kurumi
kurumi is an individual developer and deep-learning graduate student who publishes personal research, engineering notes, and open-source projects. Their stated interests are machine learning/deep learning, with additional work in computer graphics, AI agents, real-time voice, and software engineering. (Sources: https://cryer.github.io/about/ and https://github.com/cryer)
At a Glance
- AI/LLM developers and teams needing regression evaluation
- Developers who want a customizable terminal coding agent
- Hobbyists and players interested in voice-based solo TRPGs
- Researchers, students, and engineers learning deep learning, graphics, and systems programming
AI Tools by kurumi
(1)agent-eval from cryer
CI Regression Testing for LLM Agents
Discussions
No discussions yet
Be the first to start a discussion about kurumi
Latest News
agent-eval hardened statistical and calibration gates in its latest public commit
Published AI-TRPG, a voice-based solo tabletop RPG with an AI game master
Published Star CLI, a TypeScript agent command-line interface
Published a post describing a self-built modular C++/OpenGL 3D game engine
Products & Services
Open-source CI/evaluation framework for LLM agents: replay suites after prompt, model, tool, or loop changes and use statistical gates to detect regressions. It includes OpenTelemetry tracing, deterministic assertions, LLM-as-judge, simulated/adversarial users, coding-agent workspace assertions, SQLite manifests, reports, and GitHub Actions integration. MIT licensed. (Sources: https://github.com/cryer/agent-eval and https://raw.githubusercontent.com/cryer/agent-eval/master/README.md)
TypeScript command-line AI-agent interface distributed as @cryer/star-cli. It provides a streaming REPL, OpenAI/Anthropic/OpenAI-compatible providers, filesystem/bash/web tools behind permissions, plan mode, subagents, session persistence, context compaction, image input, Git integration, hooks, background tasks, and NDJSON output. (Sources: https://cryer.github.io/2026/09/starcli/ and https://github.com/cryer/star-cli)
Voice-first solo tabletop-RPG application: an AI script recommender hands off to a scenario-specific game master, with free-form player actions, code-generated dice rolls, memory, interruption/barge-in handling, NPC voices and multiple built-in genres/scripts. (Sources: https://cryer.github.io/2026/10/ai-trpg/ and https://github.com/cryer/AI-TRPG)
A hand-built, low-latency full-duplex voice-agent pipeline with independently switchable ASR, LLM, and TTS providers; the repository documents Web and Android clients, local/cloud provider choices, interruption handling, VAD, echo-control concerns, and latency measurement. (Sources: https://github.com/cryer/Realtime-Voice and https://cryer.github.io/2026/10/ai-trpg/)
Market Position
kurumi is best characterized as an independent open-source builder rather than a company competing for enterprise customers. Star CLI is positioned as a small, transparent, highly configurable alternative to agent CLIs such as Claude Code, Codex, and Gemini CLI, with particular emphasis on arbitrary OpenAI-compatible relays, permission controls, tool use, and robust streaming/error handling; agent-eval occupies the open-source LLM-agent testing/observability niche. (Sources: https://cryer.github.io/2026/09/starcli/ and https://github.com/cryer/agent-eval)
Founding Story
This is not presented as a company: the official site describes itself as a personal website and says its author is a deep-learning graduate student who likes researching new technologies. The work appears to have grown from personal experimentation and learning projects, documented on the blog since at least 2016. (Sources: https://cryer.github.io/about/ and https://cryer.github.io/2016/05/vim/)
Business Model
Revenue Model
The subject presents the work as personal and open source; public materials show self-hosted software, API-provider integrations, and MIT-licensed repositories rather than a paid commercial offering. No monetization model is stated.
Target Markets
- AI/LLM developers and teams needing regression evaluation
- Developers who want a customizable terminal coding agent
- Hobbyists and players interested in voice-based solo TRPGs
- Researchers, students, and engineers learning deep learning, graphics, and systems programming
- Regression testing and CI gating for LLM agents
- Local developer workflows using configurable AI coding agents
- Voice-first interactive entertainment and solo tabletop role-playing
- Real-time conversational voice applications
- Learning and prototyping in deep learning, computer vision, graphics, CUDA/C++, and generative AI