# Opus 4 Projects at AI Tinkerers

> Canonical HTML: https://aitinkerers.org/technologies/opus-4
> Markdown URL: https://aitinkerers.org/technologies/opus-4.md
> Technology record last updated: 2026-02-24T06:30:38Z
> Generated: 2026-09-20T16:36:19Z

Opus 4 is a high-performance open-source repository platform built for the professional management and publication of electronic academic documents.

Opus 4 serves as the primary technical foundation for institutional repositories (managed by KOBV). This PHP-driven system utilizes an Apache Solr backend to deliver sub-second search results across massive document collections. It automates the entire publication lifecycle: from initial user upload to long-term archiving and metadata distribution via OAI-PMH 2.0 interfaces. By adhering to DINI certificate requirements, it ensures that research outputs remain discoverable and persistent through standardized persistent identifiers (URNs) and rigorous metadata validation using Dublin Core standards.

- Official technology site: https://opus4.kobv.de/
- Public AI Tinkerers demos and talks: 10
- Result page: 1 of 1

## Recent Public Talks and Demos

### [Will people want your product? Bots can help you find out (and fix it)](https://seattle.aitinkerers.org/talks/rsvp_tYS3msXujf8)

I've made a set of agents that will pressure test your product idea to help it create better product/market fit. I will demo the bots that I trained on my personal 20-years of product leadership experience. I will show how they can help you sharpen your idea to focus on your ICP, do research to help validate it, and create messaging to reach that ICP.

- Event context: GTM / Growth Track - Seattle — 2026-05-27 — Seattle
- Public talk page: https://seattle.aitinkerers.org/talks/rsvp_tYS3msXujf8

### [Stateful agents with open-strix](https://raleigh.aitinkerers.org/talks/rsvp_obcR2zOygIM)

open-strix is a minimalistic open source stateful agent harness that leans on the Unix principle and uses cybernetics principles to build a tiny, solid, extensible core.

- Event context: AI Tinkerers Raleigh Meetup — May 6, 2026 — 2026-05-06 — Raleigh
- Public talk page: https://raleigh.aitinkerers.org/talks/rsvp_obcR2zOygIM

### [Multi-agent Claude Code at scale: building Parthas at Intercom](https://dublin.aitinkerers.org/talks/rsvp_qYginBx6EOk)

Parthas is a multi-agent orchestrator for Claude Code that coordinates parallel coding agents through software instead of agent-to-agent messaging. In this demo I'll run it live against a real codebase. I'll spin up several agents in isolated worktrees, show the permission queue and the web dashboard, and walk through the design principles that keep it fast without YOLO-ing your main branch.

- Event context: AI Tinkerers Dublin: Demo Night with Fin — 2026-04-28 — Dublin
- Public talk page: https://dublin.aitinkerers.org/talks/rsvp_qYginBx6EOk

### [Connecting Chicago engineers with the right companies](https://chicago.aitinkerers.org/talks/rsvp_TXG7B89_Hys)

A hyperlocal talent marketplace for engineers

- Event context: AI Tinkerers Chicago: April Meetup ft OneTwoLoop — 2026-04-14 — Chicago
- Public talk page: https://chicago.aitinkerers.org/talks/rsvp_TXG7B89_Hys

### [From 1 Agent to 11: Patterns for Multi-Agent Systems That Actually Work](https://columbus.aitinkerers.org/talks/rsvp_nZaEJ2xzX0c)

I built an AI operating system (AIOS) that runs my technology community organization, Cloud Nirvana, using 11 specialized AI agents coordinated from a Mac Mini. Each agent owns a domain: email triage, speaker pipeline QA, partnership tracking, CRM operations, content, community engagement, financial operations, and strategic planning. They share an encrypted CRM, communicate through file-based memory, and operate under a graduated trust model where each agent earns autonomy based on demonstrated reliability. Along the way, everything broke at least once. I documented those failures as reusable engineering patterns (inspired by the Gang of Four) and open-sourced them. The demo walks through the live system, shows agents handling real work in real time, and demonstrates how any builder can discover and apply these patterns against their own codebase.

- Event context: AI Tinkerers - Columbus April Meetup — 2026-04-06 — Columbus
- Public talk page: https://columbus.aitinkerers.org/talks/rsvp_nZaEJ2xzX0c

### [Building a Terminal Orchestrator for AI Coding Agents in Rust](https://houston.aitinkerers.org/talks/rsvp_AD3Q9uzasnc)

**Context Foundry** is an autonomous build loop written in Rust. Give it a task list in markdown, and it works through every task using AI agents -- scouting the codebase, planning, building, and verifying with an independent reviewer in a fresh context. Passing tasks get committed. When the list runs out, a discovery agent scans for new work and keeps going. ``` SCOUT ──▶─ PLAN ──▶─ IMPLEMENT ──▶─ DOUBT ``` **The core insight:** every agent starts with a clean context window and receives only curated artifacts from the previous stage -- not a bloated conversation history full of noise. The scout writes a structured report. The planner reads that report and writes a plan. The builder reads that plan and writes code. The verifier reads the code with zero knowledge of why it was written that way. No shared context windows, no accumulated reasoning, no inherited blind spots. Each stage gets signal, not noise. This is how foundry prevents the compounding error problem where task 3 builds on task 2's mistakes. On top of this, **pattern learning** extracts reusable lessons after each task and injects them into future runs across all projects. ``` - [x] T1.1: Set up scaffolding [SPID] ● feat - [x] T1.2: Implement auth [SPID!] ✗ WIP - [ ] T1.3: Write tests [....] ``` You can also race two models against each other -- Claude and Codex each get their own worktree and run the full pipeline independently, so you compare finished solutions, not just raw outputs. Simple tasks skip the planner and reviewer and commit in about 30 seconds. There are three run modes: Auto keeps going forever, Sprint stops when the list is done, Review creates a PR after each task and waits for approval. Foundry has completed 141 tasks on its own codebase across 33 discovery rounds. Most of the features described here were built by the loop itself.

- Event context: AI Tinkerers Houston Meetup (March 10, 2026) — 2026-03-10 — Houston
- Public talk page: https://houston.aitinkerers.org/talks/rsvp_AD3Q9uzasnc

### [Building for the Future Internet: Agent Services](https://upstate-ny.aitinkerers.org/talks/rsvp_GetoIsT3wio)

meetSpace - exploring an agent-first service for creating and finding local IRL events.

- Event context: AI Tinkerers Upstate NY: March Meetup — 2026-03-10 — Upstate NY
- Public talk page: https://upstate-ny.aitinkerers.org/talks/rsvp_GetoIsT3wio

### [How We Build Pipelines to Detect Product Issues and Fix Them with Coding Agents](https://prague.aitinkerers.org/talks/rsvp_C35CGsy9rLA)

I'll walk through how we built a pipeline that detects product issues from multiple data sources (session recordings, error logs, support tickets, LLM traces) and routes them to coding agents that generate PRs. * Step 1: Signal. Render videos from session recordings, transcribe them, and analyze, searching for issues. * Step 2: Cluster. Group signals from different sources and decide what’s worth fixing. * Step 3: Code. Preview of Twig ADE (AI Development Environment), where confirmed tasks get handed off to coding agents.

- Event context: Product Engineering: Apify x PostHog — 2026-02-26 — Prague
- Public talk page: https://prague.aitinkerers.org/talks/rsvp_C35CGsy9rLA

### [How to build good skills for LLMs](https://dhaka.aitinkerers.org/talks/rsvp_CJHoJK54W44)

Skills are the SOTA for providing capabilities to LLMs while making sure to not flood the context with all the tokens tools/ MCPs do. In this talk I will present some rules of thumb for building skills that LLMs will use and use well, sourced from Anthropic’s guidance, their course and my on-the-ground experience writing skills for my own use.

- Event context: AI Tinkerers Dhaka 2nd Meetup: Dhaka Builds with AI! — 2026-02-07 — Dhaka
- Public talk page: https://dhaka.aitinkerers.org/talks/rsvp_CJHoJK54W44

### [Writing an arXiv Paper (in a Day)](https://tokyo.aitinkerers.org/talks/rsvp_giYhKcFdlUE)

Recently, we open-sourced several of the most useful evals we used for developing our Shisa V2 models. One of the most useful was JP-TL-Bench, our Japanese/English translation eval. It's notable because it introduces a brand new methodology for doing better scoring (combining the discriminative power of pair-wise completions, but avoiding both the quadratic scaling and the score drift that come with normal scoring, like ELO). It's worth writing a paper about. How can we best use AI to help us write a paper efficiently, without it being complete slop?

- Event context: AI Tinkerers Ginza Meetup – January 15, 2026 — 2026-01-15 — Tokyo
- Public talk page: https://tokyo.aitinkerers.org/talks/rsvp_giYhKcFdlUE

## Related Technologies

- [Claude Code](https://aitinkerers.org/technologies/claude-code) ([Markdown](https://aitinkerers.org/technologies/claude-code.md)) — 214 public demos
- [Claude](https://aitinkerers.org/technologies/claude) ([Markdown](https://aitinkerers.org/technologies/claude.md)) — 173 public demos
- [AGENTS](https://aitinkerers.org/technologies/agents) ([Markdown](https://aitinkerers.org/technologies/agents.md)) — 20 public demos
- [Codex](https://aitinkerers.org/technologies/codex) ([Markdown](https://aitinkerers.org/technologies/codex.md)) — 44 public demos
- [Cursor](https://aitinkerers.org/technologies/cursor) ([Markdown](https://aitinkerers.org/technologies/cursor.md)) — 65 public demos
- [Python](https://aitinkerers.org/technologies/python) ([Markdown](https://aitinkerers.org/technologies/python.md)) — 662 public demos
- [Agentic](https://aitinkerers.org/technologies/agentic) ([Markdown](https://aitinkerers.org/technologies/agentic.md)) — 3 public demos
- [Botspring](https://aitinkerers.org/technologies/botspring) ([Markdown](https://aitinkerers.org/technologies/botspring.md)) — 1 public demo
- [Claude Code Agent SDK](https://aitinkerers.org/technologies/claude-code-agent-sdk) ([Markdown](https://aitinkerers.org/technologies/claude-code-agent-sdk.md)) — 3 public demos
- [Claude Code CLI](https://aitinkerers.org/technologies/claude-code-cli) ([Markdown](https://aitinkerers.org/technologies/claude-code-cli.md)) — 5 public demos
- [Claude Opus](https://aitinkerers.org/technologies/claude-opus) ([Markdown](https://aitinkerers.org/technologies/claude-opus.md)) — 18 public demos
- [clustering](https://aitinkerers.org/technologies/clustering) ([Markdown](https://aitinkerers.org/technologies/clustering.md)) — 3 public demos
- [comment](https://aitinkerers.org/technologies/comment) ([Markdown](https://aitinkerers.org/technologies/comment.md)) — 1 public demo
- [Database](https://aitinkerers.org/technologies/database) ([Markdown](https://aitinkerers.org/technologies/database.md)) — 8 public demos
- [Embeddings](https://aitinkerers.org/technologies/embeddings) ([Markdown](https://aitinkerers.org/technologies/embeddings.md)) — 22 public demos
- [FastAPI](https://aitinkerers.org/technologies/fastapi) ([Markdown](https://aitinkerers.org/technologies/fastapi.md)) — 181 public demos
- [Firebase Hosting](https://aitinkerers.org/technologies/firebase-hosting) ([Markdown](https://aitinkerers.org/technologies/firebase-hosting.md)) — 2 public demos
- [Flutter](https://aitinkerers.org/technologies/flutter) ([Markdown](https://aitinkerers.org/technologies/flutter.md)) — 24 public demos