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.

Every major model provider behind one ChatGPT-style interface: LibreChat spans OpenAI, Anthropic, Google, Azure, AWS Bedrock, Vertex AI, Groq, Mistral, OpenRouter, DeepSeek, and any OpenAI-compatible endpoint including local Ollama. You can switch models mid-conversation and compare providers without changing tools. Its Agents framework builds no-code custom assistants with tool access via Model Context Protocol servers, file search over uploaded documents through an optional pgvector-backed RAG service, and a sandboxed Code Interpreter that executes Python, JavaScript, Go, C++, Java, PHP, and Rust. Artifacts render React components, HTML, and Mermaid diagrams directly in chat, and image generation works through DALL-E and other configured providers. Multi-user support is enterprise-grade, with OAuth, SAML, LDAP, and two-factor authentication, per-user conversation history in MongoDB, and Meilisearch-powered search across all messages and files, plus reusable presets, forkable threads, and persistent memory across conversations. The economics favor teams: instead of a ChatGPT Plus seat per person, everyone shares one instance billed per API token, with access to every provider rather than one - and providers see individual API calls, not your accumulated organizational knowledge. Deployment is Docker Compose; API keys and endpoints are configured through .env and librechat.yaml.

OpenRouter's most-used application by token volume — over 17 trillion tokens processed — Hermes Agent is an open-source autonomous agent built by Nous Research that lives on your server and gets more capable every day. Define a goal in natural language and Hermes plans sub-tasks, executes them through tool integrations, observes results, handles errors, and refines until the job is done or it genuinely needs your input. Persistent memory with full-text search and LLM summarization lets it recall context across sessions, and an agent-created skills system self-improves after complex tasks. A messaging gateway connects Telegram, Discord, Slack, WhatsApp, Signal, and 16 more platforms with cross-channel conversation continuity. A built-in cron scheduler runs daily reports, nightly backups, and weekly audits unattended. Subagent spawning parallelizes workstreams, and six terminal backends — local, Docker, SSH, Singularity, Modal, and Daytona — fit any infrastructure. Works with any LLM provider: Nous Portal, OpenRouter for 400+ models from 70+ providers, OpenAI, Anthropic, or your own endpoint. The API key you supply powers all LLM calls; billing goes through your own account. Running on a dedicated VPS with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

With over 2,100 GitHub stars in its first month and benchmarked at 30-75% lower cost than Claude Managed Agents on enterprise task suites, TrueForge is the open-source agent harness that provides the complete runtime layer for turning any LLM into a working production agent on your own infrastructure. The TypeScript server runs the full execution loop — streaming every step, routing tool calls through MCP servers with centralized header-auth and in-chat OAuth, delegating parallelizable work to isolated subagents, and pausing for human approval on sensitive actions. Context engineering keeps token costs low: deferred tool-schema loading delays MCP schemas until invoked, large-result offloading moves oversized outputs to files, Code Mode processes structured data through sandboxed execution, and automatic compaction summarizes older history at a configurable 50,000-token threshold while preserving the full transcript. The sandbox-as-a-tool architecture provisions isolated Daytona environments only when code execution is required, allowing one server to run many concurrent agents without idle overhead. Agents are configured from shipped YAML catalogs of models, MCP servers, git-backed SKILL.md instruction packs, and sandbox providers, then saved to an Agents Library accessible via the chat UI, TypeScript SDK, or embeddable React UI SDK. Run locally with SQLite via a single npx command, or deploy for teams with Docker Compose or Helm using Postgres and Redis with OIDC authentication. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Formerly known as Danswer and now backed by over 31,000 GitHub stars with 253 releases, Onyx delivers a production-ready AI platform that turns any LLM into a context-aware enterprise assistant connected to your organization's actual knowledge. The agentic RAG pipeline combines BM-25 keyword search with prefix-aware embedding models in a hybrid index, then deploys AI agents to retrieve, verify, and synthesize answers with source citations from over 40 connected workplace tools including Google Drive, Confluence, Slack, Notion, Jira, SharePoint, GitHub, and Linear. Custom AI assistants with configurable prompts, backing knowledge sets, and document-level access control enable specialized agents for engineering, sales, support, and research workflows. The platform supports every major LLM provider — Anthropic Claude, OpenAI, Google Gemini, plus self-hosted options via Ollama, LiteLLM, and vLLM for fully air-gapped deployments. Beyond chat, Onyx provides web search with Serper, Google PSE, Brave, and SearXNG integration, an in-house web crawler, code execution, file creation, and multi-step deep research with report generation. Enterprise features include SSO via Google OAuth, OIDC, or SAML with SCIM provisioning, role-based access control, usage analytics by team and agent, query history auditing, PII removal through custom code hooks, and full whitelabeling. Deploy via Docker Compose on any infrastructure. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed (Community Edition).

DeerFlow 2.0 is ByteDance's ground-up rewrite that transforms what began as a Deep Research framework into a batteries-included super agent harness handling tasks lasting minutes to hours autonomously. Built on LangGraph and LangChain, the runtime orchestrates a lead agent that plans, decomposes, and delegates to sub-agents executing in isolated Docker or Kubernetes sandboxes with persistent filesystem access, while an extensible skills system lets developers add capabilities as Python functions or MCP servers with OAuth token flows. The harness ships with long-term memory using persistent event loops with per-agent isolation, scheduled task execution via cron, context engineering with manual compaction, and a web UI at localhost:2026 for interactive sessions. Model support spans OpenAI GPT-4o/GPT-5, Anthropic Claude via OAuth, Google Gemini, DeepSeek, Qwen via vLLM, and OpenRouter-compatible gateways with per-model pricing configuration. IM channel integration connects Telegram, Slack, Feishu/Lark, Discord, WeChat, WeCom, DingTalk, and Buzz without requiring a public IP — all using long-polling or WebSocket transports. Production deployment uses Docker Compose with nginx reverse proxy, PostgreSQL or SQLite persistence, Redis stream bridges for multi-worker scaling, and lease-based run ownership with automatic orphan recovery. The terminal workbench TUI enables headless operation. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

A personal AI assistant that remembers what it learns and reaches you wherever you are — OpenClaw is an open-source agent gateway built by the OpenClaw Foundation with 247,000+ GitHub stars. It connects to 200+ LLM models through providers like Anthropic, OpenRouter, and OpenAI, and meets you on 21+ messaging channels: Telegram, Slack, Discord, WhatsApp, Signal, iMessage, Matrix, and more. Persistent memory with full-text search lets the agent recall context across sessions, and a self-improving skills system means it gets more capable the longer it runs. Voice wake words and talk mode enable hands-free interaction on macOS, iOS, and Android. A live canvas provides an agent-driven visual workspace. Built-in browser automation, cron scheduling for unattended tasks, and subagent spawning for parallel workstreams round out the toolset. The gateway architecture keeps all sessions, credentials, and conversation history on your own server — nothing transits a third-party cloud unless you choose to connect one. The API key you provide for your chosen LLM provider powers the underlying calls; billing goes through your own account. Running on a dedicated VPS with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Langflow turns LLM application development into a visual canvas where every node maps to a real LangChain primitive (chains, agents, retrievers, memory, vector stores, and tools) that teams wire together without writing integration boilerplate. The platform supports 15+ LLM provider components including OpenAI, Anthropic, Google Gemini, Mistral, Groq, Cohere, Azure OpenAI, HuggingFace, and Ollama for fully local inference, with LiteLLM proxying to over 100 additional providers through a single OpenAI-compatible endpoint. Vector database integrations cover Pinecone, Weaviate, Chroma, Qdrant, Astra DB, OpenSearch, FAISS, and Milvus, while built-in Knowledge Bases introduced in version 1.8 allow RAG pipelines without any external vector service. Multi-agent orchestration enables agent-to-agent communication with conversation management, persistent Memory Bases for cross-session context retrieval, and step-by-step reasoning visibility in the interactive Playground. Every flow automatically becomes a callable REST API endpoint via the /run route and an MCP server exposable to Claude Desktop, Cursor, or any MCP-compatible client. Every component is a real Python class that developers can customize, extend, or replace, while Extension Bundles package third-party integrations as independent pip packages for modular installation. Tool integrations include web search, Slack, Gmail, Google Drive, GitHub, and custom REST API calls. Docker deployment with PostgreSQL persistence runs on port 7860. Backed by 153,000+ stars and DataStax. 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 14,000 GitHub stars and 41 releases since its May 2026 launch, Prime Agent delivers a fundamentally different approach to AI coding agents by treating the agent's own operating environment as programmable state that improves through use. The Recursive Language Model architecture provides the model exactly one tool — a persistent IPython kernel — where file operations, shell commands, subagent delegation via rlm() function calls, and context management all happen through code rather than rigid tool-calling schemas. Subagents launch as independent sessions with their own model, kernel, and history, communicating results through agent_message.send() without blocking the parent. The Continual Harness stores supplemental prompts, memories, skill descriptions, and reusable subagent specifications as durable state that the /refine command updates through small, evidence-backed edits with full rollback by ID. Daemon-backed sessions keep running when the terminal disconnects, with automatic context compaction summarizing older messages while preserving recent state. The TUI provides an Agent View for monitoring, switching between, and steering multiple live sessions simultaneously. Autonomous mode operates within configurable turn, token, and time budgets with user-defined quality gates. Persistent goals, heartbeats, and scheduled prompts maintain continuity across terminal sessions. Compatible with Anthropic Claude, OpenAI, Google Gemini, local models via Ollama or vLLM, and Prime Inference endpoints. Install via a single curl command on Linux or macOS. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Prompt, run, edit, and deploy full-stack Node.js applications from a browser tab: Bolt.diy is the official open-source version of Bolt.new's AI coding agent. Its foundation is StackBlitz's WebContainer technology - a sandboxed in-browser Node.js environment where the AI controls the whole stack: filesystem, npm, dev servers, terminal, and browser console. That means the agent does not just generate code; it installs dependencies, runs Vite or Next.js, reads errors, and fixes them. The defining difference from Bolt.new is model choice per prompt: 19+ providers including OpenAI, Anthropic, Gemini, DeepSeek, Groq, Mistral, Amazon Bedrock, and local models via Ollama or LMStudio, extensible through the Vercel AI SDK. Development ergonomics include live preview, a diff view of AI changes, codebase search, file locking to prevent generation conflicts, 15+ starter templates (React, Vue, Next.js, Astro, Svelte, Expo), and MCP support for external tools. Projects integrate with Git and Supabase, and deploy in one click to Vercel, Netlify, or GitHub Pages.

Archestra delivers the enterprise AI infrastructure layer that organizations need when managing multiple LLM providers, MCP servers, and AI agents across teams becomes unmanageable. The LLM gateway routes requests across Anthropic, OpenAI, Azure, Bedrock, and DeepSeek with virtual API keys, per-team cost limits, and dynamic model routing — giving every developer one token for Claude Code, Cursor, or Codex while finance tracks spend per department. The MCP gateway authenticates tool calls with OAuth 2.1 and On-Behalf-Of tokens so each tool executes as the calling user, not a shared service account, eliminating credential sprawl. The private MCP registry lets teams publish custom tool servers with approval flows promoting servers from dev through staging to production, each environment maintaining its own credentials and network egress policies. The Kubernetes operator manages MCP server lifecycle — deploying containers, scaling, health-checking, and routing gateway traffic to local servers automatically. The agent runtime supports scheduled triggers, email and webhook invocations, sub-agent delegation, reusable skills, and sandboxed code execution with a K8s-native filesystem. Deterministic guardrails including Dual-LLM verification and Lethal Trifecta protections prevent dangerous tool calls before execution. Built-in OpenTelemetry traces and Prometheus metrics provide full observability without additional tooling. Docker deployment exposes the Admin UI on port 3000 and API on port 9000 with a single command. 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.

Agents with tool use, deep research, a document editor, an IMAP/SMTP email client with AI triage, notes, tasks, and a CalDAV-synced calendar - Odysseus bundles all of it into one open-source, self-hosted AI workspace. It runs local models through Ollama, vLLM, or llama.cpp and cloud APIs like OpenAI and OpenRouter, with a hardware-aware Cookbook that scans your machine and recommends quantized models that fit. Persistent memory uses ChromaDB with hybrid vector-plus-keyword retrieval, web search runs through a bundled SearXNG instance, and agents can use MCP servers, files, and shell access with safety controls, plus custom skills and scheduled agent tasks. A blind Compare mode runs side-by-side model duels with identities hidden and accumulates Elo-style ratings from your votes, so model selection is based on your actual workloads rather than leaderboard claims. Deep research mode - adapted from the Tongyi DeepResearch approach - reads sources through SearXNG and produces cited reports, while the email client tags, summarizes, sets reminders, and drafts replies locally rather than through a third-party mail AI. The writing-first document editor adds AI edits, Markdown and HTML support, and version history. The stack is Python 3.11 with FastAPI, SQLite for state, and a vanilla JS frontend, licensed AGPL-3.0 with zero telemetry. Because agents can read email and execute commands, keep authentication enabled and never expose it as a public unauthenticated service.

Chat with your own documents: AnythingLLM, from Mintplex Labs, wraps retrieval-augmented generation (RAG) in an open-source application anyone can run. You organize content into workspaces, each an isolated namespace with its own documents, vector embeddings, chat history, and settings, so one instance can hold several separate knowledge bases. Upload PDFs, DOCX, TXT, and other formats, or scrape web pages; the built-in collector parses and chunks them into a vector database (LanceDB by default, with Pinecone, Chroma, Qdrant, and others supported). Answers cite their source documents. It works with both cloud LLMs (OpenAI, Anthropic, Gemini) and local ones via Ollama or LM Studio, and the embedding model is separately configurable. Beyond RAG chat, it includes AI agents that can browse the web and run tools, an embeddable chat widget for your website, a developer API, and multi-user mode with admin, manager, and default roles plus per-workspace access control. Context assembly is smarter than naive RAG: pinned documents, attached files, vector search hits, and recent chat history are combined under a token budget so the model's context window is filled efficiently, and each workspace supports multiple independent conversation threads against the same knowledge base. Because the embedding model, vector store, and chat LLM are all independently swappable, you can move between providers without re-ingesting a single document. The stack is Node.js with a React frontend, MIT-licensed.

Open-source alternative to OpenAI's Canvas — a collaborative writing and coding environment where AI agents help you draft, edit, and refine documents through an agentic architecture built on LangGraph. The dual-mode editor combines a BlockNote rich text editor for live-rendered markdown with a CodeMirror-based code editor supporting syntax highlighting across multiple programming languages, letting you switch between prose and code artifacts within the same session. The built-in reflection agent automatically generates style rules and user insights from your chat history, storing them in a shared LangGraph memory store that persists across sessions for increasingly personalized assistance. Pre-built quick actions provide one-click access to common writing transformations including summarize, expand, simplify, and translate, while coding actions offer explain, refactor, add comments, and convert between languages. The monorepo architecture separates the Next.js 14 frontend from the LangGraph agent backend, connecting via HTTP and WebSocket protocols through the @langchain/langgraph-sdk client. Seven LLM providers are supported out of the box — OpenAI, Anthropic Claude, Google Gemini, Fireworks AI, Groq, Azure OpenAI, and local Ollama models — with Supabase handling authentication and data persistence. Deploy via Docker 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.

Describe a component in natural language and watch it render: OpenUI, from Weights & Biases, is an open alternative to Vercel's v0. Type a prompt like "a dark-themed dashboard with a sidebar and charts" and the LLM renders working HTML with Tailwind styling live in the browser. You then iterate conversationally, asking for changes until the design is right, and convert the result to React, Svelte, or Web Components for use in a real project. The backend is Python with LiteLLM routing, so it works with OpenAI, Anthropic, Gemini, Groq, and Mistral API keys, or fully offline against local Ollama models, including vision models like LLaVA that can generate UI from screenshot input - feed a screenshot and the model reproduces or riffs on an existing interface. Generated markup is inspectable at any point, with light and dark mode toggles, theme selection, and responsive previews across device sizes. The practical effect is compressing the mockup-review-revise loop from hours to minutes: a described layout renders in seconds and iterates through follow-up prompts, and because output converts to real framework code, prototypes feed directly into production codebases instead of staying trapped in a design tool. Self-hosting keeps unreleased product interfaces and prompts on your own server, and LiteLLM routing lets you pick the model per task - a cheap fast model for rough drafts, a stronger one for final passes, or free local models for unlimited experimentation.

Dify turns the notoriously complex process of building production-grade AI applications into a visual drag-and-drop experience that teams can actually ship and maintain. With over 87,000 GitHub stars and backing from prominent investors, the platform has become the go-to open-source LLMOps solution for organizations that refuse to be locked into proprietary AI stacks. The visual workflow canvas lets developers wire together LLM calls, conditional logic, iteration loops, tool invocations, and human-in-the-loop checkpoints without writing boilerplate integration code. Its RAG pipeline engine handles the full document lifecycle from ingestion of PDFs, Word documents, and HTML through configurable chunking strategies, embedding with models from OpenAI or open-source alternatives, vector storage in Weaviate, Qdrant, Pinecone, or pgvector, and hybrid semantic-plus-keyword retrieval with citation tracking. Dify integrates with hundreds of model providers including OpenAI GPT-4o, Anthropic Claude, Google Gemini, Mistral, Llama, and any OpenAI-compatible endpoint like Ollama for fully local inference. The agent framework supports both ReAct and function-calling strategies with 50-plus built-in tools spanning Google Search, DALL-E, Stable Diffusion, WolframAlpha, and custom API definitions. Published apps can be deployed as hosted web interfaces, embedded chat widgets, REST API endpoints, or MCP-compatible tools. Enterprise features include role-based access control, SSO integration, and audit logging. A built-in marketplace enables teams to share and reuse model providers, tools, and workflow templates across projects. 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 an open-source community edition.

Instead of a separate chatbot for each channel, Synaplan runs one self-hosted AI assistant for your team, website visitors, WhatsApp line, and inbox, and shows what every reply cost. In the web app, connect OpenAI, Anthropic, Gemini, Groq, Mistral, xAI, Hugging Face, or local Ollama. Each kind of work gets its own model, from chat and image to video and audio, and every answer carries a Free, Low, Mid, or High cost badge. A planner splits a big request into a task graph and sends each step to the model that fits. On your website, one script tag adds a chat widget, and a human operator can take over any conversation live. On WhatsApp Business and email, the assistant replies on the channel where the question arrived. Behind every channel sits one knowledge base: upload PDFs, Word and Excel files, or images with OCR, and semantic search feeds them into answers. Saved tasks run on demand or on a schedule, approvals pause any action that writes, and watched pages mail you a diff when a URL changes. Connections reach Microsoft 365, Dropbox, Nextcloud, ownCloud, WebDAV, CalDAV, Jira, and Confluence. An MCP server and an Anthropic-compatible endpoint let Claude, Cursor, and Claude Code use your instance. Users and groups share folders and assistants, and admins see usage and health. 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.

An 8MB Go binary that boots in under one second, uses less than 10MB of RAM, yet delivers full AI agent capabilities across 16+ chat platforms simultaneously. PicoClaw connects to Telegram, Discord, Matrix, IRC, Slack, WeCom, DingTalk, WeChat, LINE, and QQ while supporting LLM providers spanning OpenAI, Anthropic, Gemini, DeepSeek, AWS Bedrock, Azure, and local models via Ollama. Native Model Context Protocol support enables standardized tool integration, and the built-in smart routing engine directs simple queries to lightweight models to reduce API costs while sending complex tasks to capable models. Tool capabilities include secure shell execution, filesystem access, web search, cron scheduling for recurring tasks, and sub-agent spawning with status tracking. Gateway mode transforms PicoClaw into a full AI backend with REST API endpoints accessible from any client. The Skills system loads hierarchical behavior definitions from SKILL.md files, enabling customizable agent personalities and workflows. Compiles for x86_64, ARM64, ARMv7, RISC-V, MIPS, and LoongArch, making it deployable on hardware as cheap as a $10 Sipeed LicheeRV Nano. Achieved nearly 30,000 stars within six months of its February 2026 release. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.