# The free shop

URL: https://insidethemachine.ai/shop
Data as of 2026-10-26.

152 free AI models and 264 open-source tools. The page starts with only the parts that fit the reader's memory (the "Fits my computer" switch). Add parts to My setup, then check out. What fits each memory size, in full: https://insidethemachine.ai/index.md. Skill packs ranked by stars gained: https://insidethemachine.ai/skills.md.

## Our pick per memory size

The smartest model that fits and runs with Ollama on Mac, Windows and Linux. The chat setup sheets install it; the 24/7 helper below may use a smaller brain so the ears and voice fit next to it.

- 8 GB: Qwen3.5 4B, 23% of Claude Opus 5.5, needs 5.8 GB
- 16 GB: Gemma 4 12B, 25% of Claude Opus 5.5, needs 11.2 GB
- 24 GB: Gemma 4 12B, 25% of Claude Opus 5.5, needs 11.2 GB
- 32 GB: Qwen3.8 27B, 59% of Claude Opus 5.5, needs 22.7 GB
- 48 GB: Qwen3.8 27B, 59% of Claude Opus 5.5, needs 22.7 GB
- 64 GB: Qwen3.8 27B, 59% of Claude Opus 5.5, needs 22.7 GB
- 96 GB: Qwen3.8 27B, 59% of Claude Opus 5.5, needs 22.7 GB
- 128 GB: Qwen3.8 27B, 59% of Claude Opus 5.5, needs 22.7 GB
- 192 GB: Qwen3.8-Flash-Next (125B-A6B), 69% of Claude Opus 5.5, needs 140 GB
- 256 GB: Qwen3.8-Flash-Next (125B-A6B), 69% of Claude Opus 5.5, needs 140 GB
- 512 GB: Qwen3.8-Flash-Next (125B-A6B), 69% of Claude Opus 5.5, needs 140 GB

## Your 24/7 helper (FIG. 2a)

Rule: The brain, the ears (speech to text) and the voice (text to speech) stay loaded together. Ears and voice get up to 15% of the memory for AI each (at least 1 GB). The brain is the best model that can use tools, fits in the rest and runs with Ollama on Mac, Windows and Linux. To make an image or a video, the helper pauses the brain (ollama stop) and that model uses its space.

Set it up: 1. A private ChatGPT (https://insidethemachine.ai/do/chat) installs Ollama, the brain and a chat app. 2. Speech to text (https://insidethemachine.ai/do/voice) adds the ears. It installs Whisper, a different speech-to-text model from the one on the map; both turn speech into text. 3. Voice tools (https://insidethemachine.ai/setups/c/voice) list the text-to-speech apps for the voice. We have no recipe for it yet. 4. An agent that does tasks (https://insidethemachine.ai/do/agent) adds an always-on agent app from 32 GB. On 16 GB, use the chat app from step 1. (Shown for 16 GB, the page's default size.)

- Note, 8 GB: No video model in our list fits yet. The first memory size where one fits next to the ears and voice is 32 GB.
- Note, 16 GB: No video model in our list fits yet. The first memory size where one fits next to the ears and voice is 32 GB.
- Note, 24 GB: No video model in our list fits yet. The first memory size where one fits next to the ears and voice is 32 GB.

## The shop: free AI parts, picked for your computer

Every part here is free to download and runs on your own computer. Pick an aisle, add parts to "My setup" (saved in your browser), then check out right in the My setup panel: your computer (system and memory), how you use it (one way, the chat window picked) and how you run it (fully local, or Claude conducts). One email then gives one folder to download and one prompt for Claude Code or Codex that sets up every part. For an agent, the same answers as JSON: https://insidethemachine.ai/api/checkout?packs=<a,b>&mode=<local|claude>&way=<phone|chat|terminal|always>&os=<mac|windows|linux>&gb=<memory>&brain=<model id> gives the setup's prompt, its starter name and its sheet (every step). The full checkout page is https://insidethemachine.ai/checkout (https://insidethemachine.ai/checkout?packs=<a,b>&brain=<model id>&from=cart#start opens it with the cart's packs and brain). The folder: the folder is the sheet's starter folder plus MY-CART.md, one section per part (what it is, whether it fits, how to get it), at https://insidethemachine.ai/starter/cart/<starter name>~<part ids joined by +>.zip. Nothing to pay. Our app for every setup is Hermes Agent; what it already does is never sold here as an add-on.

Fits: the memory a part needs must be at most the usable share of your memory (75%). The cards mark it with three squares: runs well; tight (it fits, but the memory is nearly full); needs more. Text scores: Artificial Analysis Intelligence Index as a share of the best paid AI (Claude Opus 5.5), 2026-10-15. Sizes: Ollama library, 2026-10-26. Pictures and video: images.json, 2026-10-16. Voice and ears: speech.json, 2026-10-23. Downloads: Hugging Face, last 30 days. Unknown says unknown.

Can (text models): Writes code: it is made for code (models.json: the coding category; it is in the Code aisle too). Sees images: you can give it pictures or screenshots (models.json: the vision tag or category, or Image in its inputs). Hears audio: you can give it sound (models.json: the audio tag or category, or Audio in its inputs). Uses tools: it can call tools, so an app like Hermes Agent can let it search, read files and more (Ollama tag: tools, or the tool-calling category). Thinks: it can think step by step before it answers (Ollama tag: thinking). Very long docs: it takes in 1 million tokens or more at once (context_tokens.native, from its HF config.json). No badge means our data does not say it can; "unknown: …" means models.json has no capability tags (and for images, no inputs) for that model.

What Hermes Agent already does (no add-on needed): Read and write your files; Remember you; Search the web and read pages (its own docs disagree); Use your Gmail, calendar and Drive; Read your GitHub; Use your Notion; Use a web browser (its own docs disagree); Run commands; Look at pictures; Speak out loud; Do jobs on a schedule; Send out helper agents (its own docs disagree); Control your desktop (its own docs disagree); Skills (ready-made know-how) (its own docs disagree). (It loads skills itself; the Skills aisle sells skill packs for it to load.)

### Text & chat (79)

Chats, writes and answers questions.

### Code (23)

Writes and fixes code.

### Images (24)

Makes pictures from your words.

### Video (6)

Makes short clips.

### Voice (22)

Speech out: reads text out loud.

### Ears (21)

Speech in: turns talk into text.

### Open-source tools

237 free, open-source tools in 19 aisles (catalog.json; 4 cloud-only or closed tools are left out of the shop). Add one to My setup and checkout puts it in MY-CART.md with its install lines from its README, which we have not tested. Memory: Hermes Agent remembers you already, so those tools are not sold as add-ons. Coding agents, General & personal agents and Chat UIs are other apps like Hermes Agent: shown to compare, not added.

- **Runtimes & inference** (19): Runs open models on your own hardware. Most stars: Ollama (in every setup already) (184,859 stars), llama.cpp (132,426 stars), PyTorch (104,960 stars), vLLM (93,863 stars), LocalAI (49,849 stars). All: https://insidethemachine.ai/setups/c/runtimes
- **Model hubs** (9): Where you find and download weights. Most stars: Transformers (167,719 stars), Unsloth notebooks (5,738 stars), Hugging Face Hub, Ollama Library, LM Studio model catalog. All: https://insidethemachine.ai/setups/c/hubs
- **Models from the channel** (6): Featured on Inside the Machine; check the fit before you download. Most stars: Qwen-Image-2.1-Turbo, Mellum2.1 12B-A2.5B, EmbeddingGemma 2, Kolibri-1, AstaBrief 8B. All: https://insidethemachine.ai/setups/c/models
- **Coding agents** (13, other apps like Hermes Agent): Turns a model into something that edits your code. Most stars: OpenCode (213,869 stars), Codex CLI (129,383 stars), Pi (pi-mono) (115,690 stars), OpenHands (90,983 stars), Cline (70,365 stars). All: https://insidethemachine.ai/setups/c/coding
- **General & personal agents** (14, other apps like Hermes Agent): Lets a model take actions for you, not just chat. Most stars: OpenClaw (392,108 stars), Hermes Agent (in every setup already) (254,350 stars), AutoGPT (187,599 stars), Browser Use (118,088 stars), DeerFlow (83,911 stars). All: https://insidethemachine.ai/setups/c/general
- **Agent frameworks** (14): Building blocks for your own agents. Most stars: AutoGen (61,426 stars), LiteLLM (61,329 stars), CrewAI (59,740 stars), LangGraph (43,319 stars), Agno (42,754 stars). All: https://insidethemachine.ai/setups/c/frameworks
- **Memory** (11, already in Hermes Agent): Lets your agent remember across sessions. Most stars: claude-mem (100,993 stars), Mem0 (67,235 stars), Cognee (32,356 stars), Graphiti (31,795 stars), Supermemory (31,236 stars). All: https://insidethemachine.ai/setups/c/memory
- **RAG & document parsing** (14): Feeds your own documents to the model. Most stars: MarkItDown (190,405 stars), RAGFlow (92,283 stars), PaddleOCR (91,178 stars), MinerU (81,815 stars), Docling (68,936 stars). All: https://insidethemachine.ai/setups/c/rag
- **Routing & gateways** (10): Sends each request to the right model. Most stars: New API (50,036 stars), One API (37,168 stars), Portkey Gateway (13,194 stars), Bifrost (8,833 stars), Plano (7,090 stars). All: https://insidethemachine.ai/setups/c/routing
- **MCP servers & tools** (14): Gives the agent hands: web, files, apps. Most stars: n8n (207,507 stars), Firecrawl (191,861 stars), Awesome MCP Servers (96,394 stars), MCP Servers (official) (91,287 stars), Crawl4AI (85,617 stars). All: https://insidethemachine.ai/setups/c/mcp
- **Chat UIs & front-ends** (14, other apps like Hermes Agent): A friendly window onto your local models. Most stars: Open WebUI (155,901 stars), NextChat (88,853 stars), LobeChat (83,250 stars), Cherry Studio (52,668 stars), text-generation-webui (47,730 stars). All: https://insidethemachine.ai/setups/c/ui
- **Voice: speech-to-text & TTS** (14): Talk to it and hear it answer, offline. Most stars: Whisper (110,561 stars), whisper.cpp (54,418 stars), ChatTTS (39,909 stars), Handy (33,870 stars), Fish Speech (33,167 stars). All: https://insidethemachine.ai/setups/c/voice
- **Image & video generation** (14): Make images and video on your own machine. Most stars: Stable Diffusion WebUI (165,059 stars), ComfyUI (137,692 stars), Fooocus (53,435 stars), Diffusers (34,726 stars), Open-Sora (29,873 stars). All: https://insidethemachine.ai/setups/c/image
- **Fine-tuning** (14): Teach a model your data and style. Most stars: Unsloth (78,064 stars), LLaMA-Factory (75,495 stars), nanoGPT (63,835 stars), nanochat (58,660 stars), Open R1 (26,481 stars). All: https://insidethemachine.ai/setups/c/finetune
- **Quantization** (11): Shrinks models so bigger ones fit in memory. Most stars: bitsandbytes (8,527 stars), Core ML Tools (5,435 stars), TensorRT Model Optimizer (5,326 stars), LLM Compressor (3,878 stars), Optimum (3,503 stars). All: https://insidethemachine.ai/setups/c/quant
- **Evals & benchmarks** (14): Measures how close you are to the frontier. Most stars: promptfoo (26,019 stars), OpenAI Evals (19,607 stars), DeepEval (18,890 stars), Ragas (16,034 stars), LM Evaluation Harness (14,234 stars). All: https://insidethemachine.ai/setups/c/evals
- **Scaffolds & templates** (11): Reusable instructions that make agents behave. Most stars: Spec Kit (141,412 stars), OpenSpec (72,037 stars), Awesome Claude Code (55,762 stars), BMAD Method (54,187 stars), Awesome Cursor Rules (40,912 stars). All: https://insidethemachine.ai/setups/c/scaffolds
- **Observability** (12): See what your agent did and what it cost. Most stars: Langfuse (35,875 stars), MLflow (28,399 stars), Opik (22,627 stars), Phoenix (11,858 stars), Evidently (7,991 stars). All: https://insidethemachine.ai/setups/c/obs
- **Hardware & system utilities** (9): Know what your machine can run, and watch it work. Most stars: Apple container (50,702 stars), Stats (42,577 stars), llmfit (37,950 stars), nvtop (11,089 stars), nvitop (7,210 stars). All: https://insidethemachine.ai/setups/c/hw

### Skills (27)

A skill is a folder with a SKILL.md file: instructions that teach an agent one job. Hermes Agent, our app, loads skills itself (`hermes skills`). Each pack below is a GitHub repo with many skills: add it to My setup and MY-CART.md tells the agent to install only the skills the person needs. Install lines come from the Hermes Agent skills docs and the `npx skills` docs (vercel-labs/skills), read 2026-10-21; we have not tested them. Skill counts: SKILL.md files in the repo on GitHub. All: https://insidethemachine.ai/setups/c/skills

#### For everyone (5)

Office files, Hugging Face and science. Good first skills.

#### Content creation (8)

Make images, video and voice, and write better prompts for them. All but HyperFrames call a paid cloud service: you need an account there, and each image or video costs money.

#### For coding (5)

Make a coding agent plan, test, review and find bugs.

#### For building on one platform (9)

Each pack teaches your agent one company's product. Most of those products are cloud services.

## The weekly email: The Local Drop

The Local Drop · free, weekly. Every new AI you can run at home. One email a week. New models, tools and hardware, and what each one needs to run. Sign up on this page: https://insidethemachine.ai/#report. Gave your email at checkout? You are already on the list.

## Where next

- Learn, with How good is local AI? (the smartest model for each memory size, as a share of the best paid AI, Claude Opus 5.5): https://insidethemachine.ai/learn.md, https://insidethemachine.ai/best.md. The plan for our own Gap Score: https://insidethemachine.ai/gap-score.md
- The directory: The shop https://insidethemachine.ai/shop · Skills https://insidethemachine.ai/skills · Models (130 open text models) https://insidethemachine.ai/models · Agent apps (36 apps) https://insidethemachine.ai/harnesses · Machines (56 machines) https://insidethemachine.ai/machines · Tools (268 tools) https://insidethemachine.ai/setups · Recipes (6 recipes) https://insidethemachine.ai/do · Scan my machine https://insidethemachine.ai/my · Best model for your RAM https://insidethemachine.ai/best · Compare apps https://insidethemachine.ai/compare · New this week https://insidethemachine.ai/changes