Awesome LLM Apps: 52 Ranked by What's Maintained (2026)

Search "awesome-llm-apps" and you land on one giant GitHub repo. Shubhamsaboo/awesome-llm-apps has crossed 115,000 stars (GitHub, June 2026), and it is genuinely excellent. But read the fine print: it bills itself as "100+ AI Agent and RAG apps you can actually run, clone, customize, ship." Those are starter templates you build from, organized by technique. They are not finished apps you install and use.

This is the other kind of directory. It is 52 real, runnable LLM apps that I either use myself or have graded against hard repo signals, sorted into 7 categories, with one column that does the actual work: a maintenance verdict. Maintained, Slowing, or Abandoned. That column is the point of the page, because the flat "awesome" lists will happily send you to a tool whose last commit landed in 2024. If you want the runtime landscape underneath all of this, the local engines and the hardware math, start with the complete 2026 guide to running LLMs locally; this directory assumes you just want the ranked list.

Key Takeaways

• The popular "awesome-llm-apps" repo (Shubhamsaboo, 115,000+ stars, June 2026) is a cookbook of templates you build, not a directory of finished apps you run. This page is the latter.
• Of the 52 apps I rank here, 42 are Maintained, 6 are Slowing, and 4 are Abandoned. That looks healthy only because I threw out the dead forks before counting.
• Last-commit recency beats star count. GPT4All sits on 77,000 stars but has been quiet since May 2025 (GitHub, June 2026); gpt-engineer and Verba are archived outright.
• Install official or actively-shipping first: Ollama or LM Studio to run models, Open WebUI or LibreChat to chat, AnythingLLM or Dify to build.

---

Table of Contents

• How Did I Rank These LLM Apps?
• Local Model Runners
• Chat UIs and Frontends
• RAG and Chat-With-Your-Docs Apps
• Agent Frameworks and No-Code Builders
• Coding Agents
• AI Search and Research Apps
• Voice, Speech, and Roleplay
• Do GitHub Stars Tell You Which App to Pick?
• What Does the Maintenance Split Tell You?
• How Do I Pick and Run One of These?
• Frequently Asked Questions
• The Bottom Line

How Did I Rank These LLM Apps?

I graded all 52 apps on one axis the awesome lists ignore: is anyone still shipping it? Four of the 52 are already dead, including two archived repos and one whose parent company folded in late 2023 (Analytics India Magazine). "Does it exist" is the wrong question in a space moving this fast. "Was it touched this month" is the right one.

So every entry carries one of three statuses, and I want to be honest about how I assigned them. I run roughly 20 of these in real projects, from Ollama and Open WebUI on my own machine to Dify and Cline at work. The rest I graded on repo signals you can check yourself in five minutes: who backs it, when the last commit landed, whether releases are tagged on a cadence or just trail off. Star counts I treat with suspicion, for reasons the chart below makes obvious.

Here is the rubric, plainly:

• Maintained. A first-party app from a company that has skin in the game, or a community project with commits in roughly the last six weeks, real releases, and a maintainer who answers issues. Production-considerable once you scope its credentials.
• Slowing. It still works, and several are still widely deployed, but the commit cadence has dropped off (no meaningful push in two to seven months) or the project is mid-transition to something new. Fine to use; keep a fallback in mind.
• Abandoned. Archived, orphaned by a shutdown, or quiet for nine months or more. Listed on purpose, so you recognize the dead fork before you copy a 2024 tutorial that points at it.

One caveat on the numbers. The star counts and last-commit dates here come straight from each project's GitHub page in June 2026, so they will drift after I publish; that drift is exactly why I lead with recency, not stars. Across the 52 apps, the split lands at 42 Maintained, 6 Slowing, and 4 Abandoned. Why does my list read 81% healthy when half of every "awesome" list is rot? Because this is a shortlist, not a census. The ratio you see is what survives the filter, which is the filter you wanted someone else to run for you.

Local Model Runners

If you want to run a model on your own hardware, this is where you start, and the category is dominated by one tool: Ollama has passed 175,000 GitHub stars (GitHub, June 2026), more than any other app in this directory except the automation platforms. These are the apps that pull weights, manage quantization, and expose a local API. Get this layer right and everything above it gets easier. For the deep dive on which runtime wins on your machine, the pillar covers llama.cpp versus Ollama versus vLLM in detail.

App	Maintainer	Status	Verdict
Ollama	Ollama (company)	Maintained	The default. One command to pull and run almost any open model, with a clean local API. If you run one tool locally, run this. Full walkthrough: the complete Ollama setup and model guide.
llama.cpp	ggml.org (community)	Maintained	The C++ engine half this ecosystem is built on, Ollama included. Run it raw only when you want to hand-tune quantization and squeeze a weak GPU.
LM Studio	LM Studio (company)	Maintained	The friendliest GUI for people who never touch a terminal. Closed source but free, with a model browser and a built-in server. Guide: downloading and running models in LM Studio.
Jan	Menlo Research	Maintained	Open-source LM Studio. A polished desktop app that runs models fully offline. The one I hand to people who distrust closed binaries.
text-generation-webui	oobabooga (community)	Maintained	The power user's cockpit. Every loader, every sampler, every extension. Overwhelming, but nothing else exposes this many knobs.
KoboldCpp	LostRuins (community)	Maintained	A single-file llama.cpp wrapper beloved by the local roleplay crowd. Tiny, fast to start, zero install ceremony.
vLLM	vLLM project	Maintained	Not a desktop app, a serving engine. When you outgrow Ollama and need real throughput across many users, this is the jump.
LocalAI	mudler (community)	Maintained	A drop-in OpenAI-compatible API you self-host. Best when you want existing apps to talk to local models with no code changes.
GPT4All	Nomic AI	Abandoned	Was the easiest on-ramp in 2023. The repo has been quiet since May 2025 (GitHub, June 2026) as Nomic moved on. Use Jan or LM Studio instead.

Chat UIs and Frontends

Once a model is running, you need something to talk to it, and this category is unusually healthy. Open WebUI leads it at over 143,000 GitHub stars (GitHub, June 2026), a full ChatGPT-style frontend that points at Ollama or any OpenAI-compatible endpoint. The rest split into self-hosted web apps and lightweight desktop clients. Pick by whether you want a server to run or a binary to download.

App	Maintainer	Status	Verdict
Open WebUI	Open WebUI (community)	Maintained	The category king. RAG, tools, multi-user, model management, all of it. The default UI for a local Ollama setup.
LibreChat	Danny Avila (community)	Maintained	The best multi-provider hub. One UI over OpenAI, Anthropic, Google, and local models, with agents and presets. My pick when I swap models all day.
Lobe Chat	LobeHub	Maintained	The prettiest of the bunch. Plugins, voice, vision, and a genuinely polished UX. A touch more opinionated than LibreChat.
Chatbox	Bin Huang (community)	Maintained	A cross-platform desktop client that just works. No server to run; point it at a key or a local endpoint and go.
Cherry Studio	Cherry HQ	Maintained	A desktop client that has quietly gotten excellent for knowledge work, with built-in agents and document chat.
big-AGI	Enrico Ros (community)	Maintained	Smaller and developer-flavored. Multi-model "beam" and persona features the bigger UIs still lack.

RAG and Chat-With-Your-Docs Apps

This is where most people actually want an LLM: pointed at their own documents. AnythingLLM has climbed past 62,000 GitHub stars (GitHub, June 2026) by being the most complete "chat with your files" app you can self-host. But this category also holds two of my four dead entries, so the maintenance column earns its keep here more than anywhere. RAG demos age fast.

App	Maintainer	Status	Verdict
AnythingLLM	Mintplex Labs	Maintained	The most complete chat-with-your-docs app. Workspaces, multiple vector DBs, desktop or Docker. The one I install for non-technical teams.
RAGFlow	InfiniFlow	Maintained	Deep-document RAG with serious parsing for tables, layouts, and OCR. Heavier to run, best results on messy PDFs.
Onyx	Onyx (company)	Maintained	Formerly Danswer. Enterprise-grade search and chat across your company's apps, with connectors for nearly everything.
Khoj	Khoj (company)	Maintained	Your second brain. Self-hostable, indexes your notes and files, and answers across them. Lovely for Obsidian users.
PrivateGPT	Zylon	Maintained	The name that launched the local-RAG wave. Still maintained by Zylon, still fully offline document Q&A.
Kotaemon	Cinnamon	Maintained	A clean open-source RAG UI with good defaults, inline citations, and an easy demo path.
Haystack	deepset	Maintained	deepset's production RAG framework, not an app. Reach for it when you are building a pipeline, not installing one.
Quivr	Quivr (company)	Slowing	Was a hot "second brain." The core repo has been quiet since mid-2025 (GitHub, June 2026) after the team pivoted to a RAG framework. Works, but watch it.
Verba	Weaviate	Abandoned	Weaviate's RAG showcase, now archived (GitHub, June 2026). A nice demo to read, not a thing to build on.

Agent Frameworks and No-Code Builders

If you want to build something rather than just chat, this is the deepest category in the directory, and the most volatile. The visual builders alone are enormous: Langflow sits above 150,000 GitHub stars and Dify above 146,000 (GitHub, June 2026). The split here is clean: code-first frameworks for developers, and drag-and-drop platforms for everyone else. Both halves are mostly Maintained, with two notable exceptions sliding toward Slowing. The model you wire into any of these matters as much as the framework, so see which open-source model to actually pick before you commit.

App	Maintainer	Status	Verdict
Dify	LangGenius	Maintained	The best open-source LLM app platform. Visual workflows, RAG, agents, and an API in one box, shipping constantly.
Langflow	Langflow (company)	Maintained	A polished node editor backed by a real company. Build agent flows visually, export an API.
n8n	n8n (company)	Maintained	Not LLM-specific, but at 194,000+ stars it has become the glue for production AI workflows. Self-hostable, endlessly useful.
LangChain	LangChain (company)	Maintained	The framework everyone complains about and still uses. 140,000+ stars. The default glue for LLM apps in Python and JS.
LlamaIndex	LlamaIndex (company)	Maintained	The data framework. If your app is mostly RAG, this is often a cleaner fit than LangChain.
CrewAI	CrewAI (company)	Maintained	The most popular multi-agent framework. Define roles, give them tools, let them collaborate. Pragmatic and well-documented.
AutoGPT	Significant Gravitas	Maintained	The 185,000-star project that started the autonomous-agent hype. Now a low-code platform; the original loop is mostly history.
Flowise	FlowiseAI	Maintained	Drag-and-drop LangChain. Build chatbots and agent flows visually, export an API. Great for prototypes and non-coders.
Letta	Letta (company)	Maintained	Formerly MemGPT. Treats long-term agent memory as a first-class concern. The interesting one to watch in stateful agents.
AutoGen	Microsoft	Slowing	Microsoft's multi-agent framework, now in maintenance mode. Microsoft directs new work to the Microsoft Agent Framework (Microsoft Learn, 2026); the community fork is AG2.
MetaGPT	DeepWisdom	Slowing	The "software company in a box" multi-agent demo. Impressive paper, but commits have thinned since early 2026 (GitHub, June 2026).

Coding Agents

Coding is the use case where open-source apps genuinely rival the paid tools, and the leader proves it: OpenHands has passed 78,000 GitHub stars (GitHub, June 2026) running as an autonomous agent that edits, tests, and iterates in a sandbox. This category lives or dies on the model behind it, so the freshness column matters less than which LLM you point it at. For that decision, I keep a separate scorecard.

App	Maintainer	Status	Verdict
OpenHands	All Hands AI	Maintained	Formerly OpenDevin. The leading open-source coding agent. Runs in a sandbox, plans, edits, tests, and iterates.
Cline	Cline (community)	Maintained	The best open-source coding agent inside VS Code. Plan and act modes, MCP support, bring your own key. My daily driver for autonomous edits.
Open Interpreter	Open Interpreter	Maintained	Lets a model run code on your machine to get things done. Powerful and a little terrifying, so sandbox it.
Goose	Block	Maintained	Block's open-source agent. Extensible via MCP, runs locally, and is genuinely good. The dark horse of this list.
Aider	Paul Gauthier (community)	Maintained	The terminal coding agent purists love. Git-aware, model-agnostic, no IDE required. Pairs beautifully with a strong model: the best LLM to drive a coding agent.
Continue	Continue (company)	Maintained	The open-source Copilot alternative for VS Code and JetBrains. Autocomplete plus chat, your model, your rules.
Tabby	TabbyML	Slowing	A self-hosted assistant with autocomplete and chat. Solid, but commits have thinned since March 2026 (GitHub, June 2026).
gpt-engineer	gpt-engineer org	Abandoned	The "describe an app, get a codebase" pioneer, now archived (GitHub, June 2026). Read it for history; use OpenHands or Aider to actually ship.

AI Search and Research Apps

This is the smallest category, but it is where open source quietly caught the commercial answer engines. These tools pair a model with a search backend to produce sourced answers or full reports instead of a wall of links. Most are Maintained, with one Stanford research project that is brilliant but should be treated as code, not a product. The model you choose drives the quality of the synthesis, which is its own rabbit hole.

App	Maintainer	Status	Verdict
GPT Researcher	Assaf Elovic (community)	Maintained	Point it at a question, get a sourced report. The most useful open research agent I have run.
Perplexica	ItzCrazyKns (community)	Maintained	An open-source Perplexity. Pairs a local model with SearXNG for private AI search. Very usable.
Morphic	Yuki Hattori (community)	Maintained	A clean generative answer-engine UI you can self-host or deploy on Vercel. Small but well-kept.
(https://github.com/stanford-oval/storm)	Stanford OVAL	Slowing	Stanford's research-to-article system. Brilliant as a paper and a demo; treat it as research code, quiet since fall 2025 (GitHub, June 2026).

Voice, Speech, and Roleplay

The last category is the most fragmented, which is why it has the lowest Maintained ratio in the directory. Speech and character apps are easy to start and hard to keep current. OpenAI's Whisper model alone has over 103,000 GitHub stars (GitHub, June 2026), but the interesting work happens in the wrappers around it, and one of the most beloved TTS projects here was orphaned when its company shut down.

App	Maintainer	Status	Verdict
whisper.cpp	ggml.org (community)	Maintained	Whisper transcription in portable C++. Runs on a laptop, a phone, a Pi. The speech-to-text backbone of countless local apps.
Whisper	OpenAI	Maintained	The open-weight STT model itself. The repo updates rarely because the model is finished, and the weights are everywhere.
SillyTavern	SillyTavern (community)	Maintained	The power-user frontend for AI roleplay and characters. Endless customization; pairs with KoboldCpp or any backend.
faster-whisper	SYSTRAN	Slowing	The fast CTranslate2 reimplementation everyone actually deploys for transcription. Still the workhorse, but pushes have slowed since late 2025 (GitHub, June 2026).
Coqui TTS	Coqui (defunct)	Abandoned	The open TTS standard, orphaned when Coqui the company shut down in late 2023 (Analytics India Magazine). Use the actively maintained Idiap community fork.

Do GitHub Stars Tell You Which App to Pick?

No, and the gap is wider than most directories admit. The top of the star chart is led by n8n at 194,000+ stars (GitHub, June 2026), a tool that is not even LLM-specific, followed by AutoGPT at 185,000, whose original autonomous loop is now mostly historical. Stars measure attention at the moment a project went viral. They do not measure whether anyone is still fixing bugs this quarter.

That mismatch is the single most useful thing to internalize before you install anything. A 50,000-star tool like Goose, shipping daily, will serve you better than a 185,000-star project coasting on a 2023 launch. The chart below ranks the eight most-starred apps in this directory. Read it as a popularity contest, not a quality ranking, and then cross-reference the maintenance column above before you commit.

What Does the Maintenance Split Tell You?

The clearest signal across all 52 apps is not which one is best. It is who stands behind it. Every Abandoned entry has the same backstory: a company that folded (Coqui), a vendor that archived its demo (Verba, from Weaviate), a pioneer that got superseded (gpt-engineer), or a single maintainer who moved on (GPT4All). Not one of them died because the idea was bad. They died because nobody had a standing reason to keep shipping.

That pattern repeats across the categories below. The deepest category, agent frameworks and builders, is also the one with the most company backing, which is exactly why it stays healthy. The thinnest categories, search and voice, carry the highest share of single-maintainer risk. Where would you rather place a production bet?

How Do I Pick and Run One of These?

Picking the app is half the job; the other half is matching it to a model and a runtime. The honest order of operations I follow: decide what you are building, pick the app from the category above, then choose the model that drives it. Most people skip the last step and wonder why a great app gives mediocre answers. The app is a shell. The model is the engine.

So if you are running anything locally, settle the runtime first. For most people that means Ollama for a clean command-line setup or LM Studio for a friendly GUI, both of which slot under the chat UIs and RAG apps above. Then pick the brain. If you are coding, the model choice swamps the app choice, so weigh it carefully in the best LLM for coding. If you are deciding between hosted frontier models to power an agent, the trade-offs in Claude Opus versus GPT-5 and DeepSeek R1 versus V3 will save you a few wrong turns. And if you want to stay fully open and self-hosted, start from the best open-source LLM and work back to the app.

Frequently Asked Questions

What is the awesome-llm-apps repo?

"awesome-llm-apps" usually means the GitHub repo Shubhamsaboo/awesome-llm-apps, a collection of 100-plus runnable AI agent and RAG app templates with full source code, now past 115,000 stars (GitHub, June 2026). It is a cookbook you build from. This directory ranks 52 finished apps you install instead.

What are the best open-source LLM apps in 2026?

For most people the best picks are the actively maintained leaders in each category: Ollama or LM Studio to run models, Open WebUI or LibreChat to chat, AnythingLLM or Dify to build, and OpenHands or Cline to code. All rank Maintained here, with Open WebUI alone above 143,000 GitHub stars (GitHub, June 2026).

How do I run these LLM apps locally?

Most of these apps need a model and a runtime underneath them. Install Ollama or LM Studio first, pull an open-weight model, then point a frontend like Open WebUI or AnythingLLM at the local endpoint. The pillar covers hardware requirements and runtime trade-offs in the guide to running LLMs locally.

Which LLM apps are abandoned or unsafe to rely on?

Four of the 52 here are dead: GPT4All (quiet since May 2025), Verba and gpt-engineer (both archived), and Coqui TTS (company shut down in late 2023, Analytics India Magazine). They still run, but nobody is patching them. For Coqui, use the Idiap community fork that picks up active maintenance.

Do GitHub stars mean an LLM app is good?

No. Stars measure attention at launch, not whether a project still ships. n8n leads this directory at 194,000+ stars yet is not LLM-specific, while AutoGPT's 185,000 stars sit on a loop that is now mostly historical (GitHub, June 2026). Check the last commit date before you trust the badge.

The Bottom Line

A useful LLM app directory is not the longest list. It is the one that already threw out the dead forks for you. The 52 apps here are the ones I would actually evaluate in 2026, and the most valuable column is the status: 42 Maintained, 6 Slowing, 4 Abandoned.

If you remember one thing, make it this: pick by who is still shipping, not by who went viral. Company-backed and actively-maintained apps cluster in the Maintained column for a reason, while the famous abandoned ones are famous precisely because they launched loud and then stopped. Start with a boring, well-kept app, add a riskier one only when it earns the slot, and check the last commit date either way. For the runtime and hardware layer underneath all of it, go back up to the complete guide to running LLMs locally.