Universal AI Agent

universal ai agent node overview the universal ai agent node is a powerful, flexible ai block that lets you add ai intelligence to your workflows think of it as hiring a smart assistant that can work with multiple ai providers (openai, anthropic, deepseek, google gemini, etc ) use external tools and integrations (send emails, read files, query databases, etc ) remember previous conversations understand and analyze images connect to custom business systems in simple terms one node, infinite ai capabilities 1\ providers tab choose your ai brain this is where you select which ai service and model will power your agent what you see provider cards showing openai, anthropic, deepseek, google gemini, x ai (grok), local/ollama each card has a "working" badge (tells you the connection is active) model list showing popular models + "+x more" to see additional options ai model dropdown to choose the exact model available providers & models provider top models total models openai gpt 5 1, gpt 5, gpt 5 mini 20+ models anthropic claude opus 4 5, claude sonnet 4 5 17+ models deepseek deepseek v3 1, deepseek v3 1 thinking 5+ models google gemini gemini 3 pro, gemini 3 flash 10+ models x ai (grok) grok beta, grok vision beta 3+ models local/ollama llama 2, llama 2 13b, llama 2 70b 30+ models key configuration fields authentication (required) use stored credential pick a saved api key from your credential manager or enter api key type your api key directly (shows red error if missing) 🔧 configuration agent name default is "universal ai agent" rename to describe what it does system prompt tell the ai how to behave example "you are a friendly customer service agent who explains things simply" user input (required) where the question/data comes from usually {{dataflow\ current()}} (pulls from previous node) output format choose how the response should look text plain text reply json structured data (good for parsing results) structured advanced formatting 2\ tools tab give your agent superpowers tools are like apps your agent can use enable only what you need how to use toggle "enable tools" on check the tools your agent should have access to shows "selected x tools" available tools by category communication email sender send emails via smtp slack message send slack messages teams message send microsoft teams messages google services google sheets read/write/manipulate spreadsheets google docs create and edit documents google drive upload, download, manage files google calendar create and manage events research web search search the internet 🔧 utilities calculator perform math calculations file reader read and analyze files data formatter transform and format data development code interpreter execute and analyze code api caller make http api calls ai tools image generator generate images from text text to speech convert text to speech speech to text convert speech to text database database query query sql databases supabase interact with supabase databases business pdf generator generate pdf documents qr generator generate qr codes ai & ml structured output parser convert messy ai responses into clean json beginner example if you want your agent to send emails and search the web, check email sender web search that's it! 3\ memory tab let your agent remember memory lets your agent recall previous conversations and context how to enable toggle "enable memory" on choose a memory type memory types explained basic (simple options) type what it does use case no memory agent forgets everything each time one off questions simple memory keeps conversation history in current session chatbots, customer support buffer window keeps only last n messages long conversations (saves tokens) advanced (powerful options) type what it does use case conversation summary ai summarizes old messages automatically long term memory with cost savings database storage saves memory to supabase database multi user apps, long term retention redis session memory fast in memory storage high speed real time apps enterprise (maximum power) type what it does long term semantic memory ai powered summarization + vector embeddings + permanent retention how memory works behind the scenes when you enable memory, the system tracks conversation data using these fields field level content example purpose agentid top level unique agent identifier support bot prod know which bot this is userid top level unique customer identifier cust 789 separate each customer's data sessionid top level unique chat session id chat 2025 12 13 14 30 organize each conversation contextid top level topic/ticket grouping order 12345 group related messages timestamp inside messages when message was sent 2025 12 13t14 30 15z track conversation flow role inside messages who sent the message user or assistant know who said what content inside messages the actual message text my order hasn't arrived the conversation text example customer chat customer "where's my order?"agent "what's your order number?" customer "12345" agent "found it! status shipped" memory stores agentid support bot produserid cust 789 sessionid chat 2025 12 13 14 30 contextid order 12345 messages \[all above messages] next time customer returns customer "when will it arrive?"agent (reads memory) "order #12345 is being delivered today" \[without memory "hi! how can i help?"] result agent remembers context ✓ 4\ vision tab let your agent see images enable vision so your agent can analyze, understand, and answer questions about images how to enable toggle "enable vision" on set image detail level (how deep to analyze) low faster, cheaper high more detailed analysis auto model decides automatically how to provide images choose one method 1\ using a variable {{dataflow\ current()}} (pulls image url or path from previous node) 2\ using a direct url https //example com/image jpg 3\ using base64 data\ image/png;base64, 4\ upload a file click "click to upload" and select png, jpg, gif, or webp (max 10mb) add multiple images use "additional images (optional)" field to add more images for batch analysis vision capabilities analyze images with ai ask questions about images works with gpt 4 vision, claude 3, gemini vision supports urls, base64, and uploaded files example user uploads a screenshot of an error vision enabled image detail auto agent analyzes screenshot and explains the error 5\ mcp tab connect custom business systems mcp = model context protocol lets your agent talk to private/custom tools how to use click "+ add server" button configure the mcp server with id unique identifier name human readable name description what this server does endpoint server url icon visual indicator what mcp can do connect to internal crm systems query custom databases use proprietary apis integrate with company specific tools note this is advanced start with the built in tools tab first use mcp only when you need private/custom integrations 6\ advanced tab fine tune the agent adjust how your agent behaves, handles errors, and responds temperature (0 2 0) controls creativity 0 = robotic, focused, consistent (for factual tasks) 0 7 = balanced (default, good for most uses) 1 5 2 0 = creative, random, unpredictable choose based on task data extraction? → set to 0 3 customer support? → set to 0 7 creative writing? → set to 1 5 max tokens response length 1000 = short reply ( 250 words) 5000 = medium reply ( 1000 words) default 10000+ = long detailed response higher = more detailed but slower and more costly response mode complete default wait for full response streaming get response word by word (good for chat uis) error handling continue default keep going even if there's an error stop halt the workflow if error occurs retry try again automatically (good for flaky apis) custom api endpoint (optional) use a custom server instead of official api https //api custom endpoint com/v1 use stored credential toggle on to use a saved api key instead of entering it each time configuration status summary at the bottom, you see api key required ✓ tools 0 selected memory buffer window mcp 0 servers 7\ node inputs & outputs inputs the agent receives data from previous nodes through user input field the main question/prompt dynamic variables data from earlier steps in the workflow output access the agent's response using {{dataflow\ getbynodeid(nodeid)}} example pass the response to send email node email body {{dataflow\ getbynodeid("universal ai agent node 123")}}