$ ./agentic --production▊
Learn to design, build and ship a production-grade agentic software product your client actually wants — using opencode, openclaw-etheroi, hermes-agent-etheroi and openclaude. No throwaway demos, no "hello world" agents. You leave with a real, deployable, monetizable product.
The market is flooded with tutorials that end at a chatbot that answers three questions. This course exists because of a gap I kept hitting in 2026 while shipping real GenAI products at etheroi.com: nobody teaches the production path — how to take four powerful open-source agentic projects and turn them into a complete, client-ready, deployable software product with users, billing, observability and a real business model.
You bring the product idea (or pick one from the course catalog). I bring the architecture, the codebase patterns and the deployment playbook. Together we ship a production app of an agentic software product the client wants — not a proof of concept.
Auth, data, multi-agent orchestration, monitoring, billing — everything a real product needs, nothing a demo skips.
opencode, openclaw-etheroi, hermes-agent-etheroi and openclaude — maintained forks of battle-tested upstream tools, taught end-to-end.
All LLMs run through API keys from providers with daily free tiers — build and ship before spending a cent on tokens.
Four pillars. Each one contributes a critical capability to your production agentic stack. Every project is open source, MIT-licensed at its core, and maintained by the instructor's account — so you fork it, read it, and make it yours.
The open-source AI coding agent — build & plan agents, LSP, MCP, 20+ providers.
github.com/developerjeremylive/opencode ↗The terminal-native coding agent that builds your product's codebase. A client/server runtime (TypeScript, Bun, Hono, SolidJS) where a build agent with full tool access and a read-only plan agent work side by side.
Your own personal AI assistant — 20+ channels, voice, canvas, skills.
github.com/developerjeremylive/openclaw-etheroi ↗The channel-agnostic control plane for your product's assistant. A TypeScript/pnpm monorepo (Node ≥ 22) that runs one assistant across WhatsApp, Telegram, Slack, Discord, iMessage, Signal and more from a single gateway.
The self-improving AI agent — a built-in learning loop.
github.com/developerjeremylive/hermes-agent-etheroi ↗The brain that grows while it works. A Python 3.11–3.13 agent runtime (uv, pydantic, httpx, fastapi) that creates skills from experience, persists memory across sessions, and delegates to subagents.
Open-source coding-agent CLI for cloud & local model providers.
github.com/developerjeremylive/openclaude ↗The model gateway of your product. A TypeScript/Bun CLI that routes one workflow across 20+ backends — OpenAI-compatible, Gemini, GitHub Models, Ollama, Bedrock/Vertex — with a PageRank Repo Map for codebase intelligence.
This isn't theory. Throughout 2026 I've been building different GenAI applications on etheroi.com using exactly these four open-source agentic projects — chat platforms, agent platforms, voice apps, translators, fintech AI reporting and more, hosted on Cloudflare. The course is the distillation of what actually worked in production: the architectures, the failure modes, the deployment patterns.
Eight modules. One production product. Every topic below is taught hands-on — you build as you learn.
What an agent actually is (and isn't), tool calling, the agentic loop, and how multi-agent systems are structured in production. We define the difference between a demo and a product: users, state, observability, failure handling, security and cost control from day one.
Install, configure and master opencode: the build vs plan agent workflow, LSP-powered code intelligence, MCP tool integration, custom subagents, and the client/server architecture. You'll scaffold and iteratively build the full product codebase with the agent — learning to direct it like a senior engineer, not babysit it.
Turn your agent into a product users can reach anywhere. Configure openclaw-etheroi's gateway, connect WhatsApp/Telegram/Slack/Discord, add voice in/out and Canvas rendering, build skills, schedule cron tasks and integrate MCP servers. You'll deploy a multi-channel assistant that persists state and answers on the channels your clients already use.
Implement the learning loop: hermes-agent's skill creation from experience, memory persistence with FTS5 search and LLM summarization, subagent delegation and parallel execution, and choosing among 6 terminal backends. Your product will get measurably better the more your users use it.
Master openclaude's provider profiles and auto-routing: OpenAI-compatible APIs, Gemini, GitHub Models, Ollama, Bedrock/Vertex — plus the Repo Map codebase intelligence, tool-driven workflows and the headless gRPC server. Learn to pick the cheapest capable model for every task and fail over when a provider drops.
Requirements discovery for agentic products, defining the client's desired product, component design, data models, API contracts and how the four projects fit together into one system. We design for scale, security and cost — deciding what runs on Cloudflare (Workers/D1/R2/KV) vs Docker vs dedicated servers.
Authentication and authorization for agent products, API key management, prompt-injection defense, rate limiting, structured logging, tracing, error budgets, retries with backoff, and data persistence (D1/SQLite/Postgres). The boring stuff that separates products from demos.
End-to-end deployment: CI/CD, containerization, Cloudflare Workers + Pages + R2 + D1, environment management, secrets, monitoring dashboards and alerting. Then the business layer: usage tracking, billing integration, a landing page for the product, and a launch checklist.
The capstone: take the product you (or your client) want — an agentic software product — and ship it as a complete production app using all four projects. Weekly 1:1 reviews, architecture feedback, code reviews and a final demo day. You graduate with a deployed product, not a certificate.
You don't need a big ML budget to ship an agentic product. All models in this course run through API keys from providers that include a daily free tier — so you build, test and even launch your first users without paying for tokens.
The provider-agnostic stack behind opencode and openclaude — plug in free-tier models directly and route across providers from one key system.
Gemini's free daily tier gives you generous quotas for text, vision and voice — ideal for production prototypes and lean startups shipping their first agentic product.
One key, hundreds of models — including free models with daily limits. The ultimate routing layer: pick the best model for each task and switch without changing code.
In every module you'll wire real API keys from these providers through openclaude's router and opencode's provider layer. You'll learn cost tracking, rate-limit handling, retries and model selection — the exact skills you need to keep a production product profitable.
If you can read code and you're serious about shipping agentic products, this course was built for you.
Engineers who want to add production agentic products to their skillset — and their portfolio. Bring your own product idea or take one from the catalog.
Launch a real agentic product with near-zero model costs and battle-tested open-source infrastructure. Skip the MVP graveyard — ship something people pay for.
Deliver client projects that aren't demos. Learn the architecture and deployment patterns to quote, build and hand over production agentic products profitably.
8 modules · 1:1 · production capstone
for teams shipping agentic products
Agencies and product teams shipping agentic software for clients get a tailored track with a shared architecture review and delivery support.
"Not an expense — infrastructure for your agentic product."
Book a 20-minute call. Tell me the product you want to build — I'll tell you the architecture, the stack and the path to production. No pitch, just engineering.
I'm a computer science engineer with 8+ years shipping software across health, finance, telecom and logistics — and throughout 2026 I've been building different GenAI applications at etheroi.com using these four open-source agentic projects. Multi-agent systems, MCP-powered tooling, voice apps and edge-hosted AI on Cloudflare — all in production, all real.
No fluff, no toy projects. I take the same client-first, production-first rigor I use to ship software and apply it to your product — so you graduate with something deployed, not something demonstrated.