k33g_org's Blog

Relearning Kotlin without installing anything: let an agent set up jbang, Kotlin and ten lessons inside an sbx sandbox, then turn that setup into a reusable kit.

Build a WASI Preview 2 component in Go with TinyGo: the WIT contract, the generated bindings, the wasip2 build, and calling it from Node.js through jco.

Spread the context out with Docker Agent: a tool-less coordinator delegates to three Rust specialists, so every agent carries less and keeps room to think.

Assemble a local Rust learning companion with Docker Agent: a cookbook RAG, a rust-analyzer LSP toolset, and the context caps that keep a small model alive.

Make a local Docker Agent easier to drive: move the system prompt into a Markdown file, and add slash commands with argument interpolation.

Give a small local model detailed procedures on demand with Docker Agent skills: an on-disk SKILL.md, an inline one, and one forked into an isolated sub-agent.

Plug external MCP tool servers into a local Docker Agent: a local stdio server, a remote streamable HTTP one, and a tools filter to keep a small model focused.

Ground a tiny local model's answers in your own documents: build a Rust cookbook RAG with Docker Agent, chunked embeddings and a local SQLite index.

Add reasoning, persistent memory and fine-grained command permissions to a Docker Agent mini code agent, with deny rules that survive yolo mode.

Add typed filesystem operations and a todo task list to a local Docker Agent, with allow_list scoping to keep the small model safely inside its working directory.

Build a mini local code agent with Docker Agent and Docker Model Runner: a single shell tool on the small Mellum2 model, the agent loop, and running it safely in an sbx sandbox.

Install and use dmr, the standalone Docker Model Runner: one binary with the inference daemon and the full model CLI, no Docker Desktop needed.

Wire NVIDIA NeMo Guardrails into a local Docker Agent to catch jailbreaks, refuse out-of-scope requests, and mask PII on-device.

Stop PII from leaking to your LLMs with Microsoft Presidio, doing 100% on-device detection and redaction inside Docker Agent.

Swap in Meta's Llama Guard 2 8B classifier to moderate a local agent's prompts through Docker Agent hooks, all on-device.

Add a 100% on-device content-safety guardrail to Docker Agent by plugging Google's ShieldGemma 2B classifier into a hook.

Final tweaks to pk before learning Rust: context packing, new models, and finishing the local coding-helper setup.

Harden the pk coding helper: secure sandbox, a web IDE, RAG-based documentation search, and language server support.

Build pk, a TUI for crafting coding helpers backed by very small local language models (<=10b parameters) for offline Rust work.

Run Mistral's Vibe coding agent CLI securely inside an sbx sandbox, with credential handling and network isolation.