hmbown

DeepSeek TUI is a terminal-native coding agent built around DeepSeek V4's 1M-token context and prefix cache. It ships as a single binary (no Node or Python runtime) with a built-in MCP client, sandbox, and durable task queue.

Skill plus MCP server that turns your agent into an RLM — load context, iterate with search/code/think tools, converge on answers.

Dialectical reasoning architecture for LLMs — runs Thesis, Antithesis, Synthesis loops to improve answer quality on hard questions.

Homebrew tap for the DeepSeek TUI terminal client.

Terminal-first control plane for NVIDIA Nemotron — agentic coding, RAG, doc-ops, and multi-model formations.

Unofficial Hermes Agent skill pack built from fantasy spell and skill names.

Triton-style kernel toolkit for MLX. Prototype, benchmark, and upstream fused kernels for Apple Silicon ML workloads.

RLM agent harness built on top of Deep Agents for running reinforcement-learning-style multi-agent workflows.

Rust-based CLI for working with MiniMax models.

Logic architecture inspired by Toulmin — forces LLMs into structured, sequential reasoning through Toulmin's argumentation model.

NVIDIA capability registry and Spark-aware routing layer for Hermes Agent and NemoClaw users.

Agentic harness for keeping research-oriented LLM runs on-task with contracts, evidence, dialectics, verification, and anti-slop gates.

Apple Silicon MLX fine-tuning toolkit — SFT, DPO/ORPO, GRPO, distillation, and OpenAI-compatible serving.

Logic architecture for abductive inference. Forces LLMs to make uncertainty visible through explicit hypothesis generation and Inference to Best Explanation.

Hermes agent profile and skill pack for structured engineering work, modeled on the ten Orders of the Knights Radiant.

Multi-perspective reasoning harness that queries multiple LLM CLIs in parallel with different perspectives.