aldegad/tenstorrent

This repository, aldegad/tenstorrent, is a small skill that teaches AI coding assistants such as Claude Code, OpenAI's Codex, and projects built on the Agent SDK how to talk to Tenstorrent's hosted API at console.tenstorrent.com. A skill in this context is a short instruction file plus a handful of examples that the assistant loads to learn a new external service. The Tenstorrent Console hosts two things the skill targets. The first is an OpenAI-compatible chat endpoint that serves a few open-weight models, including DeepSeek-R1, Qwen3-32B, and the multi-modal Qwen3-VL. The second is a text-to-video endpoint running Wan 2.2, where you submit a prompt and get back a video job. To install the skill, the README asks the user to drop the SKILL.md file into one of two known skill folders, either ~/.claude/skills/tenstorrent or ~/.codex/skills/tenstorrent. Once it is there, the matching assistant picks it up and follows the instructions inside. Authentication is one environment variable, TENSTORRENT_KEY, holding an API key generated from the Tenstorrent console. The README is explicit about not committing .env files or real keys. The two curl examples show the actual HTTP shape the skill teaches: a POST to /v1/chat/completions with a model name, messages array, and max_tokens for chat, and a POST to /v1/video/jobs with model Wan2.2-T2V-A14B-Diffusers, a prompt, and an optional negative_prompt for video. The chat example also extracts the reply with jq. The repository is small. There is a SKILL.md, a folder called examples that holds runnable versions of the commands above, a LICENSE file, and the MIT license is stated in the README. There is no build step, no runtime of its own, and no dependencies beyond curl and jq for the examples.