Back to projects
2026 · AI Toolchain · Active

AI Toolchain — OpenCode + wiki-brain

AI-assisted development ecosystem: multi-agent orchestrator, persistent knowledge wiki, and infrastructure automation

Python MCP SDK OpenRouter SQLite Bash Podman YAML systemd

Problem

Complex software development demands time and context that chat histories cannot retain. Without a structured system, knowledge is lost between sessions.

Solution

An AI-augmented development ecosystem where I design the architecture and OpenCode orchestrates specialized agents to implement it. Wiki-brain maintains persistent knowledge across sessions with full traceability. Bash/Podman scripts automate deployment.

Key takeaways

  • Multi-agent orchestrator with 7 specialists (oracle, librarian, explorer, designer, fixer, observer, council), each running the optimal model for its task
  • wiki-brain MCP: a server that exposes an Obsidian vault as search, synthesis, and linting tools for any AI assistant
  • LLM Wiki Pattern: knowledge is never lost in chat histories — it is compiled into persistent wiki pages with traceability
  • Auto-update with locking, inter-container health checks, and automatic rollback for the entire container fleet
  • Full infrastructure defined in YAML (docker-compose) and idempotent Bash scripts

Philosophy

This project is my AI working methodology. It is not a product or a service — it is the tool I use to accelerate my development when it makes sense. The AI does not define who I am; it is an accelerator under my direction.

My role is designer/orchestrator: I design the solution, make the technical decisions, and validate every outcome. The AI executes the implementation when the task allows it.

Note: not all of my code is written with AI. My embedded systems projects (C, C++, firmware) are built the traditional way. This ecosystem is complementary, not a replacement.

Components

OpenCode + oh-my-opencode-slim

An orchestrator that coordinates 7 specialized agents, each with a different AI model optimized for its function:

AgentModelRole
oracledeepseek-v4-proArchitectural decisions, complex debugging, code review
librarianminimax-m2.7Documentation and API research
explorerminimax-m2.7Fast source code search
designerkimi-k2.6UI/UX, visual design
fixerdeepseek-v4-flash-freeFast implementation of bounded tasks
observerkimi-k2.6Monitoring and analysis
councildeepseek-v4-proStrategic advisory

wiki-brain MCP

A server that implements Karpathy’s LLM Wiki Pattern: the AI assistant writes a persistent wiki and the human reads and directs it. It exposes 9 MCP tools:

  • search/read/list — wiki navigation
  • query/compile/crystallize — knowledge synthesis with [[wikilinks]] citations
  • lint/rebuild_index — consistency maintenance
  • ingest_knowledge — incorporation of external research

Infrastructure

30+ Podman containers managed with idempotent Bash automation:

  • an orchestration script with stack discovery
  • an automatic update script with locking and rollback
  • a post-deploy connectivity verification script
  • a proactive storage monitoring script

AI Ecosystem Stack

This ecosystem is built with tools that are part of my broader stack:

  • 🐳 Docker/Podman — Containerization, networks, volumes, systemd quadlets
  • 🐍 Python — Agent logic, MCP servers, automation scripts
  • 📝 YAML — Infrastructure configuration, docker-compose, agent definitions
  • 🔧 Bash — Orchestration scripts, health checks, automatic recovery

My full stack also includes C, C++, embedded firmware, and microcontrollers — visible in my electronics systems projects.