An app store for local models — pull an open-source LLM and run it entirely on your own machine, free and offline.
Pick this whenPrivacy, zero marginal cost, or uncensored models matter more than raw capability — and you have the RAM.
What makes it different
One command to pull and run a model (`ollama run <model>`), which is the whole install experience
Plugs straight into n8n as a chat-model node via a local URL on the standard port
Uncensored community models that no hosted API will serve you
What it does
Downloads and runs open-source models locally (Dolphin, Llama, Gemma, Qwen and the rest)
Serves them on a local endpoint that other tools can call
Completes the fully-local automation stack: n8n + Firecrawl + Ollama, no API bills
What it does not do
Common wrong expectationsDoesn't run well on modest hardware — a 7B model was sluggish on an 18GB MacBook, and Intel laptops were steered to a rented VM
Doesn't remove all costs — you still need a paid coding agent to build against it
Doesn't match frontier models; the file-deletion horror stories in BC5 came from open models driving tooling
Where it sits in a stack
The bottom of the model shelf: cheap, private, always-on inference for high-volume or sensitive workflow steps.
ShelfModels & model access · Models & model access
Our status · watchNo local-inference need yet, but the fully-local stack is the obvious fallback if API costs ever bite.
For your projectsPlausible for bulk, low-stakes passes over the corpus — first-pass transcript condensing or tagging — where quality matters less than volume and Fable budget matters a lot.
| Session | Coverage | Moment | What happened |
|---|
| Basecamp 3: Introduction to n8n — I | demonstrated | 2:15:55 | Installed live, pulled the Dolphin uncensored model in the terminal (`ollama run`), explained as app store for local models; n8n Ollama chat-model node credential = local URL, standard port, no API key |
| Day 3: Markdown Prompting, Content DNA and Skills, MCP Everywhere (Vapi Call, Zomato, Apify), Goose and Local Models, a Voice Agent, and the Full n8n Customer-Support Build | demonstrated | | Local models |
| Day 2: Custom GPTs and Gems, Markdown Prompting, Cloning Your Writing Style, MCP in Daily Use, a Vibe-Coded Calorie Counter, and the Consultant-to-SaaS Road | demonstrated | | Local model on his GPU |
| AI Sprint: The Hermes Agent — Day 1 (Always-On Agents, the Employee-on-Day-One System, and the Self-Improving Ghostwriter) | demonstrated | | Hooked to Hermes via Codex; 16GB RAM sizing honesty; 'my system will die if I use local models' — teaching demo only |
| Session 22: Automate Admin Tasks — Advanced RAG (Enterprise 'Internal Perplexity' Built Live) | explained | | First-class production provider in the spec; not installed on his Mac that day, so smoke-testing deferred |
| Day 5: MCP as Concierge — Easy/Medium/Hard Modes, Stacking Servers, Smithery Vetting, and RAG vs MCP | explained | | Local GPT-OSS/DeepSeek with n8n cloud |
| Day 9: Webhook vs API for Good — Apify Scraper, cURL Import, Structured Output, Gmail Drafts, then the Native-Node Rewrite | explained | | Sizing, disk usage, Claude Code support |
| Day 10: Principles for Building Agents — Three Levels of AI, the 7-Question Workflow-vs-Agent Test, Goldilocks Scope, Agent Prompting, and Tool Design | explained | | Local model swap for privacy; + LangChain for RAG |
| Day 12: MCP Ultra-Hard Mode — Editing the Config Safely, Building Your Own MCP Server in n8n, Opal, Cross-LLM Memory, and an Early Look at Claude Cowork | explained | | Local URL into n8n; no API key |
| AI Sprint: OpenClaw — Session 2 (Token Anatomy, Claude Code Inside the Server, Skills, and Multi-Agent Workspaces) | explained | | The Slido answer for local models: install into the server via Claude Code, then pull Qwen/Kimi/DeepSeek |
| Basecamp 2: Vibe Coding | mentioned | 2:31:44 | Q&A: local open-source inference; RAM-hungry, trainer found 7B slow on an 18GB MacBook; building against it still needs a paid coding agent |
| Basecamp 4: Introduction to n8n — II | mentioned | 2:54:18 | Third leg of the fully-local stack (n8n + Firecrawl + Ollama) |
| Basecamp 5: MCP | mentioned | 1:16:21 | Cited as a source of file-deletion horror stories when open-source models drive Claude-style tooling |
| Office Hours III: Backing Up Claude Chats with Obsidian, Bridging Chat and Cowork, What Fable Is For, and the Context-Rot Problem | mentioned | | Local models in Claude Code |
| Office Hours IV: PostHog over Plausible, the Launch-Fast Loop, Scraping Ethics, Agent Swarms with Memory, and Why You Never Give Credits Away Free | mentioned | | Local small models, start with Gemma |
| Office Hours XIV: Story Before Landing Page, Scoping an Agentic-Web Pitch, the Cost of Compute, Automating Past Google, and the Idea Bank | mentioned | 1:01 | Pooling idea; Ollama Cloud $20 plan |
| Session 3: Building Elegant Landing Pages | mentioned | 2:24:41 | His 7B-on-18GB experiment was 'terrible' — 24GB+ or stay on the $20 cloud plan |
| Session 4: How to Build Micro Prototypes | mentioned | 0:38:37 | The private-data lane: self-hosted models on a VPS when data can't touch cloud tools |
| Session 9: Introduction to Cursor & its principles | mentioned | 1:17:52 | Possible harness host only with serious local hardware — 'it will guzzle system power' |
| Session 12: Claude Code — Design & Building Agents (App-First: a Game, a GTM Board, and a Live Deploy) | mentioned | 1:52:54 | Learner question; possible but 'not that smart a model' for automations — not recommended |
| Session 29: Build Always-On Assistants — Autonomy Is Designed, Not Switched On (OpenClaw on a VPS, Hermes on the Desktop) | mentioned | | Named as a local-model option connectable to Hermes alongside GLM and DeepSeek. |
| Bonus + Demo Day: Self-Hosting n8n with Docker (Cloud vs Local Decision, the n8n_data Volume, Credentials, ngrok) and How the Hackathon Was Judged | mentioned | | Local LLM connection - deferred '4-hour class' |
| Day 1: APIs, Webhooks, How LLMs Predict, CoStar vs Meta Prompting, Local Models, and a Live RAG Build | mentioned | | Named, not demoed |
| Day 11: Two Alternatives to n8n — Google Workspace Studio (VIP-Email Demo) and Lyzr Agent Studio (Knowledge Bases, Sub-Agents, a Manager, and API Deployment) | mentioned | | Local GPT-OSS as the free route |
| The Five-Level AI Generalist Roadmap (+ Operating System & Monetization) | mentioned | | Local experimentation option (with LM Studio) |
| Learning Topics: Sessions 1–5 — Foundations to AI Director | mentioned | | The local-model privacy path with its hardware/maintenance trade-offs — 'local does not automatically mean compliant' |
| Day 1: What Makes an LLM an Agent - Planning, Tools, Memory, Actions; Five Design Patterns; and Cursor, Antigravity and Claude Code Live | mentioned | | Local models for privacy |
| AI Sprint: The Hermes Agent — Day 2 (Loop Engineering: the Scheduling Loop, the Delegation Loop, and the Skill Auditor) | mentioned | | Local-model lane acknowledged: 'as long as your machine supports it' |
| AI Sprint: The Open Source Playbook — Day 2 (Weights 101, OpenCode, and the Subsidy Math) | mentioned | | Local-model route attachable to OpenCode for capable hardware |
| Weekly AI Updates - October 2026 digest (Oct 7 so far: Gemini 4 'Argon', America.gov AI, Meta Muse, NVIDIA OpenShell, Grok Bot rewards, Copilot teammates, Claude Sonnet 5.5 deep dive; Mark Palmer's GBox showcase; Oct 14 so far: Google Health Coach, the White House frontier-AI accord, OpenAI Dev Day 2026, Anthropic's Oct 8 usage-policy update, Hark, Microsoft MAI voice models, Claude Haiku 5.5) | mentioned | | |
Ollama