Guides, research notes and what we learn building models.
The complete architecture for AI agencies serving law firms — from n8n orchestration to fine-tuned model inference to client-facing delivery. Component selection, deployment topology, and scaling considerations.
Read the article
You built your app with Cursor and plugged in OpenAI. Now you're locked in. Here's how to deploy AI features that you own — from prototype to production without vendor dependency.
Read the article
Enterprise clients increasingly require that their data never leaves their infrastructure. Here's how AI agencies can meet data sovereignty requirements with locally deployed fine-tuned models.
Read the article
Identical QLoRA runs on one RTX 4090 compare Llama 8B and Qwen 2.5 7B on classification, extraction and generation, plus VRAM and speed.
Read the article
Compare AI API pricing in 2026 with self-hosted open models, using real numbers at real scale, and find the cheapest way to add AI to your indie app.
Read the article
Most fine-tuned models never reach production. Here's why the gap between training in a notebook and deploying in production exists — and how to close it systematically.
Read the articleFine-tuning and retrieval-augmented generation solve different problems. This guide explains when to use each approach, the trade-offs involved, and how to combine them for the best results.
Read the articleServe every law firm client from one base model: how to train a LoRA adapter per firm, hot-swap it in Ollama or vLLM, and keep client data isolated.
Read the articleA detailed ROI analysis comparing self-hosted fine-tuned models against OpenAI API pricing for agencies — with worked examples for 3-client and 10-client scenarios and break-even calculations.
Read the articleAn honest comparison of Ertas Studio against DIY fine-tuning tools like Unsloth and Axolotl — focused on what matters for agencies: time-to-deliver, client handoff, and iteration speed.
Read the articleA composite case study showing how an automation agency used n8n, fine-tuned local models, and on-premise deployment to deliver HIPAA-compliant AI to a regional hospital network.
Read the articleHow self-hosted AI models change agency pricing strategy. Flat-rate, per-seat, and hybrid pricing models with worked margin examples at each GPU tier.
Read the articleA tactical guide for n8n agencies: collect client interaction data via workflows, clean and format it, fine-tune a model in Ertas Studio, deploy locally, and connect back to n8n for inference.
Read the articleEverything you need to know about running large language models on your own hardware — from hardware requirements and model formats to tools like Ollama, LM Studio, and llama.cpp.
Read the articleAn actionable compliance checklist for deploying fine-tuned AI models on-premise at law firms — covering data handling, access controls, audit logging, model versioning, and bar association requirements.
Read the articleA sales guide for AI agencies: how to frame on-premise AI for hospital CTOs, handle objections, structure your proposal, and navigate the healthcare procurement process.
Read the articleA hands-on tutorial: sign up for Ertas Studio, upload a sample legal dataset, configure and run fine-tuning, evaluate the output, and export your model — all in 30 minutes.
Read the articleSystem prompts, RAG context, retries and conversation history multiply your cloud AI API bill. Here is the real math, and what local inference costs instead.
Read the articleCloud LLM APIs create data processing risks under GDPR. Learn how to build AI features that comply with European privacy regulations using fine-tuned local models.
Read the articleUpload a JSONL dataset, pick a base model, train on managed cloud GPUs and export GGUF, all from a visual canvas. No Python, no terminal, no GPU setup.
Read the articleA hands-on guide to fine-tuning Meta's Llama 3 models — covering model selection, dataset preparation, LoRA configuration, training tips, and deployment as GGUF for local inference.
Read the articleErtas Studio is a canvas-driven interface for fine-tuning multiple AI models simultaneously. Upload data, configure training, and compare results — no CLI required.
Read the articleMost ML tools are built for the command line. We think fine-tuning deserves a visual workspace. Here's why we designed Ertas Studio as a canvas — and what that changes about the fine-tuning workflow.
Read the articleHow Ertas balances the convenience of cloud fine-tuning with the privacy of local deployment — giving you control over your data where it matters most.
Read the article