Coming soon
Video + text edition
67 lessons · ~18.3 hours
Build Your Own AI Coding Agent & Personal Assistant
Build a private AI coding agent and personal assistant on local models — safe tools, durable memory, voice, RAG, sandboxing, and observability — with no required cloud API dependency.
11sections
67lessons
~18.3hreading
1capstone
🎬 Watch: course introduction
What this course builds, who it's for, and how to use it — about 2 minutes.
Edition honesty: this course pairs written lessons with short video lectures — the text is the complete course, the videos teach the key ideas out loud. Work at your own pace.
Syllabus
The full curriculum, section by section
Click a section to see every lesson. Lesson links open in the learning platform once you own the course (or sign in for a preview of what's inside).
Section 1 — Sovereignty is an architectural property
- What "sovereign" means in practice 14 min read
- The reference architecture 18 min read
- Choose an inference lane 16 min read
- Hardware tiers without gatekeeping 12 min read
- Build the local skeleton 15 min read
Section 2 — Agents are loops, not magic
- Tokens, context, and chat templates 17 min read
- Tool calling is structured prediction 18 min read
- ReAct as a bounded state machine 20 min read
- Prompt engineering for agents, not chat 17 min read
- Run the first local agent turn 18 min read
- Framework or from scratch? 10 min read
Section 3 — A coding agent begins with narrow tools
- Define the workspace contract 15 min read
- Read and search tools 18 min read
- Safe writes and patch application 20 min read
- Shell execution inside Docker 25 min read
- Web fetch without creating a browser robot 15 min read
- Dependency changes require evidence 12 min read
- Assemble the coding-agent MVP 15 min read
Section 4 — Reliability comes from planning, evidence, and evaluation
- Plan only when the task earns it 17 min read
- Task decomposition and dependency graphs 18 min read
- Self-correction without infinite loops 18 min read
- Build a mini coding benchmark 24 min read
- Judge artifacts, not eloquence 15 min read
- Regression runs and comparison discipline 13 min read
Section 5 — Repository understanding is retrieval plus compression
- Build a repository map 17 min read
- Chunk code by meaning 20 min read
- Local embeddings with sentence-transformers 18 min read
- Qdrant now, pgvector when consolidation wins 20 min read
- Hybrid retrieval and reranking 18 min read
- Context compaction without amnesia 17 min read
Section 6 — Personal memory must be explicit and reversible
- Separate state, memory, and knowledge 16 min read
- PostgreSQL as the source of truth 18 min read
- Memory write policy 18 min read
- Private document ingestion 22 min read
- Retrieval-augmented generation with citations 20 min read
- Forgetting, retention, and backups 12 min read
- End-to-end private Q&A lab 9 min read
Section 7 — Voice and proactive behavior need queues and consent
- Voice is a latency budget 17 min read
- Local speech-to-text with faster-whisper 18 min read
- Local text-to-speech with Piper or Coqui 15 min read
- Jobs, schedules, and idempotency 20 min read
- Local connectors for email and calendar 20 min read
- Proactivity without surprise 10 min read
Section 8 — A simple application boundary beats framework sprawl
- FastAPI as the control plane 18 min read
- HTMX for a legible interface 14 min read
- MinIO for artifacts, not database blobs 12 min read
- Compose the sovereign application 20 min read
- Laptop, home server, bare metal 16 min read
- Backup, restore, and exit drill 5 min read
Section 9 — Sandboxing and human approval are product features
- Threat-model the agent 18 min read
- Harden the Docker runner 22 min read
- Firecracker for stronger isolation 18 min read
- Approval gates that communicate risk 18 min read
- Prompt injection and data boundaries 14 min read
- Security regression suite 10 min read
Section 10 — Production quality is measured, not guessed
- Build the telemetry model 18 min read
- Trace one slow run 14 min read
- Quantization and model routing 20 min read
- Caching and speculative decoding 17 min read
- Capacity, failure, and graceful degradation 14 min read
- Release gate and regression dashboard 12 min read
- Optional lesson — Cloud fallback without cloud dependence 10 min read
Section 11 — Capstones: Prove End-to-End Ownership
- Choose a Capstone and Freeze Scope 12 min read
- Architecture Review 12 min read
- Milestone and Demo Discipline 12 min read
- Portfolio Packaging 14 min read
- Capstone: Ship your own coding agent + personal assistant 30 min read
Outcomes
What you'll be able to do
✓ Build a private AI coding agent that runs on local models — no required cloud API
✓ Design safe, narrow tools with explicit workspace contracts and human-approval boundaries
✓ Give agents durable, explicit memory plus retrieval over your own repositories
✓ Evaluate agent work with benchmarks and evidence instead of vibes
✓ Add voice and proactive behavior safely, with queues and consent
✓ Measure production quality with telemetry — then ship a capstone you can demo end-to-end
Prerequisites
Comfortable with a terminal and basic scripting. The course meets you where you are on hardware — no expensive GPU is required to follow along.
Instructor
Instructor bio coming soon — written by the course author.
What's included
- 67 video + text lessons (~18.3 hours of reading), each ending with something to do
- Exercises built into every section — applied, not theoretical
- A capstone project with self-review rubrics — prove you can do the thing, and check your own work
- Lifetime access — buy once, keep it; founding price locked
Start building.
Video + text edition. Opens soon — founding price locked for early buyers.
Resources
Free downloads that go with this course.
The cheat sheet is a dense reference distilled from the course lessons. The starter project is a skeleton with TODOs that map to course sections — a starting point, not a finished product.