Independent AI researcher developing local-first AI systems designed to keep evidence, provenance, and operating constraints visible.
- Agent systems — bounded, observable orchestration and evaluation.
- Retrieval & memory — durable local research state, provenance, and citation-oriented document systems.
- Inference operations — resource-aware local deployment and evaluation, with conclusions scoped to the evidence and host.
- cli-sub-agent — Rust runtime and
csaCLI for headless, configuration-driven AI-agent orchestration. - Verbatim — Local, daemon-backed document retrieval with stable evidence IDs, locators, and citation-first context packs. Its public issues are research and design artifacts; prompts and proposals remain distinct from shipped code.
- gb10-services — Deployment, configuration, validation, and recovery assets for a specific GB10-class local model stack; findings are scoped to that environment.
Focused tools: hymt (translation) · mvln (move-and-link) · mempal (fork; agent memory).
- Human review — How can agent systems expose evidence and review points while keeping automation bounded and reversible?
- Local retrieval — How can lexical, vector, and provenance-oriented workflows coexist without exceeding resource or durability constraints?
- Evaluation — How should local inference and benchmarking produce reproducible comparisons without turning experiments into production claims?
- Systems: Rust, Python, SQLite, and durable local state.
- Agent workflows: MCP and configuration-driven automation.
- Operations: Local model serving, Docker, and systemd.
- Contact: Telegram
- Public work: RyderFreeman4Logos repositories