Together AI Skill Pack

18 production-ready Claude Code skills for Together AI -- real OpenAI-compatible inference, fine-tuning, and model deployment code.

What This Is

A complete skill pack for building AI applications with Together AI. Every skill contains real API code: chat completions, streaming, image generation, embeddings, fine-tuning jobs, batch inference, and model management. Uses the official together Python SDK and OpenAI-compatible Node.js client.

Installation

/plugin install together-pack@claude-code-plugins-plus

Skills

Standard Skills (S01-S12)

| # | Skill | What It Does | |---|-------|-------------| | S01 | together-install-auth | Install together SDK, configure API key, verify with model list | | S02 | together-hello-world | Chat completions, streaming, image generation, embeddings | | S03 | together-local-dev-loop | Mock responses, test fixtures, environment separation | | S04 | together-sdk-patterns | OpenAI client wrapper, model selection, typed responses | | S05 | together-core-workflow-a | Fine-tuning: JSONL prep, upload, create job, monitor, deploy | | S06 | together-core-workflow-b | Batch inference, dedicated endpoints, model comparison | | S07 | together-common-errors | Fix 401, model not found, rate limits, timeout errors | | S08 | together-debug-bundle | API connectivity, model availability, usage diagnostics | | S09 | together-rate-limits | Per-model rate limits, request throttling, queue management | | S10 | together-security-basics | API key protection, input validation, output filtering | | S11 | together-prod-checklist | Model selection, fallback chains, monitoring, health checks | | S12 | together-upgrade-migration | SDK version updates, model deprecations, API changes |

Pro Skills (P13-P18)

| # | Skill | What It Does | |---|-------|-------------| | P13 | together-ci-integration | GitHub Actions for model testing, fine-tuning validation | | P14 | together-deploy-integration | Deploy inference endpoints to Vercel/Cloud Run | | P15 | together-webhooks-events | Fine-tuning job status polling, completion callbacks | | P16 | together-performance-tuning | Streaming, model warm-up, concurrent requests | | P17 | together-cost-tuning | Model tier pricing, batch inference (50% off), caching | | P18 | together-reference-architecture | AI service layer, model routing, fallback chains |

Key Concepts

• Base URL: https://api.together.xyz/v1 (OpenAI-compatible)
• Models: Llama 3.3, Mixtral, Qwen 2.5, DeepSeek V3, FLUX.1
• Endpoints: /chat/completions, /completions, /embeddings, /images/generations
• Fine-tuning: Upload JSONL, create job, monitor, deploy custom model
• Batch: 50% cost reduction for async processing
• Pricing: $0.10-5.00 per 1M tokens depending on model size

License

MIT