Thirteen-plus LLM providers, one unified client: NextChat (formerly ChatGPT-Next-Web) is an open-source AI chat interface built on Next.js that spans OpenAI GPT-4, Anthropic Claude, Google Gemini, DeepSeek, Groq, Azure endpoints, and self-hosted backends like Ollama, LocalAI, and RWKV-Runner. Its defining trait is minimalism - the first screen loads in about 100 KB, the desktop client is roughly 5 MB, and there is no database or user system to operate; chat history lives locally in the browser with optional WebDAV or UpStash Redis sync. The Mask system saves reusable prompt-template personas you can share and debug, long conversations auto-compress to fit context windows, and Markdown rendering covers LaTeX, Mermaid diagrams, and code highlighting with streaming responses. Plugins add web search and calculators, MCP support enables external tool calling, and Artifacts previews generated content in a separate pane. Ships as a web app, Docker image, and Tauri desktop builds for Windows, macOS, and Linux, translated into 20+ languages. MIT-licensed.

Claiming 60-80% token cost reduction compared to flat single-model deployments and backed by 6,500+ GitHub stars, OpenSquilla delivers an intelligent AI agent runtime where a local ML classifier evaluates every turn on message length, code blocks, keyword patterns, and semantic embeddings before routing it to the optimal model tier from C0 through C3. The pluggable provider layer connects natively to TokenRhythm, OpenRouter, OpenAI, Anthropic, Ollama, DeepSeek, Gemini, DashScope, Moonshot, Mistral, Groq, Zhipu, SiliconFlow, vLLM, LM Studio, and additional compatible backends with primary-plus-fallback selection. The four-tier cognitive memory architecture spans working, episodic, semantic, and raw layers with vector-semantic and BM25 retrieval powered by on-device ONNX embeddings that never leave your infrastructure. Security isolation operates at the syscall level via Bubblewrap on Linux and Seatbelt on macOS, complemented by policy-based execution controls and prompt injection protections. The unified TurnRunner executes identically across the Vue-based control console Web UI, terminal CLI, and chat channel integrations including Slack and Discord, ensuring consistent tool dispatch, retry logic, and decision logging regardless of entry point. Built-in skills cover deep research, multi-search-engine queries, document generation for DOCX, PPTX, XLSX, and PDF formats, GitHub integration, cron scheduling, and bounded subagent delegation. Per-agent workspaces with durable session storage provide transcript replay, context state management, and per-call cost tracking with automatic quota enforcement. 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.

FastGPT lets you build production AI agents and knowledge base chatbots through a visual drag-and-drop workflow editor, connecting any LLM provider to your documents with retrieval-augmented generation that cites sources and reduces hallucination. The workflow canvas chains LLM calls, conditional branching, HTTP requests, code sandbox execution, and plugin nodes into complex conversation flows and agent skill pipelines without writing backend code. The knowledge base engine ingests documents in ten formats (TXT, Markdown, HTML, PDF, DOCX, PPTX, CSV, XLSX, URL scraping, and CSV batch import) then applies automatic chunking, hybrid vector retrieval with semantic reranking, and QA-pair splitting to deliver accurate, citation-backed answers. FastGPT connects to virtually any LLM provider through its AI Proxy aggregation layer: OpenAI GPT-4o, Anthropic Claude, Google Gemini, DeepSeek, Qwen, ERNIE Bot, and models hosted via Ollama all work through a unified OpenAI-compatible API. Bidirectional MCP support enables agents to call external tools and expose their own capabilities to other systems. Completed applications can be shared via login-free links, embedded as iframe widgets, or integrated with WeCom, Lark, DingTalk, and WeChat Official Accounts through the published REST API. Application operation logs, conversation annotation, and per-model usage analytics provide full lifecycle governance for compliance-sensitive deployments. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. FastGPT Open Source License (Apache 2.0 based) licensed.

With over 46,000 GitHub stars, nanobot is the ultra-lightweight personal AI agent framework that delivers full agentic capabilities — tools, persistent memory, multi-agent workflows, scheduled automation, and 10+ chat channel integrations — in approximately 4,000 lines of readable Python core code. The agent loop receives messages from any connected channel, builds context from session history and long-term memory files, calls the configured LLM provider, executes requested tools, and publishes replies back to the originating channel. Supported LLM providers include OpenAI, Anthropic, Google Gemini, DeepSeek, Qwen via DashScope, Moonshot/Kimi, Ollama, vLLM for local models, and any OpenAI-compatible API through OpenRouter or LiteLLM. Chat channels connect the agent to Telegram, Discord, Slack, WhatsApp, Feishu/Lark, DingTalk, Email via IMAP/SMTP, QQ, Matrix with end-to-end encryption, Mattermost, and the built-in browser WebUI served from the published Python wheel with no separate frontend build. Built-in tools include filesystem read/write/edit, shell execution with configurable sandboxing via bubblewrap, web search and fetch with SSRF protection, MCP server integration, cron scheduling, image generation, and subagent spawning for parallel task delegation. The Dream memory system consolidates session history into persistent markdown files for long-term context retention across conversations. Deployment runs as a CLI agent, a persistent gateway server, a Docker container with Docker Compose, or an OpenAI-compatible API server. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Aegra runs your LangGraph agents on your own infrastructure with the same SDK and APIs you would use on the managed platform, so moving in does not mean rewriting your agents. You deploy an agent once and clients connect through Agent Chat UI, LangGraph Studio, or CopilotKit, while threads, runs, and checkpoints persist between sessions instead of vanishing when a process restarts. Streaming answers arrive token by token over server-sent events, and if a connection drops, the stream reconnects and replays missed events rather than losing the run. Human-in-the-loop gates pause an agent at the step you choose, wait for a person to approve or edit, and then resume from exactly that state. Scheduled cron jobs trigger runs on standard or seconds-level expressions with timezone support, which turns agents into recurring workers for reports, checks, and follow-ups. When traffic grows, Redis-backed workers process runs with lease-based crash recovery, so a dead worker's job is picked up instead of disappearing. Authentication handlers plug in JWT, OAuth, or Firebase, custom endpoints sit beside the agent API, and traces fan out to the observability backend you already run. 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.

Document review, fixed-layout report generation, and multi-step research that ends in a PPT, Excel, or Word file all run inside BISHENG, a self-hosted workspace for enterprise AI applications. Review and report flows are drawn on a visual canvas where a circle becomes a loop, side-by-side nodes run in parallel, and a multi-item selection runs as a batch. A running flow can stop for a person to give input, confirm, or make changes, and chat and automation share one flow type. Research goes to Task Mode: give it a goal plus any files, knowledge bases, tools, and skills, and it writes a checklist, hands research to parallel sub-agents, and returns a report with clickable citation markers. Teams sharing documents use Knowledge Spaces, with folders, tags, keyword search, AI Q&A that shows its sources, and permissions down to a single file for people, departments, or user groups. For a daily news briefing, Channels turn WeChat official accounts and websites into topic feeds and save articles as knowledge in one click. Admins connect OpenAI-compatible APIs, Azure OpenAI, Anthropic, Qwen, DeepSeek, and local Ollama models, then govern access with approvals, audit logs, SSO, and LDAP. Private document parsing reads printed text, handwriting, tables, layouts, and seals, and a service-account Open API lets other systems call applications and knowledge bases. 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.

Support teams, IT departments, and researchers can put their own documents behind chatbots and AI agents built with MaxKB, an open-source platform that grounds answers in private knowledge. A support lead starts by uploading Markdown, DOCX, PDF, TXT, HTML, XLSX, XLS, or CSV files, or by syncing a web site, and MaxKB splits, vectorizes, and indexes the content on its own. An IT department can place the finished agent in any third-party site or internal tool with no coding, as a full-screen page, a mobile view, or a floating chat window. Researchers get replies that show, and let them download, the knowledge sources behind each answer. Simple agents pick tools, skills, and MCP connections from the question itself, while advanced agents are drawn as workflows with branching, multi-path recall, loops, and variables. Custom scripts, which AI can help write, plus schedules and event triggers extend what an agent can do. The model manager registers local models such as DeepSeek, Llama, and Qwen beside providers such as OpenAI, Claude, and Gemini, across language, embedding, reranking, vision, speech, and image generation. Conversation logs feed analysis for operations. Dozens of ready-made agent templates, knowledge base templates, tools, and skills in the template center and tool store shorten the first build. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. GPL-3.0 licensed.

Recognized by the CNCF Landscape in the AI Agent – Workflow Orchestration category alongside Dify and Temporal, Astron Agent is iFLYTEK's fully open-source platform for building, deploying, and managing enterprise multi-agent systems — backed by 8,900+ GitHub stars and the production infrastructure behind one of China's largest AI companies. Unlike pip-install frameworks such as LangGraph, CrewAI, or AutoGen, Astron Agent ships as a complete microservices platform spanning 10+ services across Java, Python, Go, and TypeScript: a ReactFlow-based visual workflow builder for low-code agent orchestration, native integration with the Model Context Protocol (MCP) for tool calling, a built-in model management layer supporting iFLYTEK Spark, OpenAI, Anthropic, and on-premises MaaS deployments, and a multi-tenant Go authentication service powered by Casdoor. The standout differentiator is native RPA integration via the companion astron-rpa project (7,200+ stars), providing 300+ pre-built automation capabilities spanning browser, Office document, and enterprise system interaction — enabling agents to execute physical UI actions rather than only API calls. Infrastructure includes PostgreSQL for multi-tenant data isolation, MySQL for application metadata, Kafka for event streaming, Redis for caching, and MinIO for object storage, all orchestrated through Docker Compose with explicit health checks and dependency chains or production Kubernetes Helm charts. 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.

Every prompt your application sends is a small buying decision, and OrcaRouter Lite makes it a smart one instead of a habit. Point your app at one OpenAI-compatible endpoint on your server, and each request is shopped across the providers whose keys you hold: OpenAI, Anthropic, Google, xAI, DeepSeek, and dozens more. Send model set to auto and the router picks the cheapest model that can actually do the job, checking whether the request needs tools, vision, or JSON mode before it commits your money. Prefer a different tradeoff and you can switch the whole gateway between balanced, cheapest, fastest, and quality strategies, with quality picks driven by real benchmark scores rather than vibes. Clients that speak Anthropic's or Gemini's native protocols connect directly too, so Claude Code, the OpenAI SDK, and google-genai all land on the same routing pipeline. Streaming responses flow through untouched, and every answer reports which model really served it. The built-in dashboard shows recent requests, spend, latency percentiles, unreachable providers, and a savings tile that tells you what your traffic would have cost if you had blindly used a flagship model for everything. Your provider keys stay encrypted on your own box, your analytics never leave it, and if every local provider has a bad day, the hosted OrcaRouter can step in as a failover safety net. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Inference Gateway puts a single OpenAI-compatible API endpoint in front of OpenAI, Anthropic, Groq, Cohere, Ollama, DeepSeek, Google, Mistral, MiniMax, Moonshot, Nvidia, and llama.cpp, so your application code never changes when you switch models or providers. The Go binary starts on port 8080 and normalizes authentication, streaming protocols, and response formats across all backends transparently. Native Model Context Protocol support auto-discovers tools from connected MCP servers and injects them into LLM requests without client-side management, enabling server-side tool execution across any provider that supports function calling. Agent-to-Agent protocol integration allows distributed agent communication through a declarative Agent Definition Language that generates production-ready Go or Rust servers from a single YAML manifest. The dedicated Kubernetes Operator manages Gateway, Agent, MCP, and Orchestrator custom resources with automatic HPA scaling, OIDC authentication, and service discovery that rebuilds MCP configurations when the discovered server set changes. Prometheus metrics and OpenTelemetry tracing provide full request-level observability across the entire inference pipeline. Middleware controls enable per-request provider selection, model routing, and fallback strategies. Official SDKs in Go, Python, TypeScript, and Rust provide typed client interfaces with streaming support. Docker Compose deployment requires only environment variables for API keys. A CNCF Sandbox applicant. 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.

Routerly is the one front door for every AI model your apps use: point your OpenAI or Anthropic client at it, and it chooses the right model for each request, enforces your budget, and keeps the receipts. Nothing in your code changes except the base URL, and from that moment streaming chat completions and messages flow to OpenAI, Anthropic, Gemini, Mistral, Cohere, xAI or your own Ollama models behind a single token. Nine routing policies score every candidate in parallel: cheapest, fastest, healthiest, most capable, biggest context, least rate-limited, fairest, most budget left, and one that asks a small model to judge which model fits the task. When the winner errors or times out, the next candidate takes over automatically, and the recorded routing trace shows exactly why each model won or lost. Guardrails can block or log content that breaks your rules, and phone numbers, card numbers, SSNs, IBANs and emails can be scrubbed out before anything reaches a provider. Every call is logged with tokens, USD cost and latency, and hard limits per model, project and token reject overspending requests before a provider can charge for them. Projects keep separate tokens, model lists and budgets for each team or tool, role-based permissions decide who may change what, and alerts land in Slack, email, webhooks or PagerDuty when budgets or providers misbehave. 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 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.

Plain chunk retrieval hands an LLM disconnected excerpts; LightRAG builds a knowledge graph from your documents so answers can join facts spread across many files. Ask about one named person, product, or clause and local mode pulls its details. Ask for themes across the whole library and global mode follows the links between entities. Hybrid blends the two, naive falls back to plain chunk search, and mix combines all three. Mix is the default. Upload text, Markdown, PDF, Word, PowerPoint, and Excel files; MinerU, Docling, or the native parser can extract images, tables, and formulas. The admin console manages documents, explores the graph visually, and debugs queries, while a query-only workspace page gives everyday colleagues a clean chat screen that works on phones. A REST API covers indexing and queries, and an Ollama-compatible endpoint lets chat front ends such as Open WebUI use the knowledge base as a model. New files merge into the existing graph without a full reindex, and deleting a document rebuilds only the entities and relationships it touched from cached results. Separate model roles let an inexpensive model extract entities while a stronger one writes answers. An API key or accounts protect the server, and named workspaces keep several knowledge bases apart. Pair it with local Ollama models and no document text leaves the server. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Five teammates sharing three AI subscriptions usually means a password in a group chat; SubLane replaces that with a gateway you host, where accounts are pooled and every person gets a managed key. Connect Codex, Claude, Antigravity, and Grok Build subscriptions once, by browser OAuth or credential import, and the gateway snapshots each account's quota windows and allowances so routing knows who still has room. Add API channels for OpenAI-compatible services alongside the subscriptions, then group everything into resource pools with model allowlists, conversation affinity, concurrency controls, and cooldown recovery for accounts that hit a limit. Each member receives a personal key scoped to one group, with its own expiry, rate and concurrency limits, and an allowance measured in tokens, internal dollars, or estimated quota share; keys can be paused or revoked without touching anyone else. Whatever the client speaks, the gateway answers: OpenAI Responses and Chat Completions, Claude Messages, and Gemini formats flow over HTTP, SSE, or WebSocket, independent of which subscription actually serves the request. Administrators watch per-member and per-model usage charts, activity heatmaps, and request diagnostics, bind a fixed outbound proxy to any account, and apply workspace content rules before dispatch. It all runs as one Go process with SQLite and an embedded dashboard, backed up and restored from the interface. 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.

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.

ENZO is a self-hosted AI workspace where you bring your own provider API keys and chat with more than 300 models from nine providers in one streaming interface with normal, thinking, research, and coding modes. You describe an agent in plain English and a two-pass builder drafts its full operating manual, picks the tools the task needs, and keeps improving from your activity through a per-agent neural layer you can inspect. Seventy-four bundled skills give agents ready-made domain playbooks, and a skill editor lets you write your own. Agents can read Gmail, check Calendar, search the web, and run a deep-research loop that writes its own queries under hard budgets so it cannot burn your key. Coding mode writes a multi-file project, boots it, and previews it live while telling you when something is broken. A unified model marketplace compares providers with live health checks, download counts, and benchmarks, and a built-in music player streams songs with a five-band equalizer. Your keys stay sealed in a passphrase-encrypted browser vault and travel only to the provider you picked, so there is no ENZO account, usage meter, or subscription taking a cut. 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.

Self-hosting a language model is only half the job; people still need somewhere pleasant to talk to it, and Chat UI is that window, the same interface that powers HuggingChat. Point it at any OpenAI-compatible endpoint, your own Ollama or llama.cpp server, the Hugging Face router, or a hosted gateway like OpenRouter, and the models it serves appear in the picker automatically, with no per-model configuration. Conversations are saved to your database and stay browsable in the sidebar, so a thread from last week can be reopened, continued, and compared against a fresh answer from another model. Multimodal models accept image uploads right in the composer, and models that support it can call MCP tools mid-conversation to fetch data or run actions. For a shared team deployment, OpenID Connect sign-in keeps individual histories and settings separate, and per-user defaults travel with each account. Everything a consumer chatbot would keep on someone else's servers, prompts, uploaded files, and usage history, stays in the MongoDB you control on the same machine. 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.

Accepted as a NeurIPS 2025 Spotlight paper and rapidly approaching 6,000 GitHub stars, AI-Researcher from the Hong Kong University Data Science Lab delivers the first fully autonomous scientific research system — a multi-agent platform that takes a list of reference papers and returns a complete research contribution with working code, validated experiments, and a formatted academic manuscript. The pipeline orchestrates five distinct phases: a Resource Collector systematically gathers materials from arXiv, IEEE Xplore, ACM Digital Library, Google Scholar, GitHub, and Hugging Face; an Idea Generator performs gap analysis against semantic embeddings to produce 3-5 novel hypotheses with feasibility scores; an Algorithm Designer transforms concepts into functional implementations; a Validation Engine automates testing, performance evaluation, and iterative optimization; and a Manuscript Creator generates polished full-length papers with figures, tables, and citations. The Gradio-based web GUI provides intuitive tabs for environment configuration, example selection, and real-time monitoring of research progress, while the production deployment at novix.science offers immediate browser access without local setup. Scientist-Bench provides a standardized benchmark comprising state-of-the-art papers across diverse AI research domains for evaluating autonomous research capabilities. The system supports multiple LLM providers including OpenAI, Anthropic, Google Gemini, and OpenRouter models with per-task routing for cost optimization. Deploy via Python with pip dependencies or Docker containerization. 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.