3. LLM Nodes

LLM nodes let you run AI models inside your workflow. They are used for text summarization, classification, translation, content generation and decision-making steps. They are located under the LLM tab in the node catalog.

LLM nodes are not used on their own: they receive input from a previous node, send that input to the selected model and pass the generated result on to the next node. The basic flow is the same in every example:

Text Input → LLM node → Text Output

• Text Input carries the text typed by the user or the data coming from a previous node.
• LLM node processes this data according to the selected model and operation.
• Text Output displays the generated response or passes it on for use in later steps.

Node Groups​

LLM nodes in SkyStudio are organized into four groups:

Connection Setup​

Built-in API nodes and GOAT do not require a credential connection. Model access is provided by SkyStudio; you do not need to enter your own API key.

Note: Among the built-in nodes, only the Claude Web Search node may require a connection to access the web search provider. In that case, the node shows a "Credential connection required" warning.

BYOK and BYOM nodes, on the other hand, connect to your own provider account or model infrastructure. For these nodes, you first create a connection in the App Connections area and then select that connection in the node. Step-by-step setup is described in each provider's own section (see the BYOK — Using a Provider with Your Own API Key and BYOM — Bringing Your Own Model sections).

Built-in API Nodes​

The built-in API nodes group contains 3 nodes:

Common Panel Layout​

All three nodes use the same panel layout:

OpenAI API Call​

Sends a request to OpenAI models and passes the response into the flow.

Parameters​

Model Selection​

Usage Example​

To summarize a meeting note:

• Model: GPT-5.4 Mini
• System Prompt: You are a meeting summary assistant. Respond in English and use at most 5 bullet points.
• Query: Click the field and select the previous node's text output with /

Sample Output​

{
  "answer": "Three decisions were made in the meeting: the budget was increased by 15%, the launch was postponed, and a new team member will be hired.",
  "model": "gpt-5.4-mini",
  "tokens_used": 412
}

Anthropic API Call​

Sends requests to Anthropic Claude models. Preferred for long document analysis and tasks that require careful reasoning.

Parameters​

Model Selection​

Usage Example​

To analyze a long contract:

• Model: Claude 4.8 Opus
• System Prompt: You are a contract analysis expert. Identify risky clauses and write a justification for each. Do not speculate on topics you are unsure about.
• Query: Select the file reading node's output with /

Claude Web Search​

Runs the Claude model together with web search capability. Used for questions that require up-to-date information.

Parameters​

Usage Example​

To gather up-to-date industry information:

• Query: What is the latest status of AI regulations in Türkiye?
• System Prompt: List the findings in chronological order and cite a source for each item.

Note: Web search takes longer than a normal LLM call. If up-to-date information is not needed, using the Anthropic API Call node is faster.

Using the System Prompt​

Click the + Add System Prompt button to give the model a role and rules. The Query field defines "what to do"; the System Prompt defines "how to do it".

A Good System Prompt​

Example System Prompt​

You are a customer feedback classification assistant.
Assign each incoming text to one of these categories: Complaint, Suggestion, Praise, Question.
Return only the category name as your answer. Do not write an explanation.

Note: For tasks like classification, writing a clear format rule in the System Prompt such as "return only the category name" makes it easier for the next nodes to process the output.

Using Tools​

When you open the Tools section, you can select tools that the model can call automatically when needed.

How It Works​

1. Click the Tools heading to open the section.
2. Click the + sign next to the tool you want to add from the list.
3. If the tool says "Connection required", set up the relevant app connection.

If the model needs this tool while answering your query, it calls it on its own and includes the result in its response.

Usage Example​

When you write "Summarize the latest email from Ahmet" in the Query, if the Outlook Mail Search tool is added, the model first searches for the email, finds it and then summarizes it.

Note: Adding tools slows the model down and increases cost. Only add tools that are truly needed.

Passing Data into the Query Field​

The "Add parameters with /" hint on the right of the Query field indicates that you can insert previous node outputs.

Methods​

Combining with static text​

Summarize the meeting note below in 5 bullet points:

{{output.text}}

Common Model Parameters​

In GOAT, BYOK and BYOM nodes, the request text is written in the Message field (in some operations Prompt, Query or Input Text(s)) instead of Query. In these fields too, you can type / to insert a previous node's output. The following parameters recur in the chat operations of these nodes:

Note: When numeric fields such as Temperature and Max Tokens are left blank, the model's or provider's default value is used. For workflows moved to production, setting these values explicitly is recommended for consistent results.

GOAT​

GOAT is the built-in chat model available directly within SkyStudio. It can be used for basic text generation, question answering, summarization, drafting and classification flows without requiring an external provider account or API key.

Connection Setup and Management​

The GOAT node does not require a connection. After adding the node to the canvas, the "App Connections" field does not appear in the right panel; you can fill in the parameters directly.

Operations​

The GOAT node offers a single operation:

Chat​

This operation sends the given message to the GOAT model and generates a text response.

Parameters​

Message Example​

Write a polite and short reply to the customer message below:
/

Here, / is used to select the output of the Text Input node.

System Prompt Example​

You are a customer support specialist. Write in English. Do not give uncertain information. Keep your answer to at most three sentences.

Temperature Guide​

0.2–0.5 is recommended for summaries, classification and support replies; around 0.7 for idea generation and marketing copy.

Node Connections​

Text Input is selected as the incoming connection. Text Output is selected as the next node.

Sample Output​

{
  "response": "Hello, we apologize for the issue you experienced. Your request has been forwarded to our team and we will get back to you as soon as possible.",
  "model": "goat",
  "tokens_used": 42
}

Example Workflow with GOAT​

Text Input → GOAT / Chat → Text Output

1. Type, for example, What is AI-powered customer service? into the Text Input node.
2. In the GOAT node, insert the Text Input value into the Message field with /.
3. Write the desired response format in the System Prompt field.
4. Set Text Output as the GOAT node's next connection.
5. Save the workflow and start the run action on the workflow screen.
6. The result is displayed in the Text Output node.

Common Errors​

Notes​

• Since GOAT requires no external account, it can be the first choice for rapid prototyping and test scenarios.
• For high-volume scenarios or those requiring highly specialized domain knowledge, switching to one of the BYOK/BYOM providers is recommended.
• When the System Prompt is left blank, the model's default, general-purpose behavior applies.

BYOK — Using a Provider with Your Own API Key​

BYOK (Bring Your Own Key) lets you use models by connecting your own provider account and API key to SkyStudio. The key is obtained from the relevant provider, added as a secure connection in the App Connections area of SkyStudio, and then selected in the relevant LLM node. Usage and costs are billed according to the connected account's own provider plan; SkyStudio only brokers the connection.

General Setup Process​

The following steps are common to every BYOK node:

1. Obtain an API key from the provider.
2. Create an App Connection for the provider in SkyStudio (App Connections → Add New Connection → select the provider → enter Connection Name and API Key → Save).
3. While creating the connection, set the access level in the Who can use field (Only me / Workspace / Selected users / Integrations).
4. In the node, select the saved connection from the App Connections dropdown.
5. Specify the connection to use in the Connection field (if there is more than one).
6. Fill in the model and operation fields.
7. Connect the Text Input node to the LLM node, and the LLM node to Text Output (or the relevant next node).
8. Run the workflow and check the node result.

Important: API keys must only be entered in the connection field. They must never be written in Message, System Prompt, Prompt or other text fields, as these fields may become visible in the workflow history and logs.

Note: Each provider connection works with that provider's own API key. You can define multiple connections for the same provider for different environments (test/production) or different teams.

Mistral AI​

Mistral AI is a multi-purpose model provider that can be used for chat, embedding, document reading (OCR), code completion and content moderation.

Connection Setup and Management​

Getting an API Key​

1. Sign in to your Mistral AI account.
2. Select the relevant workspace.
3. Go to the API Keys page from the left menu.
4. Click Create new key to create a new key.
5. Copy the generated key immediately.

Note: The key may be shown in full only at the moment it is first created; if you forget to save it, you will need to create a new key. Official guide: Mistral API keys.

Adding the Connection to SkyStudio​

1. Go to App Connections from the SkyStudio left menu.
2. Click Add New Connection.
3. Select Mistral AI as the provider.
4. Fill in the following fields:

5. Click Save.

Assigning the Connection to the Node​

After adding the Mistral AI node to the canvas, select your saved connection from the App Connections dropdown in the right panel.

Operations​

The Mistral AI node offers 5 operations:

Chat​

Used for text generation, question answering, summarization, content writing and information extraction.

Parameters​

Example Workflow​

Text Input → Mistral AI / Chat → Text Output

A customer question is typed into Text Input. This value is linked to the Message field in the Mistral Chat node. Text Output is selected as the next node. When the workflow runs, the response appears in Text Output.

Sample Output​

{
  "response": "Your return process is completed within 14 days. You do not need to upload any additional documents.",
  "model": "mistral-large-latest",
  "finish_reason": "stop",
  "usage": { "prompt_tokens": 28, "completion_tokens": 19 }
}

Generate Embedding​

Converts text into a numeric vector. Used in similarity search, document matching, RAG (Retrieval-Augmented Generation) and semantic search scenarios.

Parameters​

Example Usage​

Text Input → Mistral AI / Generate Embedding → Next vector or search node

The output of this node is embedding data rather than ordinary user text. Therefore, instead of connecting it directly to Text Output, it is better to connect it to a search or data node that accepts embeddings.

Sample Output​

{
  "embeddings": [
    { "index": 0, "vector": [0.0123, -0.0456, 0.0789, "..."], "dimensions": 1024 }
  ],
  "model": "mistral-embed"
}

OCR / Read Document​

Used to extract text and structured information from PDF, image or document content.

Parameters​

Example Workflow​

Text Input containing a document URL → Mistral AI / OCR - Read Document → Text Output

The document URL from Text Input is inserted into the Document / Image URL field with /. The result is displayed in Text Output as the document content.

Sample Output​

{
  "text": "INVOICE\nDate: 01.08.2026\nTotal Amount: 12,500 TRY\n...",
  "pages": 1,
  "model": "mistral-ocr-latest"
}

Code Completion (FIM)​

FIM stands for "fill in the middle". It is used to complete the missing part of a code snippet.

Parameters​

Prompt (Prefix) Example​

def calculate_total(products):

Example Workflow​

Text Input → Mistral AI / Code Completion (FIM) → Text Output

The code beginning from Text Input is linked to the Prefix field. The model completes the code; the result is displayed in Text Output.

Sample Output​

{
  "completion": "\n    total = 0\n    for product in products:\n        total += product['price'] * product['quantity']\n    return total",
  "model": "codestral-latest"
}

Content Moderation​

Used to evaluate text for policy, safety or unwanted content.

Parameters​

Example Workflow​

Text Input → Mistral AI / Content Moderation → Text Output

The output contains the moderation result for the text. This result can also be used in the next step together with an If-Else Condition for conditional routing (e.g. reject the message if inappropriate content is detected).

Sample Output​

{
  "flagged": false,
  "categories": {
    "hate": false,
    "violence": false,
    "self_harm": false,
    "sexual": false
  }
}

Common Errors​

Notes​

• Your Mistral AI account's usage quota and billing are managed through the Mistral platform; SkyStudio only brokers the API calls.
• In the OCR / Read Document operation, processing time may increase for large PDF files; if possible, send files containing only the relevant pages.
• The Code Completion (FIM) operation is suitable for automations that mimic code editor integrations; filling in the Suffix field improves result quality.

DeepSeek​

DeepSeek is a model provider known for its cost-performance balance, used especially in chat and code completion scenarios.

Connection Setup and Management​

Getting an API Key​

1. Sign in to your DeepSeek Platform account.
2. Go to the API Keys section from the left menu.
3. Click Create new API key.
4. Copy the generated key and store it in a safe place.

Note: Official API documentation: DeepSeek API.

Adding the Connection to SkyStudio​

1. Go to App Connections from the SkyStudio left menu.
2. Click Add New Connection.
3. Select DeepSeek as the provider.
4. Fill in the following fields:

5. Click Save.

Assigning the Connection to the Node​

After adding the DeepSeek node to the canvas, select your saved connection from the App Connections dropdown in the right panel.

Operations​

The DeepSeek node offers 2 operations:

Chat​

Parameters​

Example Workflow​

Text Input → DeepSeek / Chat → Text Output

The question in Text Input is linked to the DeepSeek Message field. The output is shown in the Text Output node.

Sample Output​

{
  "response": "ARIMA and Prophet are the most commonly used methods for time series forecasting.",
  "model": "deepseek-chat",
  "usage": { "prompt_tokens": 15, "completion_tokens": 22 }
}

Code Completion (FIM)​

Parameters​

Example Workflow​

Text Input → DeepSeek / Code Completion (FIM) → Text Output

Sample Output​

{
  "completion": "\n    return sum(p['price'] * p['quantity'] for p in products)",
  "model": "deepseek-coder"
}

Common Errors​

Notes​

• DeepSeek models are a good option for cost-sensitive, high-volume workflows.
• The Code Completion (FIM) operation gives the best results with models from the deepseek-coder family.

Groq​

Groq is a platform that provides fast model inference. It is preferred for low-latency chat, support bots and fast text processing flows.

Connection Setup and Management​

Getting an API Key​

1. Sign in to your Groq Console account.
2. Select the relevant project.
3. In the API Keys area, click Create API Key.
4. Copy the generated key.

Note: Groq keys are project-based; you need to create a separate key for each project.

Adding the Connection to SkyStudio​

1. Go to App Connections from the SkyStudio left menu.
2. Click Add New Connection.
3. Select Groq as the provider.
4. Fill in the following fields:

5. Click Save.

Assigning the Connection to the Node​

After adding the Groq node to the canvas, select your saved connection from the App Connections dropdown in the right panel.

Operations​

The Groq node offers a single operation:

Chat​

Parameters​

Example Workflow​

Text Input → Groq / Chat → Text Output

For example, a support request is typed into Text Input, and the Groq model is asked for a short, professional reply. After the workflow runs, Text Output is checked.

Sample Output​

{
  "response": "Your request has been received; we will get back to you within 24 hours at the latest.",
  "model": "llama-3.3-70b-versatile",
  "latency_ms": 180
}

Common Errors​

Notes​

• Groq offers noticeably lower response times than other providers for the same model family; it can be preferred for real-time chatbots.
• Some models may be decommissioned after a certain period; it is recommended to periodically compare your workflows with Groq's current model list.

xAI (Grok)​

xAI provides the Grok models. It can be used for chat and image generation.

Connection Setup and Management​

Getting an API Key​

1. Sign in to your xAI Console account.
2. Go to the API Keys section.
3. Create a new key and copy it.

Adding the Connection to SkyStudio​

1. Go to App Connections from the SkyStudio left menu.
2. Click Add New Connection.
3. Select xAI (Grok) as the provider.
4. Fill in the following fields:

5. Click Save.

Assigning the Connection to the Node​

After adding the xAI (Grok) node to the canvas, select your saved connection from the App Connections dropdown in the right panel.

Operations​

The xAI (Grok) node offers 2 operations:

Chat​

Parameters​

Example Workflow​

Text Input → xAI (Grok) / Chat → Text Output

Sample Output​

{
  "response": "Three key findings stand out in your competitor analysis report: price, delivery time and customer support.",
  "model": "grok-4"
}

Generate Image​

Used to generate an image from a text description.

Parameters​

Prompt Example​

A team working in a modern technology office, natural light, corporate and realistic photo style

Example Workflow​

Text Input → xAI (Grok) / Generate Image → Output or next node that accepts images

The Prompt is taken from Text Input. After the workflow runs, check that the image generation result appears in the relevant output node or in the node result.

Sample Output​

{
  "images": [
    { "url": "https://cdn.x.ai/generated/img_8f21ac.png", "index": 0 }
  ],
  "model": "grok-image"
}

Common Errors​

Notes​

• In the Generate Image operation, the more detailed the Prompt (subject, style, lighting, composition, unwanted elements), the closer the generated image will be to the request.
• Keeping Number of Images high increases processing time and cost; start with 1 in your first attempts.

Together AI​

Together AI makes various open-source models available for chat, embedding and image generation.

Connection Setup and Management​

Getting an API Key​

1. Sign in to your Together AI account.
2. Go to the API Keys section under Settings.
3. Create a new key and copy it.

Adding the Connection to SkyStudio​

1. Go to App Connections from the SkyStudio left menu.
2. Click Add New Connection.
3. Select Together AI as the provider.
4. Fill in the following fields:

5. Click Save.

Assigning the Connection to the Node​

After adding the Together AI node to the canvas, select your saved connection from the App Connections dropdown in the right panel.

Operations​

The Together AI node offers 3 operations:

Chat​

Parameters​

Example Workflow​

Text Input → Together AI / Chat → Text Output

Sample Output​

{
  "response": "Open-source models provide flexibility in terms of customization and cost control.",
  "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo"
}

Generate Embedding​

Parameters​

Example Workflow​

Text Input → Together AI / Generate Embedding → Vector search node

Sample Output​

{
  "embeddings": [
    { "index": 0, "vector": [0.0211, -0.0187, "..."], "dimensions": 768 }
  ],
  "model": "togethercomputer/m2-bert-80M-8k-retrieval"
}

Generate Image (FLUX)​

Parameters​

Example Workflow​

Text Input → Together AI / Generate Image (FLUX) → Image output node

Sample Output​

{
  "images": [
    { "url": "https://together.ai/generated/flux_3a91.png", "width": 1024, "height": 1024 }
  ],
  "model": "black-forest-labs/FLUX.1-schnell"
}

Common Errors​

Notes​

• Together AI offers many open-source models through a single connection; choose a model by comparing them against your business needs (speed, cost, quality).
• As the Steps value increases, image quality usually improves, but processing time and cost also increase; a low Steps value is enough for quick drafts.

Perplexity​

Perplexity is used in web search–powered chat scenarios. It is preferred for research, source scanning and summarization tasks that require up-to-date information.

Connection Setup and Management​

Getting an API Key​

1. Sign in to your Perplexity account.
2. Go to the API section under Settings.
3. Create a new key and copy it.

Adding the Connection to SkyStudio​

1. Go to App Connections from the SkyStudio left menu.
2. Click Add New Connection.
3. Select Perplexity as the provider.
4. Fill in the following fields:

5. Click Save.

Assigning the Connection to the Node​

After adding the Perplexity node to the canvas, select your saved connection from the App Connections dropdown in the right panel.

Operations​

The Perplexity node offers a single operation:

Chat (with Web Search)​

Parameters​

Example Workflow​

Text Input → Perplexity / Chat (with Web Search) → Text Output

A current research question is typed into Text Input. The model runs and the response is displayed in Text Output. Since up-to-date information is involved, it is recommended to check the sources in the result separately.

Sample Output​

{
  "response": "As of 2026, the three biggest players in the industry are ... (see sources)",
  "citations": [
    "https://example-source1.com/report",
    "https://example-source2.com/analysis"
  ],
  "model": "sonar-pro"
}

Common Errors​

Notes​

• Since Perplexity responses are based on web sources, it is recommended to verify the returned sources (citations) separately for critical decisions.
• It gives more reliable results than other providers for time-sensitive questions such as current events, prices and regulations.

Cohere​

Cohere is used for chat, embedding and document reranking.

Connection Setup and Management​

Getting an API Key​

1. Sign in to your Cohere Dashboard account.
2. Go to the API Keys section.
3. Create a new key and copy it.

Adding the Connection to SkyStudio​

1. Go to App Connections from the SkyStudio left menu.
2. Click Add New Connection.
3. Select Cohere as the provider.
4. Fill in the following fields:

5. Click Save.

Assigning the Connection to the Node​

After adding the Cohere node to the canvas, select your saved connection from the App Connections dropdown in the right panel.

Operations​

The Cohere node offers 3 operations:

Chat​

Parameters​

Example Workflow​

Text Input → Cohere / Chat → Text Output

Sample Output​

{
  "response": "When rerank and embedding are used together, search relevance increases significantly.",
  "model": "command-r-plus"
}

Generate Embedding​

Parameters​

Example Workflow​

Text Input → Cohere / Generate Embedding → Vector search node

Sample Output​

{
  "embeddings": [
    { "index": 0, "vector": [0.0342, -0.0119, "..."], "dimensions": 1024 }
  ],
  "model": "embed-multilingual-v3.0"
}

Rerank Documents​

Used to select and rank the documents most relevant to a search query.

Parameters​

Query Example​

How many days is the return period?

Documents (JSON) Example​

["The return period is 14 days from the delivery date.", "Shipping takes 2-4 business days."]

Example Workflow​

Text Input → Cohere / Rerank Documents → Text Output

Text Input is linked to the Query field. The list of documents to evaluate is entered in the Documents field. After the run, the most relevant documents are displayed in the result.

Sample Output​

{
  "results": [
    { "index": 0, "document": "The return period is 14 days from the delivery date.", "relevance_score": 0.94 },
    { "index": 1, "document": "Shipping takes 2-4 business days.", "relevance_score": 0.11 }
  ]
}

Common Errors​

Notes​

• Adding the Rerank Documents operation as a layer on top of classic keyword search significantly improves the relevance of search results.
• In the Generate Embedding operation, choosing the correct Input Type (query or document) directly affects result quality.

Gemini​

Gemini is added to use Google's text and image generation models.

Connection Setup and Management​

Getting an API Key​

1. Sign in to your account via Google AI Studio.
2. Go to the Get API key section.
3. Create a new key and copy it.

Note: Official guide: Creating a Gemini API key.

Adding the Connection to SkyStudio​

1. Go to App Connections from the SkyStudio left menu.
2. Click Add New Connection.
3. Select Gemini as the provider.
4. Fill in the following fields:

5. Click Save.

Assigning the Connection to the Node​

After adding the Gemini node to the canvas, select your saved connection from the App Connections dropdown in the right panel.

Operations​

The Gemini node offers 2 operations:

Chat​

Parameters​

Example Workflow​

Text Input → Gemini / Chat → Text Output

Sample Output​

{
  "response": "Recommended structure for the presentation: Introduction, Problem, Solution, Evidence, Closing.",
  "model": "gemini-2.5-pro"
}

Generate Image​

Parameters​

Aspect Ratio Guide​

Example Workflow​

Text Input → Gemini / Generate Image → Image output node

Sample Output​

{
  "images": [
    { "url": "https://generativelanguage.googleapis.com/generated/img_c391.png", "aspect_ratio": "1:1" }
  ],
  "model": "gemini-2.5-flash-image"
}

Common Errors​

Notes​

• Gemini API keys are tied to your Google account; for team use, creating a separate service account is recommended.
• Aspect Ratio should be planned in advance according to where the output will be used (social media, banner, presentation); cropping afterwards may reduce image quality.

OpenRouter​

OpenRouter lets you use models from different providers through a single connection.

Connection Setup and Management​

Getting an API Key​

1. Go to the OpenRouter Keys page.
2. Create a new key and copy it.

Adding the Connection to SkyStudio​

1. Go to App Connections from the SkyStudio left menu.
2. Click Add New Connection.
3. Select OpenRouter as the provider.
4. Fill in the following fields:

5. Click Save.

Assigning the Connection to the Node​

After adding the OpenRouter node to the canvas, select your saved connection from the App Connections dropdown in the right panel.

Operations​

The OpenRouter node offers a single operation:

Chat​

Parameters​

Model Examples​

anthropic/claude-sonnet-4

or

deepseek/deepseek-chat

Note: The model name must be written in its current OpenRouter format (including the provider prefix); otherwise you will get a Model not found error.

Example Workflow​

Text Input → OpenRouter / Chat → Text Output

Sample Output​

{
  "response": "For the comparative analysis, I suggest three scenarios: ...",
  "model": "anthropic/claude-sonnet-4",
  "provider": "openrouter"
}

Common Errors​

Notes​

• Since OpenRouter gives access to models from dozens of providers with a single connection and API key, it is a practical option for model comparison and A/B testing scenarios.
• Some models can only be used via OpenRouter with a specific payment method or quota; if you get an error, check your OpenRouter account balance first.

Hugging Face​

Hugging Face is preferred for using open-source models.

Connection Setup and Management​

Getting an API Key (Access Token)​

1. Go to the Hugging Face Access Tokens page.
2. Click New token.
3. Give the token a name and select the required permission scope (at least read/inference).
4. Copy the generated token.

Adding the Connection to SkyStudio​

1. Go to App Connections from the SkyStudio left menu.
2. Click Add New Connection.
3. Select Hugging Face as the provider.
4. Fill in the following fields:

5. Click Save.

Assigning the Connection to the Node​

After adding the Hugging Face node to the canvas, select your saved connection from the App Connections dropdown in the right panel.

Operations​

The Hugging Face node offers a single operation:

Chat​

Parameters​

Model Example​

meta-llama/Llama-3.3-70B-Instruct

Note: The model to be used must be accessible on Hugging Face and compatible with the relevant connection type (Inference API / Inference Endpoints). Some models may additionally require license approval before use.

Example Workflow​

Text Input → Hugging Face / Chat → Text Output

Sample Output​

{
  "response": "When comparing models, accuracy, latency and cost should be evaluated together.",
  "model": "meta-llama/Llama-3.3-70B-Instruct"
}

Common Errors​

Notes​

• On the first call, the model may respond more slowly than usual due to a cold start; this can also appear as a 503 Model loading error.
• Some large models may only be available on paid Inference Endpoints plans; check the free Inference API coverage on Hugging Face.

BYOM — Bringing Your Own Model​

BYOM (Bring Your Own Model) lets you use your own model endpoint or your custom model deployment on your cloud provider in a SkyStudio workflow. The connection fields in this section vary depending on the infrastructure where the model is hosted (custom endpoint, Azure resource deployment or AWS Bedrock region).

Custom Model​

Custom Model is used to connect to OpenAI-compatible or internally hosted custom model endpoints.

Connection Setup and Management​

Adding the Connection to SkyStudio​

1. Go to App Connections from the SkyStudio left menu.
2. Click Add New Connection.
3. Select Custom Model as the provider.
4. Fill in the following fields:

Note: This information is obtained from the system hosting the model (OpenAI-compatible server, internal model server, etc.).

5. Click Save.

Assigning the Connection to the Node​

After adding the Custom Model node to the canvas, select your saved connection from the App Connections dropdown in the right panel.

Operations​

The Custom Model node offers a single operation:

Chat​

Parameters​

Model Example​

company-assistant-v1

Example Workflow​

Text Input → Custom Model / Chat → Text Output

Sample Output​

{
  "response": "Your request has been answered based on the internal knowledge base.",
  "model": "company-assistant-v1"
}

Common Errors​

Notes​

• Custom Model is designed to work with any endpoint compatible with the OpenAI Chat Completions API (internal LLM server, self-hosted open-source model server, etc.).
• If the endpoint is hosted on a non-public network (VPN, internal network), make sure SkyStudio can reach it.

Azure OpenAI​

Azure OpenAI is added to use model deployments created on Azure.

Connection Setup and Management​

Adding the Connection to SkyStudio​

1. Go to App Connections from the SkyStudio left menu.
2. Click Add New Connection.
3. Select Azure OpenAI as the provider.
4. Fill in the following fields:

5. Click Save.

Note: This information is obtained from the Keys and Endpoint page of the relevant Azure OpenAI resource in Azure Portal.

Assigning the Connection to the Node​

After adding the Azure OpenAI node to the canvas, select your saved connection from the App Connections dropdown in the right panel.

Operations​

The Azure OpenAI node offers 2 operations:

Chat​

Parameters​

Deployment Name Example​

production-chat-model

Example Workflow​

Text Input → Azure OpenAI / Chat → Text Output

Sample Output​

{
  "response": "Per corporate policy, the request should first be routed to the relevant department.",
  "deployment": "production-chat-model"
}

Generate Embedding​

Parameters​

Example Workflow​

Text Input → Azure OpenAI / Generate Embedding → Vector search node

Sample Output​

{
  "embeddings": [
    { "index": 0, "vector": [0.0156, -0.0298, "..."], "dimensions": 1536 }
  ],
  "deployment": "text-embedding-ada-002"
}

Common Errors​

Notes​

• In Azure OpenAI, Deployment Name is used instead of the model name; this lets you manage multiple deployments of the same model under different names (e.g. test/production).
• For projects with corporate data retention and region requirements, Azure OpenAI keeps data in the region where your Azure resource is located.

AWS Bedrock​

AWS Bedrock is added to use model families offered through AWS (Anthropic, Meta, Amazon and others).

Connection Setup and Management​

Adding the Connection to SkyStudio​

1. Go to App Connections from the SkyStudio left menu.
2. Click Add New Connection.
3. Select AWS Bedrock as the provider.
4. Fill in the following fields:

5. Click Save.

Note: The IAM user or role used must have the bedrock:InvokeModel permission. The connection details depend on your AWS account, the selected region and the relevant access permissions.

Assigning the Connection to the Node​

After adding the AWS Bedrock node to the canvas, select your saved connection from the App Connections dropdown in the right panel.

Operations​

The AWS Bedrock node offers a single operation:

Chat​

Parameters​

Model ID Example​

anthropic.claude-3-5-sonnet-20241022-v2:0

Note: The Model ID must be accessible in the selected AWS region; not every model is offered in every region.

Example Workflow​

Text Input → AWS Bedrock / Chat → Text Output

Sample Output​

{
  "response": "Three recommended steps for cost optimization: usage monitoring, auto scaling and reserved capacity.",
  "model_id": "anthropic.claude-3-5-sonnet-20241022-v2:0"
}

Common Errors​

Notes​

• In AWS Bedrock, some models must be enabled through a model access request in the AWS console before use; otherwise, the call is rejected even if the model exists.
• For production workflows requiring high-volume, predictable load, using provisioned throughput reduces the risk of ThrottlingException.

Quick Start Examples​

Example 1: Meeting Note Summarization​

Text Input → OpenAI API Call → Text Output

OpenAI API Call settings:

• Model: GPT-5.4 Mini
• System Prompt: Summarize the meeting notes. Give decisions and action items under separate headings.
• Query: Text Input output via /

Example 2: Support Request Classification and Routing​

Webhook → Anthropic API Call → If-Else Condition ─┬─ Create Issue (Jira)
                                                  └─ Send Mail (Outlook)

Anthropic API Call settings:

• System Prompt: Classify the incoming request as "technical" or "general". Return only one of these words.
• Query: Webhook body via /

Example 3: Report with Up-to-Date Information​

Text Input → Claude Web Search → Artifact → Send Mail (Outlook)

Example 4: Multi-Step Analysis​

A flow that first summarizes a long document and then interprets it:

File Input → Read Excel → Anthropic API Call (summarize) → OpenAI API Call (generate recommendations) → Text Output

Provider Selection Guide​

A general summary of which node may be better suited to which need:

Running the Workflow and Using the Result​

The working logic is the same for every provider example:

1. Type the text to be tested into the Text Input node.
2. In the LLM node, check that the correct App Connection, Connection and model are selected.
3. Link the Text Input value to the Message, Prompt, Input Text(s) or relevant operation field with /.
4. Make sure the incoming connection is Text Input and the next connection is Text Output (or the relevant next node).
5. Save the workflow.
6. Start the run action on the workflow screen.
7. First check the LLM node's result, then the final response in Text Output.
8. If an error occurs, first check the API key in the connection, the selected model, model access permissions and the required fields.

The output of chat nodes can be shown directly to the user. The output of embedding, rerank, moderation and image generation nodes, however, should be connected to next nodes suited to its purpose (e.g. embedding output to a vector search node, rerank output to a ranked list node).

Common Errors​

The following errors can occur in all LLM nodes. Provider-specific errors are listed in each provider's own section.

Notes​

• Built-in API nodes and GOAT do not require a credential connection; BYOK and BYOM nodes connect to your own account/infrastructure, and usage is billed according to that account's plan.
• The Query (or Message) field defines what to do; the System Prompt defines how to do it. Using both together gives the best results.
• In classification and decision-making steps, write a clear format rule in the System Prompt so that the next If-Else node can read the output correctly.
• Small models (Mini, Nano) are both faster and more economical for simple tasks; do not choose the largest model for every task.
• The Tools feature slows the model down and increases cost; only add the tools you need.
• Claude Web Search should only be used when up-to-date information is needed; otherwise, a normal LLM node is faster.
• Enter API keys only in the connection field; never write them in the Message, System Prompt or Query fields.
• Most errors in BYOK and BYOM nodes stem from connection details (API Key, Endpoint, Deployment Name); when troubleshooting, check the connection first, then the model/parameter fields.
• In workflows moved to production, explicitly setting Temperature and Max Tokens improves consistency across runs.
• You can chain multiple LLM nodes in a flow — for example, to summarize first and then interpret.
• To avoid depending on a single provider in critical processes, design the same workflow to run redundantly with an alternative node (e.g. GOAT, or a different model via OpenRouter).
• Give nodes meaningful names (such as "Summarize", "Classify"); this makes it easier to find the right output in the / menu.