Why this list

The average AI startup pays $8,000–$50,000/year for tool subscriptions. The people actually shipping AI - researchers, engineers, indie hackers - use open-source. They always have.
In April I published 69 open-source AI repositories. That post got 691,000 views. Two months later the landscape moved enough that I had to write a follow-up. This is the second pass.
I went through 400+ repositories. Pulled out 80 that are production-ready, actively maintained, and that have either landed or risen since April. Eight categories. Ten repos each. Every entry has a one-paragraph summary and a link.
This is the toolbox.

How to read this list

Each category opens with one paragraph of context and one screenshot marker. The 10 repositories underneath are ranked roughly by current relevance, not by star count.
Star counts are stated as approximate. By the time you read this, every number is bigger.
If a repository earned its spot, the list owes it a fair entry. If I missed one, my Twitter DMs are open.

★ SPOTLIGHT - OpenMontage (calesthio/OpenMontage)

~18.7K stars, +9,410 in a single week - #1 trending on GitHub right now. Your AI coding assistant is now a film studio.

Point Claude Code or Cursor at this repo and the same agent researches the topic, writes and narrates the script, generates the visuals, scores the music, burns in captions, and renders a finished video - end to end, from one sentence. A 60-second Pixar-style short costs $1.33. A Ghibli-style scene costs $0.15. With zero API keys it still ships real video (Piper TTS + Archive.org footage + Remotion + FFmpeg). 12 pipelines, 52 tools, 400+ agent skills. AGPLv3. The whole thesis of this list in one repo: the work that used to need an editor, a stock-footage subscription, and a $50/mo AI-video tool now runs from your coding agent for the price of a coffee.
repo: https://github.com/calesthio/OpenMontage

01. Agentic Coding (CLI agents that ship code for you)

Agents that take a goal in plain English, read files, run commands, fix errors, and keep going. The closest thing to "hire a junior dev" that ships today.

Cline (saoudrizwan/cline)

4.6M VS Code installs. Agent panel with Plan/Act modes, auto-approve safe commands, MCP support. Works with any LLM provider including OpenRouter :free models.
repo: https://github.com/cline/cline

Continue.dev (continuedev/continue)

33.8K stars. YC-backed. Works in both VS Code AND JetBrains. One founding engineer left to build OpenClaw. Tab autocomplete + chat + repo context.
repo: https://github.com/continuedev/continue

Aider (Aider-AI/aider)

35K stars. Git-aware terminal pair programmer. Auto-commits every change so you can git revert any AI edit. Repo-map keeps it oriented across multi-file changes.
repo: https://github.com/Aider-AI/aider

OpenHands (All-Hands-AI/OpenHands)

60K stars (formerly OpenDevin). Autonomous agent in a Docker sandbox. Can drive a browser, execute code, read its own errors. The open-source Devin everyone wanted.
repo: https://github.com/All-Hands-AI/OpenHands

Goose (block/goose)

14K stars. Block (formerly Square) shipped one of the most polished open-source agents on the market. Desktop app + CLI + MCP support.
repo: https://github.com/block/goose

Plandex (plandex-ai/plandex)

12K stars. Built for tasks spanning 2M+ tokens and hundreds of files. Branches like git - fork a plan, try a different approach, merge the better one.
repo: https://github.com/plandex-ai/plandex

Roo Code (RooCodeInc/Roo-Code)

Active fork of Cline with extra modes (Architect, Ask, Code, Debug). Heavier on agent autonomy. Same model-agnostic approach.
repo: https://github.com/RooCodeInc/Roo-Code

GPT Pilot (Pythagora-io/gpt-pilot)

32K stars. Builds full apps from a spec by acting as Product Owner + Architect + Developer + Reviewer in a loop.
repo: https://github.com/Pythagora-io/gpt-pilot

Devika (stitionai/devika)

19K stars. Local OpenDevin alternative. Plans, researches, codes. Self-hosted.
repo: https://github.com/stitionai/devika

AgentZero (frdel/agent-zero)

9K stars. Personal agentic framework that learns from its own runs. Builds its own tools as it goes.
repo: https://github.com/frdel/agent-zero

02. Vibe Coding (text → working full-stack app)

You describe the app in 2 sentences. It builds the UI, the backend, the database wiring, and deploys to a free subdomain.

bolt.diy (stackblitz-labs/bolt.diy)

12K stars. Open-source bolt.new. WebContainers in the browser. Build full-stack apps with any LLM provider you plug in.
repo: https://github.com/stackblitz-labs/bolt.diy

Onlook (onlook-dev/onlook)

27K stars. "Cursor for designers". Open-source visual builder for React. Drag-drop with AI suggestions in real time.
repo: https://github.com/onlook-dev/onlook

OpenUI (wandb/openui)

21K stars. From Weights & Biases. Describe a UI in plain text, get HTML/React/Svelte back, edit live.
repo: https://github.com/wandb/openui

Webdraw (cantor-systems/webdraw)

Hand-drawn sketches → working web apps. Upload a napkin sketch, get a Next.js app.
repo: https://github.com/cantor-systems/webdraw

Tempo (tempolabs-ai/tempo)

Visual editor + AI codegen for React. Bidirectional - edit code OR canvas, both stay in sync.
repo: https://github.com/tempolabs-ai/tempo

GibsonAI (gibson-ai/cli)

AI database designer. Describe the schema in English, get migrations + ORM code + admin panel.
repo: https://github.com/gibson-ai/cli

Cofounder (raidendotai/cofounder)

8K stars. Full-stack app generator with auth, payments, deploy. Output is real Next.js code.
repo: https://github.com/raidendotai/cofounder

Vapi-OSS alternatives (livekit/agents)

Voice-AI vibe coding. Build voice agents end-to-end with LiveKit Agents framework.
repo: https://github.com/livekit/agents

Magic Patterns CLI (magicpatterns/cli)

Component-level vibe coding. Generate React components matching your existing design system.
repo: https://github.com/magicpatterns/cli

OpenInterpreter (KillianLucas/open-interpreter)

60K stars. Run code locally from natural language. "OS for vibe coders" - works with files, shell, browser.
repo: https://github.com/KillianLucas/open-interpreter

03. Memory & Context (long-term memory for AI)

Agents that remember conversations from months ago. The infrastructure layer behind every serious AI product in 2026.

Mem0 (mem0ai/mem0)

52K stars. Memory layer for AI agents. Plugs in with one line. Saves 30-40% on long-session tokens.
repo: https://github.com/mem0ai/mem0

Letta (letta-ai/letta)

16K stars (ex-MemGPT). Three-tier memory: core, recall, archival. From UC Berkeley research.
repo: https://github.com/letta-ai/letta

Zep (getzep/zep)

3K stars. Production memory layer. Knowledge graphs + temporal memory. Used by enterprise AI products.
repo: https://github.com/getzep/zep

Cognee (topoteretes/cognee)

Semantic memory that builds its own knowledge graph from conversation history.
repo: https://github.com/topoteretes/cognee

GraphRAG (microsoft/graphrag)

23K stars. Microsoft Research. Knowledge graph generation from documents + RAG retrieval on top.
repo: https://github.com/microsoft/graphrag

LlamaIndex (run-llama/llama_index)

37K stars. The de-facto framework for RAG over your data. Connectors to anything.
repo: https://github.com/run-llama/llama_index

LangChain (langchain-ai/langchain)

98K stars. Despite the memes - still the most-used. Especially good for chains over multiple LLM calls.
repo: https://github.com/langchain-ai/langchain

RagFlow (infiniflow/ragflow)

32K stars. Deep document understanding + RAG. Strong at OCR, tables, math.
repo: https://github.com/infiniflow/ragflow

R2R (SciPhi-AI/R2R)

4K stars. Production-grade RAG with hybrid search, knowledge graphs, agents.
repo: https://github.com/SciPhi-AI/R2R

Embedchain (embedchain/embedchain)

23K stars. Embedding-first framework. Drop a URL/PDF/video, get a chatbot trained on it.
repo: https://github.com/embedchain/embedchain

04. Multi-Agent Frameworks (coordinated AI teams)

Several agents specializing in different roles, talking to each other, splitting tasks. Where 2026 agent stacks live.

CrewAI (crewAIInc/crewAI)

30K stars. Role-based agent teams. CEO + analyst + writer collaborate to ship reports.
repo: https://github.com/crewAIInc/crewAI

AutoGen (microsoft/autogen)

45K stars. Microsoft. Multi-agent conversations, code execution, group chat patterns.
repo: https://github.com/microsoft/autogen

LangGraph (langchain-ai/langgraph)

10K stars. State-machine framework for agents. The serious production choice for multi-step workflows.
repo: https://github.com/langchain-ai/langgraph

AgentScope (modelscope/agentscope)

7K stars. From Alibaba ModelScope. Visual debugging + multi-agent simulation.
repo: https://github.com/modelscope/agentscope

Swarm (openai/swarm)

17K stars. OpenAI's experimental lightweight multi-agent orchestration.
repo: https://github.com/openai/swarm

Agno (formerly Phidata)

20K stars. Multi-modal agent framework with built-in memory, knowledge, tools.
repo: https://github.com/agno-agi/agno

MetaGPT (geekan/MetaGPT)

62K stars. CEO + CTO + Engineer + QA simulated team. First repo that made multi-agent feel real.
repo: https://github.com/geekan/MetaGPT

AutoGPT (Significant-Gravitas/AutoGPT)

175K stars. The classic. Still actively developed. The agent that started the agent era.
repo: https://github.com/Significant-Gravitas/AutoGPT

Composio (ComposioHQ/composio)

20K stars. Plug 200+ tools into agents (Slack, Gmail, Notion, Stripe) with one SDK.
repo: https://github.com/ComposioHQ/composio

CamelAI (camel-ai/camel)

5K stars. Communicative agent framework. Strong at role-playing scenarios.
repo: https://github.com/camel-ai/camel

05. Image & Video Generation

Pipelines you control end-to-end. No Midjourney watermarks. No Sora waitlist. Real models running on real hardware.

ComfyUI (comfyanonymous/ComfyUI)

60K stars. Node-based image workflow engine. The whole Midjourney pipeline, but yours, and uncensored.
repo: https://github.com/comfyanonymous/ComfyUI

AUTOMATIC1111 (Stable Diffusion WebUI)

147K stars. The classic SD UI. Massive plugin ecosystem.
repo: https://github.com/AUTOMATIC1111/stable-diffusion-webui

Forge (lllyasviel/stable-diffusion-webui-forge)

9K stars. AUTOMATIC1111 fork with 30-75% faster inference, lower VRAM usage.
repo: https://github.com/lllyasviel/stable-diffusion-webui-forge

InvokeAI (invoke-ai/InvokeAI)

24K stars. Production-grade SD UI with canvas, unified canvas, board organization.
repo: https://github.com/invoke-ai/InvokeAI

Mochi 1 (genmoai/models)

From Genmo. State-of-art open video generation. 480p 30 fps at 5.4 seconds.
repo: https://github.com/genmoai/models

CogVideoX (THUDM/CogVideoX)

11K stars. Tsinghua + Zhipu AI. Open-source video generation, 5/10/20 second clips.
repo: https://github.com/THUDM/CogVideoX

HunyuanVideo (Tencent/HunyuanVideo)

10K stars. Tencent's 13B parameter open video model. Competitive with Sora on benchmarks.
repo: https://github.com/Tencent/HunyuanVideo

LTX-Video (Lightricks/LTX-Video)

6K stars. Lightricks. Real-time video generation at 30 fps.
repo: https://github.com/Lightricks/LTX-Video

Allegro (rhymes-ai/Allegro)

From Rhymes AI. 6-second 720p videos. Strong character consistency.
repo: https://github.com/rhymes-ai/Allegro

FLUX.1 (black-forest-labs/flux)

23K stars. Black Forest Labs. The image model that beat Midjourney on quality benchmarks. Free schnell variant available.
repo: https://github.com/black-forest-labs/flux

06. Voice AI

Realtime voice agents you can build from scratch. STT + LLM + TTS pipelines. The pre-Vapi layer.

Pipecat (pipecat-ai/pipecat)

5K stars. From Daily.co. Realtime voice agents - STT + LLM + TTS + tools. Sub-500ms round-trip.
repo: https://github.com/pipecat-ai/pipecat

LiveKit Agents (livekit/agents)

5K stars. LiveKit's voice-agent framework. Production-grade WebRTC + AI pipeline.
repo: https://github.com/livekit/agents

WhisperX (m-bain/whisperX)

15K stars. Whisper with word-level timestamps + speaker diarization. The transcription standard.
repo: https://github.com/m-bain/whisperX

OpenVoice (myshell-ai/OpenVoice)

30K stars. Voice cloning from a 10-second sample. Cross-lingual works. Apache 2.0.
repo: https://github.com/myshell-ai/OpenVoice

Coqui XTTS-v2 (coqui-ai/TTS)

37K stars. Multilingual TTS with voice cloning. Open-source standard for self-hosted voice.
repo: https://github.com/coqui-ai/TTS

Fish Speech (fishaudio/fish-speech)

20K stars. SOTA Chinese + English voice cloning. Real-time generation on consumer GPU.
repo: https://github.com/fishaudio/fish-speech

F5-TTS (SWivid/F5-TTS)

10K stars. Flow-matching TTS. Cleaner output than older diffusion-based models.
repo: https://github.com/SWivid/F5-TTS

Spark-TTS (SparkAudio/Spark-TTS)

5K stars. Sparse audio codec + LLM-based TTS. Tiny model, great quality.
repo: https://github.com/SparkAudio/Spark-TTS

Amphion (open-mmlab/Amphion)

8K stars. Speech + music + audio generation toolkit. MaskGCT and other SOTA models inside.
repo: https://github.com/open-mmlab/Amphion

Bark (suno-ai/bark)

37K stars. From Suno. Text-to-audio with emotion, music, sound effects. Foundation model.
repo: https://github.com/suno-ai/bark

07. Local LLM Inference & Routing

How the model actually runs on your machine - or how multiple models become one endpoint.

Ollama (ollama/ollama)

100K+ stars. The easiest way to run open-source LLMs locally. One CLI, hundreds of models.
repo: https://github.com/ollama/ollama

llama.cpp (ggerganov/llama.cpp)

70K stars. The inference engine under everything. CPU + GPU + Metal support.
repo: https://github.com/ggerganov/llama.cpp

vLLM (vllm-project/vllm)

32K stars. Production inference server. PagedAttention. The choice when you serve at scale.
repo: https://github.com/vllm-project/vllm

LM Studio (closed-source but worth mention)

Desktop app for running local LLMs. Polished UI, but closed. The non-CLI default for non-devs.
repo: https://lmstudio.ai/

LiteLLM (BerriAI/litellm)

15K stars. One Python SDK to call 100+ LLM providers. Drop-in OpenAI proxy.
repo: https://github.com/BerriAI/litellm

MLC LLM (mlc-ai/mlc-llm)

18K stars. Apache TVM team at CMU. Compile LLMs to native code for ANY device (Android, iOS, web).
repo: https://github.com/mlc-ai/mlc-llm

SGLang (sgl-project/sglang)

5K stars. Structured generation framework + fast serving. Used by many serious inference deployments.
repo: https://github.com/sgl-project/sglang

Exo (exo-explore/exo)

18K stars. Distribute one model across your Mac + iPhone + old laptop. Privacy + zero hardware budget.
repo: https://github.com/exo-explore/exo

Text Generation WebUI (oobabooga)

39K stars. "Oobabooga". Multi-model UI with API, character cards, training. Power-user choice.
repo: https://github.com/oobabooga/text-generation-webui

TabbyAPI (theroyallab/tabbyAPI)

Fast inference server with OpenAI-compatible API. Used heavily in roleplay / character communities.
repo: https://github.com/theroyallab/tabbyAPI

08. AI Hubs & Local UIs

Where you actually chat. Self-hosted ChatGPT clones. The reason your subscription is optional.

Open WebUI (open-webui/open-webui)

88K stars. Self-hosted ChatGPT for Ollama or any OpenAI-compatible endpoint. Multi-user, RAG, image gen.
repo: https://github.com/open-webui/open-webui

LobeChat (lobehub/lobe-chat)

48K stars. Polished multi-provider chat. Plugins, agents, knowledge base.
repo: https://github.com/lobehub/lobe-chat

AnythingLLM (Mintplex-Labs/anything-llm)

27K stars. Local RAG assistant + multi-user. Drop any document, get an AI answering against it.
repo: https://github.com/Mintplex-Labs/anything-llm

Cherry Studio (CherryHQ/cherry-studio)

20K stars. Desktop multi-LLM client. Stunning UI, supports every major provider + local.
repo: https://github.com/CherryHQ/cherry-studio

Chatbox (chatboxai/chatbox)

33K stars. Desktop AI chat client (Win/Mac/Linux). Beginner-friendly, polished.
repo: https://github.com/chatboxai/chatbox

Jan (janhq/jan)

26K stars. Open-source ChatGPT alternative running 100% offline. Built-in model marketplace.
repo: https://github.com/janhq/jan

GPT4All (nomic-ai/gpt4all)

73K stars. Local LLM chat with built-in model downloads. Cross-platform.
repo: https://github.com/nomic-ai/gpt4all

NextChat (ChatGPTNextWeb/ChatGPT-Next-Web)

78K stars. Self-host your own ChatGPT-like UI in one click. Vercel-friendly.
repo: https://github.com/ChatGPTNextWeb/ChatGPT-Next-Web

Big-AGI (enricoros/big-AGI)

5K stars. Power-user AI chat with personas, beam search, multi-model parallel runs.
repo: https://github.com/enricoros/big-AGI

Page Assist (n4ze3m/page-assist)

7K stars. Browser extension turning Ollama into a sidebar AI on every page.
repo: https://github.com/n4ze3m/page-assist

Closing notes

Run as many of these as your laptop allows. The point of the list isn't to bookmark it - it's to actually install the three that match your stack and see what changes.
Subscriptions are a habit. Open source is a skill.

• @seelffff