The reference guide · 25 chapters · asset pack included · free updates for life
Your code copilot,
100% local
Copilot Pro is $120/year. Cursor Pro is $240/year. This is $27 — once.
The Local Copilot Kit is THE reference guide for building your own AI copilot on your own machine — from the first install to an agent that codes inside your editor. Your code and your data never leave your disk.
Get the kit — $27$27 · one-time payment · free updates for life
✓ This is for you if…
you want your own AI copilot on your own machine — professional developer, freelancer, student, or simply comfortable with a computer. Privacy, cost, or cloud dependence bother you, and you've got a GPU with a few GB of VRAM or an Apple Silicon Mac.
✗ This isn't for you if…
you want an agent that autonomously codes entire multi-repo projects on its own, or every millisecond of latency costs you real money. The cloud keeps the edge there — and the guide tells you so.
Not sure your machine is enough? Check in 30 seconds with the free configurator →
Why you can trust this
This guide is written and maintained by BestLLMfor — an independent, continuous benchmark pipeline tracking 185 locally-runnable models and 43 guides, run in the open on a published methodology. No VC, no content farm behind it. Every config in this kit is tested end to end the same way, on real consumer hardware — not templated from a vendor lab, not a random compilation of tutorials found online. What you read here is what we actually run. Who's behind BestLLMfor →
The problem
As of April 2026, GitHub Copilot trains on your code by default across its Free, Pro, and Pro+ tiers. Cursor keeps getting more expensive and bills you by usage. And on every prompt, your proprietary code travels to a third party.
You already know a local LLM fixes all of that. But between "which model fits my card", quantization, Ollama crashing, and a dozen contradictory tutorials, you still don't have a setup that actually works. The kit is THE reference method: the complete path, tested end to end, configs included — nothing left to hunt down elsewhere.
Local copilot vs GitHub Copilot vs Cursor
Same work in VS Code — without a subscription and without sending your code anywhere. The honest comparison:
| Local copilot | GitHub Copilot | Cursor | |
|---|---|---|---|
| Your code stays on your machine | Yes | No | No |
| Trained on your code | Never | Yes, by default | Depends on settings |
| Cost | $0/month | ~$10/month | ~$20/month |
| Works offline | Yes | No | No |
| Chat + agent on your repo | Yes (Cline) | Yes | Yes |
| Autocomplete | Yes (with a GPU) | Yes | Yes |
| Heavy multi-repo reasoning | Limited | Strong | Strong |
The guide shows you how to cover ~90% of your Copilot/Cursor usage locally — and tells you plainly where the cloud still wins.
What you get
- Your members area, for life — all 25 chapters, always the current edition, every command copy-pasteable in one click (exotic FIM tokens survive copy-paste, which a PDF alone can't guarantee).
- The full guide as a PDF (175 pages) — the same content, yours forever, offline.
- The 21-file asset pack — Modelfiles, setup scripts, a prompt library, and an IT decision memo, ready to paste or hand to your team.
- Ready-to-paste editor configs — Cline (VS Code), Aider (terminal), Tabby (autocomplete), tiered by VRAM.
- 8 advanced modules — benchmarking your own machine, multi-project workflows, local RAG on your docs, a local pre-commit hook, hardening your setup, the prompt library — plus two new ones: your private ChatGPT (Open WebUI / AnythingLLM) and local AI at work (privacy, SOC2, HIPAA, a ready-to-send IT memo).
- The "when it breaks" chapter — OOM, GPU not used, slow output: the diagnosis and the fix for each failure mode.
- The cheat-sheet — your VRAM, the right model, the exact command, on one page.
- Free updates — every monthly edition (new models, retuned configs) at no extra cost.
Stack: Ollama + Cline + Aider + Tabby. Models: Qwen2.5-Coder, Qwen3-Coder, Devstral, DeepSeek-Coder. 100% local, $0/month.
The 25 chapters of the kit
175 pages, from the 30-minute quick win to advanced modules you won't find anywhere else on the site. Nothing to skip: every chapter links to the next.
- Why go local in 2026
- Your first local copilot in 30 minutes
- What local gives you — and where it loses
- Installing Ollama the right way
- The right model for your VRAM
- Is your machine enough, and when to upgrade
- Cline + Ollama inside VS Code
- Aider in the terminal, your git-aware pair programmer
- FIM autocomplete: inline Copilot, running locally
@codebase: RAG on your own repo- Modelfiles by model × role
- The settings that actually matter
- When it breaks: diagnose and fix
- The real cost model
- A concrete threat model: what actually leaves your machine
- Reading a dated edition over time
- Update log + Appendix A: going further
- Benchmark YOUR machine in 10 minutes (real VRAM + real tok/s, not the spec sheet)
- Multi-project workflow: one Ollama, many repos
- Local RAG on your docs (not your code — your PDFs, your specs)
- A 100% local pre-commit hook: a reviewer before every push
- Hardening your setup: from “nothing leaves” proven to a standing guarantee
- A disciplined prompt library, ready to copy-paste
- Your private ChatGPT: a local assistant for everything, not just code (Open WebUI / AnythingLLM)
- Local AI at work: making the case to your company (privacy, SOC2, HIPAA, a ready-to-send IT memo)
Every monthly edition enriches this list (dated update log, ch.16–17) — you never repurchase to get the next version.
What you'll build in 30 minutes
No expertise required. The guide walks you from the first command to a copilot answering inside your editor:
1 · Install Ollama
One command. The guide tells you which model to pull for your card (VRAM table above).
2 · Pull the right model
ollama run qwen2.5-coder:7b — and confirm your GPU is actually doing the work.
3 · Wire up Cline
Chat + agent inside VS Code, reading and refactoring your repo — fully local.
4 · Turn on FIM autocomplete
The inline Copilot experience, running on your machine, with the correct fill-in-the-middle tokens.
5 · Local RAG on your repo (optional)
@codebase-style retrieval over your own code, still 100% offline.
- Your members area, for life: all 25 chapters always current, commands copyable in one click (full list above)
- The PDF guide + the 21-file asset pack (Modelfiles, scripts, prompt library, IT memo) to download
- The cheat-sheet + the full troubleshooting chapter
- Every monthly edition, for life — they show up inside your members area automatically
- Instant access after payment, access key emailed to you
Secure payment via Stripe. You're out the door with the PDF and the asset pack in under a minute. Receipt emailed instantly · invoice available on request.
The honest truth
Local replaces ~90% of your Copilot/Cursor usage — not 100%. We'll tell you plainly where it wins, and where the cloud keeps the edge. A reference guide owes you that honest map:
By the end, you'll know how to
- Pick the right model for YOUR machine (and know which ones to avoid) in 10 seconds.
- Run a chat + agent inside VS Code (Cline) that reads your repo, refactors, and explains — locally.
- Get autocomplete that actually works (right FIM tokens, right model), not one that hallucinates.
- Diagnose and fix it yourself when it breaks (OOM, GPU, slow output).
- Code offline, without a subscription, without a single line of your code leaving your machine.
Living edition
"Free updates for life" isn't a slogan — the kit tracks the ecosystem (new models, deprecated tools, retuned configs) for you, and every update ships inside your members area at no extra cost.
v2026.09 — the launch edition. New editions ship as the stack moves (new models, tool changes, retuned Modelfiles) — you'll see them the moment you log back into your members area. Nothing to re-download by hand, nothing to repurchase.
Frequently asked questions
How do I replace GitHub Copilot with a local AI?
Install Ollama, pull a coding model (e.g. Qwen2.5-Coder), then connect Cline inside VS Code (or Aider in the terminal) to Ollama. The kit gives you the exact path plus ready-to-paste configs -- budget 30 minutes.
Which local AI should I pick for coding?
Depends on your VRAM: Qwen2.5-Coder 7B (8GB), 14B (12GB), Devstral 24B (16GB), 32B (24GB). The kit's cheat-sheet gives you the right model and the exact command for your machine.
Can you really use Ollama inside VS Code?
Yes: the Cline extension (chat + agent) for reasoning and refactors, and Tabby or Twinny for autocomplete. The exact configs -- and the correct FIM tokens -- are included in the pack.
Does it work offline?
Yes. Once the model is downloaded, everything runs without a connection: chat, refactors, generation, autocomplete -- 100% local.
Do I need to be a developer to follow the kit?
No. The kit starts from zero: installing Ollama, picking the model that fits your machine, wiring up your editor -- every step is guided, with copy-paste commands. If you can install software and open a terminal, you can follow it. The advanced modules are waiting for when you're ready.
What if my machine isn't enough?
The “is your machine enough” chapter tells you in 5 minutes (tiers starting at 8GB of VRAM, Apple Silicon included). If it isn't enough, you find out before buying, not after.
Is this suitable for business use and compliance-conscious teams?
That's exactly the point: your code never leaves your machine, and nothing is sent to a third party or used to train a model. The kit also sorts which licenses are safe to use at work (Apache 2.0 / MIT) from the ones to avoid, and includes a ready-to-send IT decision memo for privacy, SOC2, and HIPAA-conscious teams.
Is it free?
The LLM itself runs locally: $0/month, no subscription, no usage cap. The kit (guide + asset pack + lifetime members area) is a one-time $27 purchase.
What if it doesn't work on my machine, after I buy?
30-day money-back guarantee -- one email, no questions asked.
How is it delivered?
Instant access after payment: your lifetime members area (always the current edition, personal link emailed to you) plus the PDF and the asset-pack zip to download. Access and downloads are permanent -- and if you lose the email, you can request your link again in seconds.
Can I get a receipt or invoice for expensing?
Yes. Payment runs through Stripe: you get an emailed receipt immediately, and a formal invoice on request -- just reply to the confirmation email.
What if BestLLMfor shuts down one day?
The PDF and the asset-pack zip are yours, downloaded to your machine -- they don't depend on the site staying up. The members area is a reading convenience, not a condition of access.
How is this different from the free guides on the site?
The site's guides explain each piece separately. The kit is THE reference method: all 25 chapters in the right order, tested and maintained configs (Cline, Aider, Tabby, Modelfiles), full troubleshooting, and advanced modules you won't find anywhere else on the site -- kept current every month. It's the complete system for building your local copilot, not a collection of articles.
Build your local copilot.
THE reference guide, from your first download to a copilot answering inside your editor — private, no subscription, on your machine. Today, not "someday."
Get the kit — $27