What this article is
A no-marketing landscape of self-hostable AI agent and assistant projects in mid-2026, with star counts and categories from public GitHub data and Hostinger's application marketplace. Useful if you are choosing a project to deploy, comparing alternatives to OpenClaw or Hermes Agent, or just trying to make sense of a noisy category.
The seven categories
The self-hosted agent space is not one market; it is seven overlapping ones. The product that wins for a single user with a laptop is rarely the same as the one that wins for a 50-person customer support team. Use the categories below to narrow the shortlist before comparing features.
- Personal AI assistants (single-user, multi-channel, always-on): OpenClaw, Hermes Agent.
- LLM application platforms (build chatbots and agents with RAG + workflows): Dify, Flowise.
- Chat UI layers (ChatGPT-style front ends for any LLM): Open WebUI, LibreChat, Big-AGI.
- Stateful agent frameworks (production agents with memory and persistence): Letta, Cheshire Cat, Agent Zero.
- Workflow automation with AI (visual graphs that include AI steps): n8n.
- Self-hosted ChatGPT alternatives (team chat with RAG, RBAC, audit): Bionic GPT, AnythingLLM.
- Specialised agents (one job, done well): GPT-Researcher, Dograh, Paperclip.
The shortlist
Star counts captured from the GitHub API on the date of writing. They move fast in this category.
| Project | Category | Stars | One-line differentiator |
|---|---|---|---|
| openclaw/openclaw | Personal assistant | 379k+ | 22 channels, OpenAI/NVIDIA/GitHub-sponsored, npm-installable, lobster mascot |
| NousResearch/hermes-agent | Personal assistant | 199k+ | Self-improving learning loop, Honcho user modelling, Modal/Daytona hibernation |
| n8n-io/n8n | Workflow automation | 193k+ | Visual workflow tool with first-class AI nodes; not strictly an agent but used as the agent layer |
| langgenius/dify | LLM application platform | 146k+ | End-to-end RAG + agent + workflow builder; popular enterprise self-host |
| open-webui/open-webui | Chat UI layer | 142k+ | ChatGPT-style UI for any LLM (Ollama, OpenAI, etc.). Heavy plugin ecosystem. |
| Mintplex-Labs/anything-llm | Self-hosted ChatGPT alt. | 62k+ | All-in-one RAG + agents + chatbots, multi-user, with workspace permissions |
| FlowiseAI/Flowise | LLM application platform | 54k+ | Drag-and-drop low-code agent builder. Production deploys via a single Docker image. |
| danny-avila/LibreChat | Chat UI layer | 40k+ | Multi-LLM chat with plugins, OpenAI-compatible API. Closest to ChatGPT in feel. |
| letta-ai/letta | Stateful agents | 23k+ | Successor to MemGPT. Persistent memory layer with explicit memory blocks. |
| Cheshire Cat AI | Stateful agents | (active) | Production agent framework with built-in RAG, plugins, function calling |
| Agent Zero | Stateful agents | (active) | Open-source multi-agent cooperation with persistent memory |
| Bionic GPT | Self-hosted ChatGPT alt. | (active) | On-premise ChatGPT alternative with team chat, RAG assistants, RBAC |
| Big-AGI | Chat UI layer | (active) | Multi-model AI workspace with parallel comparison across 500+ LLMs |
| GPT-Researcher | Specialised agent | (active) | Autonomous deep-web research with cited reports |
| Dograh | Specialised agent | (active) | No-code platform for building AI voice agents |
| Paperclip | Stateful agents | (active) | AI/ML orchestration platform for autonomous teams |
Projects marked "(active)" are listed in the Hostinger application marketplace as one-click installs. Their GitHub stars were not separately queried for this article; check the upstream repo for the current number.
A short decision guide
Pick by job, not by star count.
- You want a personal assistant that talks to you on every channel and runs on your laptop or a $5 VPS: OpenClaw, then Hermes Agent.
- You want a team-grade chat platform on your infra (RAG, multiple users, RBAC): AnythingLLM, then Bionic GPT, then Open WebUI.
- You want to build a product feature (RAG over docs, agent over your data): Dify or Flowise. Dify if your team is comfortable with the more opinionated, full-platform approach; Flowise if you want a visual node editor.
- You need real long-term memory in your agent: Letta. Built explicitly for stateful conversations.
- You want to orchestrate AI inside broader business workflows: n8n. It is a workflow tool, not strictly an agent, but the AI nodes are first-class.
- You want comparison shopping across many LLMs in one UI: Big-AGI for parallel model comparison; LibreChat for ChatGPT-feel multi-provider chat.
- You want a research agent that produces cited reports: GPT-Researcher.
- You want voice agents: Dograh.
Six trends the data confirms
- Personal AI assistants are now the largest category. OpenClaw at 379k stars is larger than most general-purpose dev tools. The category was niche two years ago.
- Channel sprawl is table stakes. 22 messaging channels for OpenClaw, six plus CLI for Hermes Agent. iMessage, WeChat, QQ, Matrix, Nostr are all in scope, not just the obvious three.
- Skill marketplaces are emerging as platforms. The
agentskills.ioopen standard is now compatible with both OpenClaw and Hermes Agent;VoltAgent/awesome-openclaw-skillscatalogues 5,400+ filtered and categorised skills. This is the App Store moment for personal AI. - Memory is the new moat. Hermes Agent's closed learning loop, Letta's stateful blocks, OpenClaw's persistent conversations, Honcho's dialectic user modelling. Products without a memory story are losing.
- Cost-aware hibernation is becoming standard. Daytona, Modal, Cloudflare Containers'
SANDBOX_SLEEP_AFTER. Agents idle 90% of the time, so hibernation drops monthly cost from $30 to $5 to $10. - Big-name sponsorship signals the category has crossed the chasm. OpenAI, GitHub, NVIDIA, and Vercel openly sponsoring an MIT-licensed self-hosted agent project means this is no longer hobby tier; it is strategic for them.
What I would actually install today
If you have an hour and a credit card:
- For yourself, on your laptop:
npm install -g openclaw@latest && openclaw onboard. Pair Telegram. Done. - For your support team, on a managed VPS: Anything LLM via the Hostinger one-click. Multi-user, RAG over your docs, in under an hour.
- For your product, when you need bring-your-own-LLM customer support specifically: Open Agent (the Kiebot product). Drop a snippet, plug in your model key, live in ten minutes.
Where to read next
For step-by-step deploys of OpenClaw and Hermes Agent on AWS Lightsail, Hostinger, and Cloudflare, see three one-click paths to a self-hosted AI agent. For the architecture decisions when you actually integrate an agent into your product, see Designing AI Applications That Survive Production. For the specifics of the Kiebot side, see Inside Open Agent and Inside Kiework.
