Инструменты

Download and run open models such as Gemma locally with one command. Provides a REST API and Python/JS libraries, and ollama launch can point Claude Code, Codex and other agents at local models.

Reference for the Claude API and Anthropic SDK: model selection and pricing, streaming, tool use, prompt caching, Managed Agents and more, plus workflows for model migration and prompt audits. Official Anthropic skill.

Scaffolds a Claude Agent SDK project in Python or TypeScript interactively with /new-sdk-app, then a verifier agent checks dependencies, config and code against official best practices. Maintained by Anthropic.

Local LLM inference engine in plain C/C++ with a command-line chat and an OpenAI-compatible HTTP server; pulls and runs GGUF models straight from Hugging Face. Supports Apple Silicon, CUDA, Vulkan and other hardware, plus low-bit quantization.

Google's official open-source terminal agent using Gemini models with a 1M-token context. Built-in Google Search, file and shell tools, plus MCP support. Personal Google accounts get a free quota.

High-throughput LLM inference and serving engine: vllm serve starts an OpenAI- and Anthropic-compatible server in one command. Supports 200+ model architectures from Hugging Face, quantization and distributed inference. Aimed mainly at NVIDIA, AMD and other GPUs.

Provides a sequential_thinking tool that guides the model to break complex problems into revisable, branching steps. Good for planning, analysis and tasks that need careful iteration. Official MCP reference server.

All-in-one LLM fine-tuning framework: llamafactory-cli handles pretraining, SFT, PPO/DPO alignment and LoRA/QLoRA training, plus chat inference, weight merging and export, and OpenAI-style API serving for 100+ open models. Has a Web UI; training usually needs a GPU.

Detects your CPU, memory and GPU and scores open models on whether they run, speed, quality and context, then works with local runtimes like Ollama, llama.cpp and MLX to download and benchmark them. recommend --json gives agent-friendly output.

Local model gateway that routes requests from Claude Code, Codex and other coding agents to providers like DeepSeek, Kimi, Gemini and OpenRouter, with failover, rule-based routing and request logs. Bring your own API keys.

ModelScope's official training framework for LLMs and multimodal models: the swift command covers pretraining, fine-tuning, RL, inference, evaluation, quantization and deployment for 600+ text and 400+ multimodal models, with Megatron parallelism and a web UI. Usually needs a GPU.

Writes precise prompts for a specific AI tool: identifies the target tool, extracts your intent, asks at most three questions if details are missing, then outputs a lean, paste-ready prompt. Covers dozens of tools including ChatGPT, Claude Code, Cursor, Midjourney and Sora.

Nearly 100 skills spanning the AI research workflow: ideation and literature review, fine-tuning and post-training, distributed training, inference serving, evaluation, interpretability and paper writing, tied together by an autoresearch orchestration skill.

Plugin marketplace for AI research with nearly 100 skills in 23 categories, from ideation, fine-tuning and post-training to distributed training, inference serving, evaluation and paper writing, coordinated end to end by the autoresearch skill.

Simon Willison's LLM CLI and Python library: call OpenAI, Claude, Gemini and local models from the terminal, with conversations logged to SQLite automatically, plus embeddings, structured extraction, tool calling and a plugin system.

Teaches agents the Hugging Face Hub CLI hf: search and download models, manage datasets, Spaces, Buckets and repos, and run Jobs and inference endpoints. The skill content is generated from your local CLI, so install hf first. Maintained by Hugging Face.

Hugging Face's official skills: manage Hub models and datasets with the hf CLI, fine-tune language and vision models, run model evaluations, and build Gradio apps deployed to Spaces. Needs an HF account; cloud training on HF Jobs requires a paid plan.

Lets agents work with the Hugging Face Hub through the hf CLI: search and download models and datasets, manage repos and Spaces, and run training and evaluation jobs on cloud GPUs, with more workflow skills installable as needed. Maintained by Hugging Face.

Lets agents fine-tune language and vision models with TRL or Unsloth on Hugging Face Jobs cloud GPUs, supporting SFT, DPO, GRPO and reward models, and convert results to GGUF for local use. Requires an HF Pro or higher paid plan. Maintained by Hugging Face.

NVIDIA's open-source LLM vulnerability scanner that probes for prompt injection, jailbreaks, data leakage, hallucination and harmful content, supporting OpenAI, Hugging Face, Bedrock and most REST-accessible models. Scanning cloud models needs their API keys.

Manage Kubernetes clusters in natural language: translates intent into kubectl operations and runs them, and can serve as an MCP server for other agents. Uses Gemini by default (API key required) and also supports OpenAI and local models. Maintained by Google Cloud.

Open-source LLM evaluation platform: one opencompass command picks the model and datasets and runs the evaluation across hundreds of benchmarks. Works with local Hugging Face models and API models, with vLLM and LMDeploy for faster inference.

Build with OpenAI best practices inside Codex: call the OpenAI API, build and deploy agents with the Agents SDK, create ChatGPT apps and prepare submission materials, and create API keys directly. Maintained by OpenAI; Codex only.

LM Studio's official CLI: list, load and unload local models, start and stop the local API server, and check status and logs; ls, ps and other commands support --json output for scripts and agents. Requires the LM Studio desktop app.