Hunter Bown founded ShannonTek and created Codewhale, an open-source terminal coding agent written in Rust. His work includes tools that connect models to code, persistent working state and larger investigations.[1][2]
Start with Codewhale →
X Career
Contributions
An agent loop in the terminal
Codewhale reads a repository, edits files and runs commands while using the model provider the user chooses. It supports hosted and local endpoints, with configurable approval modes. Its receipts interface lists changed files, executed commands and external calls, giving the operator a record to inspect after a task. The software’s current repository identifies DeepSeek-TUI as its earlier name.[2][3][5]
Working state outside the model prompt
Aleph keeps loaded content, search indexes, evidence and code-execution state in a Python process rather than repeatedly copying raw material into a model’s limited prompt window. The assistant calls search, inspection and execution tools to obtain smaller results, and can break questions into subqueries. This changes how a long investigation accesses its source material; it does not expand the model’s context window itself.[4]
Tools, writing & talks
Tools
Start here
Codewhale
github.comAn open-source Rust coding agent that reads projects, changes files and runs commands through a chosen model provider.
Start with the current README and approval options. You need a terminal, a project folder and a model connection; the receipts view helps inspect what a run actually did.
- Knights Radiantgithub.com
Hermes profile and skill pack that structures engineering work around the ten Knights Radiant Orders.
Catalog workOriginal source - Autodialecticsgithub.com
Agentic harness that keeps research LLM runs on task using contracts, evidence, verification, and anti-slop gates.
Catalog workOriginal source - NeMoCodegithub.com
Terminal control plane for NVIDIA Nemotron covering agentic coding, RAG, doc-ops, and multi-model formations.
Catalog workOriginal source - Midtrygithub.com
Reasoning harness that queries several LLM CLIs in parallel, each prompted from a different perspective.
Catalog workOriginal source - Peircean Abductiongithub.com
Logic architecture that forces LLMs to make uncertainty visible through explicit hypothesis generation.
Catalog workOriginal source - MLXSmithgithub.com
Fine-tuning toolkit for Apple Silicon MLX covering SFT, DPO, GRPO, distillation, and OpenAI-compatible serving.
Catalog workOriginal source - NemoHermesgithub.com
Routing layer for Hermes Agent and NemoClaw that tracks NVIDIA capabilities and stays Spark-aware.
Catalog workOriginal source - Toulminitoulmini-web.pages.dev
Forces LLMs into structured sequential reasoning based on Toulmin's argumentation model.
Catalog workOriginal source - Homebrew DeepSeek TUIgithub.com
Homebrew tap that distributes the DeepSeek TUI terminal client through the brew package manager.
Catalog workOriginal source - MiniMax CLIgithub.com
Command-line client written in Rust and Python for driving MiniMax models from a terminal.
Catalog workOriginal source - RLM Agentsgithub.com
Harness layered over Deep Agents for reinforcement-learning-style workflows across multiple agents.
Catalog workOriginal source - ZMLXgithub.com
Kernel toolkit for MLX to prototype, benchmark, and upstream fused kernels for Apple Silicon ML workloads.
Catalog workOriginal source - Wizards of the Ghostsgithub.com
Unofficial skill pack for the Hermes Agent built from fantasy spell and skill names.
Catalog workOriginal source - Alephgithub.com
An MCP server and skill that keeps searchable context and working state outside an assistant’s prompt window.
Repository owner[4]Original source - Hegelionhegelion-web.pages.dev
Runs thesis, antithesis, synthesis loops with LLMs to improve answer quality on hard questions.
Catalog workOriginal source - DeepSeek TUIgithub.com
Terminal-native coding agent built on DeepSeek V4's 1M-token context, shipped as a single binary with an MCP client.
Catalog workOriginal source