--- title: "Together AI: review, setup & limitations · Undominated.ai" canonical: https://undominated.ai/skills/togethercomputer-together-ai/ description: "Routes Together AI work to domain guides, with rules against retrying unrecoverable errors and leaving billable resources running." --- # Together AI: review, setup & limitations · Undominated.ai > Routes Together AI work to domain guides, with rules against retrying unrecoverable errors and leaving billable resources running. [← Explore all skills](/skills/) AI APPLICATION DEVELOPMENT / Together AI # Together AI Routes Together AI work to domain guides, with rules against retrying unrecoverable errors and leaving billable resources running. [See setup guidance ↓](#setup)[Original source ↗](https://github.com/togethercomputer/skills/tree/f59f28e7e7d5937e50581fcfef87b957e8ee8636/skills/together-ai) SOURCE REVIEW Reviewed 2026-10-13 Evidence 4 linked sources Publisher Together AI Licence [MIT ↗](https://github.com/togethercomputer/skills/blob/f59f28e7e7d5937e50581fcfef87b957e8ee8636/LICENSE) Revision f59f28e7e7d5 [Read what was—and wasn’t—checked ↓](#review) “Together AI platform skill covering every product area: chat completions and streaming text generation, tool calling, structured outputs, reasoning and vision models; image generation and editing (FLUX, Kontext); video generation; text-to-speech and speech-to-text; embeddings and reranking; fine-tuning (LoRA, full, DPO, VLM, BYOM); batch inference; LLM-as-a-judge evaluations; code sandboxes; dedicated model inference (tg beta endpoints, deployments, autoscaling, traffic splits, model and LoRA uploads); dedicated containers (Sprocket, Jig); GPU clusters (Kubernetes, Slurm) with Kueue or Volcano scheduling; and TogetherLink (tlink, tclaude) for running Claude Code, Codex, OpenCode, or desktop apps on Together models. Reach for it whenever the user builds, debugs, deploys, trains, evaluates, or operates anything on Together AI, or mentions the together SDK, the tg CLI, TogetherLink, api.together.ai, TOGETHER_API_KEY, or a Together-hosted model id. The skill body routes each request to the matching domain guide.” [Together AI · upstream description ↗](https://github.com/togethercomputer/skills/blob/f59f28e7e7d5937e50581fcfef87b957e8ee8636/skills/together-ai/SKILL.md) Our analysis follows below. 01 / THE REASONING ## Why this made the selection. - Dedicated endpoints, containers and clusters bill while they run; the file says to create them only when asked, delete what was created and confirm it is gone. - Every claimed action, duration, count or split must trace to a tool result or API response, and 400, 401, 402, 403 and 404 errors are not retried. ### A good fit for - Calling Together serverless models, batch inference or fine-tuning from code - Checking whether a Together model ID is still served before building on it ### Weigh up before choosing - Model IDs in the guides are examples checked 2026-10-07, per the file. The frontmatter still lists embeddings and reranking, while domains/embeddings.md says Together does not currently offer them. - Only SKILL.md and domains/embeddings.md were read; the other domain guides, references and scripts were not. The install command does not pin this revision. 02 / THE REVIEW RECORD ## What we actually inspected. Source review has boundaries. A clear record is more useful than a “safe” badge. ### Material inspected - skills/together-ai/SKILL.md (setup, guides table, rules for every area) - skills/together-ai/domains/embeddings.md - README.md (skills overview, installation) - LICENSE MIT, copyright 2025 Together AI ### Our findings - client.models.list() and tg beta models public cannot confirm serverless availability, per the file; a max_tokens=1 chat request can. - The file lists v1-to-v2 SDK renames, such as create_batch to batches.create. ### Not established by this review - The installer was not run, and the skill was not executed against a project or service. No Together API call was made. - The other domain guides, references/ and scripts/ were not opened. The review applies to the material and revision named here. A newer upstream release can change its behavior. 03 / PUT IT TO WORK ## Add a skill to your workflow. [Upstream setup instructions ↗](https://github.com/togethercomputer/skills/blob/f59f28e7e7d5937e50581fcfef87b957e8ee8636/README.md) DOCUMENTED COMMAND npx skills add togethercomputer/skills Copy command ↗ Run in your terminal to choose your AI tools, project or global scope, and installation method. This short command follows the current upstream version; the review covers the revision shown above. - The README documents npx skills add togethercomputer/skills, or copying skills/together-* into the host's skills directory. - The command does not pin revision f59f28e7e7d5937e50581fcfef87b957e8ee8636. Compare the installed directory with that revision. - Set TOGETHER_API_KEY in the environment. tg login is interactive and the file asks the user to run it. ### Before you start - Node.js and npm for the documented Skills installer - A Together AI API key, and uv for the Python SDK and the tg CLI ### Compatibility Agent Skills-compatible hosts · Claude Code · Cursor · Codex · Gemini CLI · Together AI API ### Implementation Python · TypeScript ### Together AI account API and CLI calls - Calls Together APIs with TOGETHER_API_KEY, and the tg CLI for dedicated inference, clusters and fine-tuning. - Can create billable dedicated endpoints, containers and clusters when asked; the file requires deleting what it created. ### Cost model Source is MIT. Serverless tokens, batch jobs, fine-tuning, dedicated endpoints and GPU clusters are billed by Together AI. Those charges were not measured. 04 / FOLLOW THE EVIDENCE ## The source trail. Our notes are separate from the original resource. Check upstream before adopting a new version. - [Skill definition ↗](https://github.com/togethercomputer/skills/blob/f59f28e7e7d5937e50581fcfef87b957e8ee8636/skills/together-ai/SKILL.md) Checked 2026-10-13 https://github.com/togethercomputer/skills/blob/f59f28e7e7d5937e50581fcfef87b957e8ee8636/skills/together-ai/SKILL.md Supports: summary, upstreamDescription, whySelected, limitations, access, review - [Embeddings domain guide ↗](https://github.com/togethercomputer/skills/blob/f59f28e7e7d5937e50581fcfef87b957e8ee8636/skills/together-ai/domains/embeddings.md) Checked 2026-10-13 https://github.com/togethercomputer/skills/blob/f59f28e7e7d5937e50581fcfef87b957e8ee8636/skills/together-ai/domains/embeddings.md Supports: limitations, review - [Install instructions and skill index ↗](https://github.com/togethercomputer/skills/blob/f59f28e7e7d5937e50581fcfef87b957e8ee8636/README.md) Checked 2026-10-13 https://github.com/togethercomputer/skills/blob/f59f28e7e7d5937e50581fcfef87b957e8ee8636/README.md Supports: install, compatibility - [Repository licence ↗](https://github.com/togethercomputer/skills/blob/f59f28e7e7d5937e50581fcfef87b957e8ee8636/LICENSE) Checked 2026-10-13 https://github.com/togethercomputer/skills/blob/f59f28e7e7d5937e50581fcfef87b957e8ee8636/LICENSE Supports: license KEEP COMPARING ## Other approaches to consider. Related by category or shared topics. These are alternatives to inspect, not a measured quality order. [### Agent Platform GenAI Inference ↗ Calls Gemini and third-party OpenMaaS models on Google Cloud Agent Platform, asking for model, project and region instead of defaulting.](/skills/google-agent-platform-inference/)[### Amazon Bedrock ↗ Bedrock model invocation, Knowledge Bases, Guardrails and AgentCore guidance, with quota, prompt-caching and cost-tracking workflows.](/skills/aws-amazon-bedrock/)[### Anthropic Claude API ↗ Routes Claude API work to language-specific SDK guidance and distinguishes self-hosted tool loops from managed agent execution.](/skills/anthropics-claude-api/) [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/) ## Continue your investigation - [Choose a toolkit](/tools/) - [Inspect execution hosts](/agents/) - [Check tool access](/mcp-servers/) - [Check project dependencies](/check-repo/)