80 Open-Source AI Repos You Should Actually Use in 2026 (Agentic Coding, Memory, Video & More)

The average AI startup pays $8k–$50k/year on tool subscriptions. The people actually shipping — researchers, engineers, and indie hackers — have always relied on open source.

This is the updated 2026 edition of the open-source AI toolbox. I went through 400+ repositories and pulled the 80 most production-ready, actively maintained projects that have either landed or risen significantly since April.

Eight categories. Every entry includes a short summary and direct link. Star counts are approximate — they only go up.

Spotlight: OpenMontage (calesthio/OpenMontage)

~18.7K stars and currently #1 trending on GitHub.

Point Claude Code or Cursor at this repo and it researches a topic, writes + narrates a script, generates visuals, scores music, adds captions, and renders a finished video — end to end. A 60-second Pixar-style short can cost as little as $1.33. Fully local options available (Piper TTS + Remotion + FFmpeg). 12 pipelines and 400+ agent skills. AGPLv3.

This repo perfectly captures the thesis of the entire list.

01. Agentic Coding (CLI agents that actually ship)

Agents that take a goal, explore codebases, run commands, fix errors, and iterate until the task is done.

Cline — 4.6M VS Code installs. Plan/Act modes + MCP support.
Continue.dev — Works in VS Code & JetBrains. Excellent repo context.
Aider — Git-aware terminal pair programmer with auto-commits.
OpenHands (formerly OpenDevin) — Autonomous agent in Docker sandbox. Can use a browser.
Goose (Block) — Polished desktop + CLI agent with strong MCP support.
Plandex — Built for massive codebases (2M+ tokens).
Roo Code — Strong fork of Cline with extra autonomy modes.
GPT Pilot — Full app builder that acts as PM + architect + dev + reviewer.
Devika — Solid local OpenDevin alternative.
AgentZero — Learns and builds its own tools over time.

02. Vibe Coding (Text → Working Full-Stack App)

Describe what you want → get a real, editable application.

bolt.diy — Open-source bolt.new. WebContainers in the browser.
Onlook — "Cursor for designers" — visual React builder with AI.
OpenUI (Weights & Biases) — Text → HTML/React/Svelte with live editing.
OpenInterpreter — "OS for vibe coders." Run code, control files/browser from natural language.
Cofounder — Full-stack apps with auth, payments, and deploy.

03. Memory & Context (Long-term Memory for Agents)

The infrastructure layer every serious 2026 AI product needs.

Mem0 — One-line memory layer. Saves 30-40% on tokens in long sessions.
Letta (ex-MemGPT) — Three-tier memory (core + recall + archival).
Zep — Production memory with knowledge graphs and temporal awareness.
GraphRAG (Microsoft) — Knowledge graph generation + RAG from documents.
LlamaIndex — The go-to framework for RAG over your data.
LangChain — Still dominant for complex multi-step agent workflows.
RagFlow — Excellent at deep document understanding (tables, OCR, math).

04. Multi-Agent Frameworks

Coordinated teams of specialized agents.

CrewAI — Role-based teams (CEO + analyst + writer, etc.).
AutoGen (Microsoft) — Multi-agent conversations and group chat patterns.
LangGraph — State-machine framework. The production choice for complex workflows.
MetaGPT — Simulated company (CEO → CTO → Engineer → QA).
AutoGPT — The original agent that started the modern era. Still actively developed.
Composio — Plug 200+ real tools (Slack, Gmail, Stripe, etc.) into any agent.

05. Image & Video Generation (Local & Controllable)

No watermarks. No waitlists. Real models you control.

ComfyUI — Node-based workflow king. The full Midjourney pipeline, uncensored.
AUTOMATIC1111 — The classic Stable Diffusion WebUI with massive plugin ecosystem.
Forge — Faster and lower VRAM than A1111.
FLUX.1 (Black Forest Labs) — Currently one of the strongest open image models.
Mochi 1, CogVideoX, HunyuanVideo, LTX-Video — Strong open video generation options.

06. Voice AI (Realtime Agents)

Build your own voice agents from scratch.

Pipecat — Realtime STT + LLM + TTS pipelines (sub-500ms).
LiveKit Agents — Production-grade WebRTC + AI voice framework.
WhisperX — Word-level timestamps + speaker diarization.
OpenVoice — Voice cloning from a 10-second sample.
Fish Speech — Excellent real-time multilingual voice cloning.

07. Local LLM Inference & Routing

Run models locally or route intelligently across providers.

Ollama — Easiest way to run open models locally.
llama.cpp — The inference engine that powers almost everything.
vLLM — Production serving with PagedAttention.
LiteLLM — One SDK to call 100+ LLM providers (OpenAI-compatible proxy).
Exo — Run one model distributed across your devices (Mac + iPhone + old laptop).

08. AI Hubs & Local UIs

Self-hosted ChatGPT-style interfaces.

Open WebUI — Self-hosted ChatGPT for Ollama or any compatible backend. Multi-user + RAG.
LobeChat — Beautiful multi-provider interface with plugins and agents.
AnythingLLM — Local RAG + document chat. Multi-user ready.
Jan — 100% offline ChatGPT alternative with model marketplace.
GPT4All — Easy local LLM chat with built-in downloads.

Final Thoughts

Subscriptions are a habit.

Open source is a skill.

The best way to use this list is to pick three repos that match your current stack and actually install them this week. The landscape moves extremely fast — these tools are what serious builders are shipping with right now.

If you have more questions, please feel free to contact me at any time: https://t.me/FatherSon97