Starter Kit · One-Time Purchase
Local AI Automation Scripts
Ready-to-run Python scripts for local AI
Worth $29 — you pay $9 today.
Buy now — $9Instant delivery — download from your in-app library right after checkout. Linked to your account; sign in any time to re-download. All sales final.
Overview
Local AI Automation Scripts is a self-contained Python toolkit that turns documents and code into useful output — summaries, spreadsheets, tags, reports, drafted emails, code reviews — using a local Ollama model instead of any cloud API. Eleven ready-to-run scripts cover document summarization, field-and-entity extraction to CSV/JSON, classification, report synthesis, email drafting, code review, test generation, plus a folder watcher and scheduler for hands-off automation. Everything runs offline on your own machine: no API keys, no accounts, no per-token bill, and no data ever leaving the network.
The real value isn't the code — it's the service the code lets you sell. There's a whole category of buyers who legally or contractually can't paste their files into ChatGPT: law firms, accountants, clinics, HR teams, insurance and claims offices, property managers, government subcontractors. They have piles of documents, no in-house engineer, and a hard "no cloud AI" rule. This kit removes the build barrier so you can install a private document processor on hardware they own and charge a real setup fee for it.
Be clear-eyed: this is not passive income. The scripts handle the engineering; you still scope the job, configure the fields, validate the output, train the user, and provide support. That work is exactly what clients pay for — and the included MONETIZE.md walks you through the pitch, pricing, sample scope-of-work, and how to find clients without cold outreach.
What's included
- 11 working Python scripts, each with a --help flag, organized into Documents, Extraction, Classification, Automation, Writing, and Code categories
- summarize_document.py — summarizes PDF/DOCX/TXT/MD at 3 levels (tldr/standard/detailed), map-reduce for long docs so nothing is truncated
- extract_fields.py — pulls a fixed set of fields (e.g. invoice number, vendor, total, due date) from one file or a whole folder into CSV + JSON
- extract_entities.py — extracts people, orgs, dates, amounts, and locations as reliably-parseable JSON (JSON mode)
- classify_documents.py — sorts/tags documents into your own labels, optional --move into subfolders, writes a CSV manifest
- generate_report.py — synthesizes a folder of sources into a structured Markdown report (exec summary, findings, risks, actions)
- batch_summarize.py — summarizes every document in a folder into one combined report
- watch_folder.py — watches a 'drop-in inbox' folder and auto-runs an action on every new file
- scheduler.py — runs jobs on a schedule from a small JSON config (e.g. daily 8am digest), with a --run-now option
- email_drafter.py — drafts emails from bullet points in any tone
- review_file.py + generate_tests.py — code review for bugs/security/perf and unit-test generation (pytest/Jest/Vitest)
- MONETIZE.md — the full service playbook: one-line pitch, buyer table, pricing ranges, copy-paste Scope of Work, delivery checklist, and warm/inbound client-finding channels
- README, requirements.txt, .env.example, jobs.example.json, and a hardware-to-model sizing guide (qwen3:4b → qwen3:8b → llama3.3:70b), all model names verified current as of June 2026
Who it's for
- Freelancers and independent consultants who want a concrete, sellable AI service rather than a vague 'AI for business' pitch
- Developers and tech-savvy generalists who can run a Python script and want to package that into paid setup work for local firms
- IT consultants and MSPs whose small-business clients keep asking 'can we use AI safely with our data?'
- Bookkeepers, virtual assistants, and ops people who already process documents for clients and want to do it faster and charge more
- Anyone who needs private, offline document processing for their own regulated or confidential workflow
Use cases
- Turn a folder of vendor invoices or receipts into a clean spreadsheet (CSV/JSON) for a bookkeeper or accounting client
- Summarize and extract clause/entity data from contracts for a law firm — privately, on their own machine
- Classify and auto-sort an inbox of mixed documents (invoices, contracts, resumes, other) into folders with a manifest
- Stand up a 'drop folder' so every new file a clinic or office adds gets summarized or triaged automatically
- Generate a scheduled daily digest report from a research or intake folder at 8am with no human in the loop
- Screen and summarize a stack of resumes for an HR team or recruiter without uploading candidate PII anywhere
- Run private code review and test generation for a dev shop that can't send proprietary code to a cloud LLM
Sell privacy-safe AI document processing to firms that can't touch cloud AI
The service
Install and configure a fully local AI document processor on hardware the client owns — so their contracts, invoices, patient notes, resumes, or claims get summarized, classified, and extracted into spreadsheets without a single byte leaving their network. You charge a one-time setup fee for scoping, configuring the fields, validating output, and training their team, plus an optional monthly support retainer. The whole pitch is one fact: the data never leaves the building.
What to charge
Pilot / proof-of-value: $400–$900 one-time. Standard setup (2–3 workflows + drop-folder + runbook): $1,200–$3,500. Setup + scheduler/automation: $2,500–$6,000. Monthly support retainer: $150–$600/mo. Per-batch processing you run yourself: $0.50–$3 per document or $300–$1,500 per batch. The setup fee is the engine; the retainer compounds. A part-time freelancer landing ~1 client/month realistically books roughly $15k–$25k of first-year side income — scaling with your time, not passively.
How to find clients
- Start with your existing network — every accountant, lawyer, clinic owner, or recruiter you already know is a prospect. One message: 'I built a way to run AI on documents 100% privately — want me to show you on one of your files?'
- Partner with IT consultants and MSPs (offer a 10–20% referral cut). They already manage these firms' computers and get asked 'can we use AI safely?' — one MSP can feed you for a year. This is the single highest-leverage channel.
- Show up at local professional groups — chambers of commerce, bar/CPA association events, BNI breakfasts — and offer a free 20-minute 'Private AI for [profession]' lunch-and-learn. These rooms are full of regulated-data firms.
- Publish content that ranks for the fear: 'Is it safe to put client documents into ChatGPT?', 'HIPAA-safe document AI', 'on-prem AI for law firms.' People searching that are pre-qualified buyers.
- After your first happy client, ask for one intro to a peer firm — regulated professionals trust peer referrals far more than ads. Be helpful (not pitchy) in industry-specific LinkedIn/Slack/Discord groups and become 'the private-AI person.'
The delivery steps
- Confirm the client’s machine can run a useful model (16GB+ RAM is comfortable for qwen3:8b; 8GB → qwen3:4b/gemma3:4b; a server → llama3.3:70b), and get 3–5 representative sample documents under NDA if needed.
- Agree the exact fields, categories, and report format in writing — use the copy-paste Scope of Work in MONETIZE.md (objective, scope, out-of-scope, privacy statement, fee, accuracy note).
- Install Ollama, pull the right-sized model, copy the kit, pip install -r requirements.txt, and set OLLAMA_MODEL in .env to match their hardware.
- Run a real document through extract_fields.py / classify_documents.py and tune fields and labels until the output is clean; demo it live on one of THEIR documents — the CSV demo lands every time.
- Set up the drop-folder (watch_folder.py) and/or scheduler (scheduler.py) if in scope and make it auto-start (Task Scheduler on Windows, launchd/systemd on Mac/Linux).
- Process the agreed sample batch and review accuracy WITH the client, write the runbook, train and record the call, confirm in writing that no data leaves their network, then invoice (50/50) and ask for one peer referral.
How to market it
- Lead with the fear you remove, not the tech: 'You've been told not to put client documents into ChatGPT — that's the right call. I set up an AI that runs entirely on a machine you own.' Say 'nothing leaves your network' twice.
- Demo, don’t describe. Offer a free live demo running one of the prospect’s own (or representative) documents through extract_fields.py to a CSV — seeing their invoice become a spreadsheet in 30 seconds closes deals.
- Niche down to one profession at a time (e.g. 'Private AI for accountants') so your message, sample docs, and case study all reinforce each other and referrals stay inside the niche.
- Publish fear-keyword content on a site or LinkedIn ('Is ChatGPT safe for client files?', 'HIPAA-safe document AI') to attract pre-qualified inbound rather than cold outreach.
- Build an MSP/IT-consultant referral network with a clear cut — they already have trust with these firms and field the 'can we use AI safely?' question for you.
- Turn your first delighted client into a one-page case study (privacy + hours saved) and trade it for a single peer introduction; peer trust converts far better than ads in regulated industries.
- Resell the kit’s deliverables as productized tiers — Pilot, Standard, Setup+Automation, plus a support retainer — so prospects pick a package instead of negotiating an hourly rate.
Frequently asked questions
Do I need to be a developer to use or sell this?
You need to be comfortable running Python scripts from a terminal and installing Ollama. You do NOT need to write the pipeline — that’s already built. The paid work is scoping, configuring fields, validating output, and training the client, which is exactly what they pay you for. MONETIZE.md spells out every step.
Is this really private, or does it call a cloud API?
Genuinely private. Every script runs against a local Ollama server on your (or the client's) own machine. There are no API keys, no accounts, and no data sent to OpenAI, Anthropic, or anyone else. That 'data never leaves the building' fact is the entire reason regulated buyers will pay you instead of using ChatGPT.
What hardware does it need?
Python 3.10+ and Ollama with one chat model pulled. 8GB RAM runs qwen3:4b or gemma3:4b; 16GB comfortably runs the qwen3:8b default; a workstation/server can run llama3.3:70b. The README includes a hardware-to-model sizing table so you match the model to the client’s box.
What file types and outputs does it handle?
It reads PDF, DOCX, TXT, and MD (text layer of PDFs — scanned/image-only PDFs need OCR first, which isn’t included). Outputs include CSV and JSON spreadsheets from extraction, Markdown reports, classification manifests, and plain-text summaries/emails, all written to an output folder and printed to stdout.
How much can I realistically charge and earn?
Realistic 2026 freelance ranges: $400–$900 for a pilot, $1,200–$3,500 for a standard setup, up to $6,000 with automation, and $150–$600/month for support. A part-time freelancer landing about one client a month books roughly $15k–$25k of first-year side income. This is active service income, not passive — and the setup fee, not the retainer, is the engine.
Are the models and scripts current?
Yes. All model names (qwen3:8b, qwen3-coder:30b, llama3.3:70b, nomic-embed-text, etc.) and dependency versions were verified against the Ollama library as of June 2026, and the kit is built to let you swap in newer models via a single .env setting as Ollama moves.
After you buy
Purchases are linked to your account — sign in and head to your product library to download anytime. Bought without an account? Check your email for the download link and a one-click way to set a password.
← Back to all kits, tools & codebases