All courses
Pick a course and start with a free lesson.
Build a minimal read-only MCP server
Build a server with a narrow lookup tool, static resource, current SDK factory and model-free protocol tests.
OpenAI tools and agents without paid execution
Build an offline harness for current Responses function calling, structured output, approvals and evals with explicit API-budget boundaries.
LangGraph: a stateful local agent with review
Build a reproducible graph with a narrow read tool, checkpoints, review interrupts and tests without model calls.
OpenClaw in a read-only lab
Configure an isolated Linux VM, local Gateway and Ollama, restrict the agent to reading and verify a custom skill without external actions.
Agent security and prompt injection
Build a testable boundary between what an agent reads and what it is authorized to do. Work from trust mapping through hostile fixtures to incident recovery.
Evaluate AI agents in practice
Prepare reproducible fixtures, artifact and action checks, a rubric and an honest version-comparison report. Complete the practice locally using stubs without paid APIs.
Run AI workflows reliably
Design states, safe logs, traces, bounded retries, budgets, actionable alerts and verifiable recovery for a private workflow.
Free local transcription with Whisper and FFmpeg
Prepare your recording, build whisper.cpp, select model and language, review words against audio and organize private batching. Free software does not provide free compute resources.
Automatic subtitles from local ASR
Move from owned video to reviewed SRT/VTT and export. Learn segmentation, synchronization, necessary sound cues and text-track versus burned delivery.
Free video editing with Kdenlive
Build your own short lesson: organize media/proxies, cut precisely, balance audio, configure or import subtitles and verify the render.
Automate simple video editing with FFmpeg
Inspect owned files, trim, normalize, join, adjust audio and caption. Build a local CLI project with QC while preserving originals.
Local text-to-speech with Piper
Prepare a private, bounded Piper workflow for synthetic workshop announcements, from license evidence to reviewed WAV files. Commands target an isolated Linux training machine; no Python is run on a Mac in this course authoring. Requires basic terminal skills, supported Python, disk/RAM and optional FFmpeg tools. CPU execution needs no paid API, but hardware, electricity and any rented machine still cost money. No cloning, impersonation or publication is part of the lab.
Local image workflows with ComfyUI
Build and review a bounded seven-node local image workflow for a fictional unbranded ceramic bowl. Requires basic file management, a supported Windows practice machine, adequate disk/RAM and a compatible GPU or slower CPU path. The course uses official ComfyUI Portable and an inspected SD1.5 checkpoint as a modest teaching baseline, not a quality recommendation. No model is executed while authoring, no paid service is required, and hardware, electricity or rented GPUs still have costs.
Build and hand off a small private AI portal
Build a connected Node prototype with UI, HTTP, auth, object access, grounded retrieval, adapter, limits, logs and tests. Initial extractive-stub mode uses no real model or paid API.