows how your profile matches it, then drafts a tailored resume and cover letter that use only facts you have given it. Requirements your profile cannot back up appear as gaps instead of invented experience, and a chat lets you refine any part, undo AI changes, or return to an earlier version. Twelve resume themes and a theme builder print to PDF on the server. Find jobs through JSearch, Adzuna, or Germany's Bundesagentur für Arbeit, score a posting against your profile, and start an application in one click. An optional job agent runs one search a day and can prepare drafts that wait for your approval; it never sends or applies. A board and a table track each application from draft to offer, and the dashboard charts your funnel week by week. Connect Gmail and Google Calendar to send applications from your own address, have the AI read employers' replies and draft answers, and keep interviews in step both ways, with nothing sent until you confirm. Choose a local model through Ollama or LM Studio, or a hosted OpenAI-compatible, Anthropic, or Mistral provider. Everything exports as one file, with nightly backups. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. AGPL-3.0 licensed.

Reaching #1 on GitHub Trending with over 48,000 stars, OpenMontage is the first open-source agentic video production system — transforming AI coding assistants like Claude Code, Cursor, Copilot, Windsurf, and Codex into complete video studios that handle research, scripting, scene planning, asset generation, editing, and final rendering through natural language prompts. Twelve production pipelines cover animated explainers, cinematic trailers, documentary montages, talking heads, screen demos, podcast repurposing, character animation, localization and dubbing, avatar spokesperson videos, hybrid productions, clip factory batch processing, and animation workflows. Over 100 registered Python tools connect to 60+ providers including Kling, Runway Gen-4, Google Veo 3.1, FLUX, Google Imagen 4, ElevenLabs, and Suno AI for cloud generation, plus Piper TTS, WAN 2.1, Hunyuan, and CogVideo for fully local GPU rendering — while free footage from Archive.org, NASA, Wikimedia Commons, Pexels, and Unsplash powers the documentary montage pipeline's CLIP-indexed retrieval system for real-motion video without paid generation APIs. Two composition engines — Remotion for React-based programmatic video and HyperFrames for HTML/GSAP motion graphics — render final output with spring animations, word-level captions, kinetic typography, and SVG character rigs. A seven-dimension scored provider selector, pre-compose validation gates, post-render ffprobe self-review, slideshow risk scoring, configurable budget caps with per-action approval thresholds, and the Backlot live web dashboard for visual production monitoring enforce production-grade quality at every stage. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. AGPL-3.0 licensed.

OmniRoute is an AI gateway, aggregating 338 LLM providers including OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Kimi, MiniMax, and GLM into a single OpenAI-compatible endpoint at localhost:20128. The gateway catalogs over 1,200 models across 90 free-tier providers and 40 free-forever providers, automatically rotating through tier-1, tier-2, and tier-3 fallback chains when any provider exhausts its quota or returns errors. RTK plus Caveman stacked token compression reduces eligible context by 15 to 95 percent before forwarding requests, cutting API costs dramatically without degrading output quality. OmniRoute exposes its full routing engine through a built-in MCP server with 104 tools across 31 scopes over stdio, HTTP, and SSE transports, plus an A2A protocol server with six autonomous agent skills and JSON-RPC 2.0 streaming. The gateway integrates directly with Claude Code, Cursor, GitHub Copilot, Codex CLI, OpenCode, and Cline through standard base-URL configuration. Seventeen routing strategies include latency-optimized, cost-minimized, and auto-scoring modes that evaluate candidates on success rate, context fit, model fitness, quota state, and circuit-breaker health. The Next.js dashboard provides real-time provider status, usage analytics, combo chain configuration, and model catalog browsing via a responsive PWA. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

With over 164,000 GitHub stars and one of the fastest-growing open-source projects in the AI tooling ecosystem, Firecrawl is the web context API that turns any website into clean markdown, structured JSON, or screenshots optimized for large language models. The Scrape endpoint converts a single URL into LLM-ready output with approximately 67% fewer tokens than raw HTML, handling JavaScript rendering, rotating proxies, anti-bot bypasses, and dynamic content extraction with zero configuration. The Crawl endpoint recursively scrapes entire websites from a single request with configurable depth, URL filters, and concurrent page limits. The Map endpoint discovers all URLs on a domain instantly for sitemap generation. The Search endpoint performs web searches and returns full page content from results. The Interact endpoint scrapes a page then continues working with it — clicking buttons, filling forms, and extracting dynamic content using AI prompts or code. The Agent endpoint provides autonomous web data gathering where users describe what they need in plain English. SDKs are available for Python, Node.js, Go, Rust, Ruby, PHP, Java, C#/.NET, and Elixir, with an MCP server for connecting to any AI agent or MCP client. Self-hosting deploys via Docker Compose and requires Redis and a Playwright-based browser service for JavaScript rendering. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. AGPL-3.0 licensed.

God's Eye View turns your browser into a real-time spatial intelligence command center, rendering thousands of live aircraft, ships, satellites, earthquakes, traffic flows, and public cameras on a photorealistic 3D Earth powered by CesiumJS and Google Photorealistic 3D Tiles. Click any aircraft to see its transponder telemetry from OpenSky and adsb.lol, including route history, altitude, speed, and callsign; track live vessel positions worldwide through AIS beacon data from AISStream; or follow roughly 840 satellites color-coded by class using orbital elements from CelesTrak. A hands-free voice agent powered by the OpenAI Realtime API lets you ask the planet questions in natural language, and the globe annotates your answer directly in 3D space. Toggle FLIR mode for a thermal camera aesthetic, layer in NASA FIRMS wildfire data, switch between Google 3D, Bing aerial, and OpenStreetMap base layers, or tune into a geolocated world radio dial. Public CCTV cameras are projected into 3D city geometry with viewshed cones and direct-manipulation calibration. The cockpit mode provides a pilot-style briefing surface with mission-specific overlays. Entity inspection panels show detailed metadata for every tracked object, and shareable links let you send any scene configuration to a colleague. Ten of the thirteen live data layers work with zero API keys, while the required Google Maps key offers 1,000 free 3D tile sessions per month. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Getting a language model into the chat apps people already use takes a single self-hosted server running AstrBot, which answers in QQ, Telegram, Discord, Slack, Lark, DingTalk, LINE, and WeCom. The model itself is a setting. OpenAI-compatible endpoints, Anthropic, Google Gemini, DeepSeek, Ollama, and LM Studio all work, with fallback models and retry limits for failed requests, and Dify, Coze, or Alibaba Cloud Bailian can take over a conversation too. A bare model cannot act, so the built-in agent calls MCP servers and Skills, delegates to SubAgents, runs scheduled tasks, and executes code in an isolated sandbox. It cannot read your files either, so a knowledge base takes PDF, Word, Markdown, and EPUB uploads and searches them with hybrid dense and BM25 retrieval. Personas shape each bot's personality, and Whisper or Edge TTS adds speech. When an answer looks wrong, the ChatUI shows thinking, tool calls, token counts, and latency, and a per-message trace timeline shows how the bot got there. Because the panel steers the bot, administrator IDs, session whitelists, optional TOTP login, and HTTPS protect it. Backups live under System Settings, and the plugin market installs any of more than 1,000 extensions in one click. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. AGPL-3.0 licensed.

With over 84,000 GitHub stars since its April 2026 launch, Open Design has emerged as the definitive open-source alternative to Anthropic's Claude Design. Rather than shipping its own language model, Open Design acts as an agent-agnostic design orchestration layer that auto-detects 25+ coding CLI executables on your PATH — including Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Qwen, GitHub Copilot CLI, Hermes, and Kimi — or connects to any OpenAI-compatible endpoint via its built-in BYOK proxy at /api/proxy/stream. The platform introduces a file-based protocol where SKILL.md files define composable design workflows and DESIGN.md files establish version-controlled brand systems, currently shipping 31 first-party skills and 72 brand-grade design systems. Output spans web and mobile prototypes, live dashboards, presentation decks, raster images, video, and HyperFrames motion graphics, with export pipelines for HTML, PDF, PPTX, MP4, and ZIP. The daemon architecture uses SQLite for project state and serves a sandboxed iframe preview renderer with vendored React and Babel for JSX artifacts. Deployment options include the native desktop app for macOS, Windows, and Linux, local daemon mode via pnpm tools-dev, a single-container Docker Compose path serving both API and Next.js frontend on port 7456, and Vercel deployment for the web layer. On RepoCloud, deploy Open Design on a dedicated VPS with full root SSH access, persistent storage for your design systems and generated artifacts, and complete control over which AI providers and agent CLIs your instance connects to, all under the Apache-2.0 license.

When the connection drops, Project NOMAD keeps serving offline Wikipedia, Khan Academy courses, maps, and an optional private AI assistant from a single Debian-based Linux server. The core needs only about 5 GB of disk and under 1 GB of RAM. Content is what fills the drive, and a setup wizard sorts it into Essential, Standard, and Comprehensive tiers while a storage bar tracks the running total. Wikipedia runs from a quick-reference set of the top 100,000 articles to the complete English edition with images at about 121 GB, and Kiwix also serves medical references, survival and repair guides, and Project Gutenberg ebooks. Kolibri adds Khan Academy courses with multi-user progress tracking, and any browser on the network can study. Offline maps cover the whole globe, single countries, or regions, with markers you can search, color, and navigate to by coordinate. The AI assistant is the heavy part, bringing the install to roughly 25 GB. It chats through Ollama or any OpenAI-compatible server such as LM Studio, searches uploaded documents, lists sources with dates when known, and lets vision models read attached images. CyberChef handles encoding and hashing, and FlatNotes keeps Markdown notes. The Supply Depot adds one-click apps or your own Docker containers. Opt-in auto-updates for the Command Center, installed apps, and downloaded content run inside a window you set. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache 2.0 licensed.

With over 330 GitHub stars and described as "OpenRouter, but for agent tools instead of models," Treg is the unified tool registry that gives your AI agent one base URL, one token, and instant access to approximately 2,800 cataloged endpoints across 56 providers — SEO and backlink analysis via Semrush and Moz, social media intelligence from TikTok and Instagram, people and company enrichment through Crunchbase and Hunter, media buying, scraping, and SERP data — all priced per call from fractions of a cent with no provider signup required. The FastAPI backend with PostgreSQL stores credential bindings and injects them server-side into every proxied request, ensuring API keys and OAuth tokens never touch the agent or local machine while every call is audit-logged. Teams register their own paid API accounts, OAuth connections, vendor CLIs like Stripe, GitHub CLI, Vercel, and gcloud, plus custom SKILL.md definitions — all callable by every teammate's agent through a single treg token with automatic credential injection. The evidence-based tool selector shows success rate, response speed, and last-seen timestamp for each provider, so agents pick tools on measured performance rather than guesswork. Bring-your-own-key routes bypass metering entirely, and the one-dollar free credit on every new team lets agents start calling immediately. The Vue.js web dashboard provides team management, audit logs, catalog browsing, and guided onboarding. Self-host with pip install tools-registry[server] on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache-2.0 licensed.

Reaching 45,000 GitHub stars within seven months of launch, Multica is the fastest-growing open-source platform for managing AI coding agents as first-class teammates — assign an issue to Claude Code, Codex, Cursor, Copilot, Kimi, or any of 21 supported agent CLIs and it picks up the work, comments progress in real time via WebSocket, raises blockers, and hands the result back for human review before anything merges. The Go backend (Chi router, sqlc-generated type-safe queries, gorilla/websocket) connects to PostgreSQL 17 with pgvector for semantic search across workspace history, while the Next.js 16 App Router frontend delivers workspace dashboards showing per-agent token spend, execution time, daily cost charts, and runtime status across unlimited connected machines. Agent Skills provide reusable methods, reference material, and supporting files that compound across runs — a persistent knowledge layer that makes each subsequent task faster and more accurate. Squads let a leader agent select the right specialist for subtasks, creating multi-agent workflows without manual orchestration. Review gates ensure no AI-generated code ships to main without explicit human approval. Self-host via Docker Compose or Kubernetes with full Git integration across GitHub, GitLab, Gitea, and Forgejo including self-hosted instances. The CLI and REST API make every surface scriptable, and Autopilot automations trigger agent runs from events. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache 2.0 licensed with additional conditions.

Designed as a unified personal AI workstation, QwenPaw operates as an autonomous digital copilot that coordinates scheduled automations, interactive coding sessions, and multi-channel team communication across everyday messaging apps. Users can dispatch long-running research tasks, process office documents including PDF, Excel, and Word files, and execute browser-based data collection without manual intervention. The built-in web console and terminal interface provide full visibility into agent reasoning, allowing operators to inspect intermediate thinking steps, review proposed file edits, and approve sensitive tool calls. Through direct connectors for Discord, Telegram, DingTalk, Lark, and WeChat, teams can trigger specialized skills or query shared workspaces directly from their existing group channels. A self-evolving personal memory system continuously indexes chat interactions and local resources into editable Markdown files, ensuring knowledge persists across restarts and task delegations. Built-in guardrails including a sandboxed execution runtime, tool access policies, and automated skill scanners protect underlying host files from unauthorized modifications during autonomous scripting runs. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache 2.0 licensed.

Toonflow, an open-source AI studio for short dramas, animated comics, and short videos, lets you plan, generate, and direct every shot as nodes on a canvas that runs on your own server. Text nodes hold the story, characters, and shot descriptions. Image generation nodes create character sheets with close-up, front, side, and back views. Video generation nodes take numbered reference images. Aspect ratio, resolution, and duration options depend on the selected model. Connectors link the nodes, so a character sheet made once is reusable in every shot. The 3D director's studio blocks a scene with camera framing, lighting, and keyframes. Export the scene as a video node. An assistant panel reads the project and reports which assets and videos have been generated. It offers next steps: generate a shot, revise a prompt, or start a new canvas. Finished images and clips preview on the canvas and download for further production work. Models come from third-party APIs, local ComfyUI workflows, or local language models. Plugins for nodes, skills, tools, and agents install from the marketplace. MCP links external tools and services. Projects, settings, and plugins live in one data directory. The interface offers 21 languages. There is no built-in login, so a firewall or an authenticating reverse proxy decides who can reach it. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

The open-source PaaS with 55,000+ GitHub stars and the largest one-click service catalog in its class — Coolify gives you the Vercel push-to-deploy workflow on a dedicated server you control. Connect a GitHub, GitLab, Bitbucket, or Gitea repository and every push builds and deploys automatically via Nixpacks, a Dockerfile, or Docker Compose, with Traefik or Caddy reverse proxying, automatic Let's Encrypt certificates, and per-branch preview deployments with their own URLs. Databases — PostgreSQL, MySQL, MariaDB, MongoDB, Redis — provision in a few clicks, and a catalog of 280+ one-click service templates covers WordPress, n8n, Grafana, MinIO, Plausible, Ollama, and more. One dashboard manages the entire server, with Docker Swarm available for clustering multiple nodes. Backups go to any S3-compatible storage with one-click restore, and a full API supports CI/CD integration. All configuration lives on your own server, so resources keep running even if you stop using Coolify — no vendor lock-in by design. Running on a dedicated VPS with guaranteed CPU, RAM, and SSD, full root access via SSH and browser console. Apache 2.0 licensed.

Large language models get a polished front end that can run fully offline: Open WebUI is the self-hosted front end of choice. It talks to local model runners, primarily Ollama, and to any OpenAI-compatible API, so LM Studio, vLLM, Groq, Mistral, OpenRouter, and cloud providers all plug into the same chat interface and can be mixed per conversation. RAG is built in: upload files to knowledge bases or reference them in chat with the # command, backed by a choice of nine vector databases (ChromaDB and PGVector officially maintained) and multiple extraction engines including Tika and Docling, with hybrid BM25-plus-vector search and cross-encoder reranking. Web search results from providers like SearXNG, Brave, and Tavily inject directly into conversations. Extensibility comes from Python tools and functions that run inside the chat, a Pipelines plugin framework, and native MCP support. Multi-user features include RBAC, SSO, and group permissions, and the instance itself exposes an OpenAI-compatible API your own apps can call.

Backed by 56,000+ GitHub stars and over 240 million Docker pulls, LiteLLM delivers the open-source AI gateway trusted by Netflix, Lemonade, Rocket Money, and thousands of engineering teams to route every LLM request through one unified API. The Rust-core gateway adds sub-millisecond overhead per request with 8ms P95 latency at 1,000 RPS, 15x throughput improvement and 11x lower memory footprint compared to Python-only proxies. A single OpenAI-compatible endpoint connects to 100+ providers and 1,800+ models spanning OpenAI, Anthropic, Google Gemini, AWS Bedrock, Azure OpenAI, Vertex AI, Hugging Face, vLLM, Nvidia NIM, Ollama, and Mistral with day-zero support for new model releases. The Auto Router V2 classifies request complexity across four tiers using rule-based scoring, semantic keyword matching, and adaptive Thompson sampling to route each request to the most cost-effective model without API calls or training data. Virtual API keys enable multi-tenant governance with per-team, per-user, and per-project cost tracking, budget caps with automatic fallback rerouting, and role-based access control. Built-in guardrails provide PII masking, prompt injection detection, and model-graded evaluation before requests reach providers. The Agent Gateway extends routing from model calls to agent workflows with MCP server integration. Observability integrates with Langfuse, Arize Phoenix, OpenTelemetry, and MLflow for complete request tracing. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.