Importable n8n workflows for AI agents and prompt systems, with the description and a diagram of each one.
| Workflow | File | What it does |
|---|---|---|
| Multi AI Agent (Telegram AI Bot) | Multi Ai Agent.json |
One Telegram bot that routes each message by keyword to a different AI persona and model, or to all three at once for a combined answer |
What it does. A Telegram Trigger receives every message. A Switch node lowercases the text and checks, in order, whether it contains force, nexus, victor or team. The first match picks the route; anything else goes to the fallback output (named Sizu), which is handled by the default agent.
| Route | Trigger word | Agent(s) | Model |
|---|---|---|---|
| Force | force |
Ai Force: creative, bold, direct | Google Gemini node, models/gemma-4-26b-a4b-it |
| Nexus | nexus |
Ai Nexus: analytical, structured | OpenRouter, inclusionai/ring-2.6-1t:free |
| Victor | victor |
Ai Victor: fast and concise | Groq, llama-3.3-70b-versatile |
| Team | team |
Team Force, Team Nexus and Ai Victor1 run in parallel as a three-member "council" (2-4 line answers, each prefixed with its name); a Merge node (3 inputs) and a Code node combine them into one message | Gemini models/gemini-2.5-flash-lite, OpenRouter inclusionai/ring-2.6-1t:free, Groq llama-3.3-70b-versatile |
Fallback (Sizu) |
anything else | Ai Default (persona "Sizu"), which can search and read Gmail through a tool | Groq, llama-3.3-70b-versatile |
All agents replying to the same Telegram chat share one Simple Memory node (window of 15 messages, keyed by the Telegram chat id). Every route ends in a single Send a text message Telegram node that replies to the chat the message came from. The agents answer in the language of the user (Bengali, English or mixed).
Gmail tool. Get many messages in Gmail is attached to Ai Default as a tool. The agent fills in limit and a Gmail search query itself (for example is:unread or newer_than:1d), so "check my email" or "summarise my last 5 emails" works from Telegram.
Nodes used. Telegram Trigger, Switch, AI Agent (x7), Google Gemini Chat Model (x2), OpenRouter Chat Model (x2), Groq Chat Model (x3), Simple Memory, Gmail Tool, Merge, Code, Telegram.
Credentials needed (create them in n8n, then select them on each node after import):
| Credential type | Used by |
|---|---|
| Telegram API (bot token from @BotFather) | Telegram Trigger, Send a text message |
| Google Gemini (PaLM) API key | Both Google Gemini Chat Model nodes |
| OpenRouter API key | Both OpenRouter Chat Model nodes |
| Groq API key | All three Groq Chat Model nodes |
| Gmail OAuth2 | Get many messages in Gmail |
The credential names stored in the file (for example "Groq account") belong to the original instance; n8n asks you to pick your own when you open each node. No keys or tokens are stored in the JSON.
Setup.
- In n8n choose Workflows -> Import from file and select
Multi Ai Agent.json. - Create the five credentials above and attach them to the nodes that show a warning.
- Check the model names. The free OpenRouter and Gemini/Gemma model ids in the file can be retired by the providers, so swap them for any model you have access to.
- Edit the trigger words in the Switch node if you want different routes.
- Activate the workflow and message your bot, for example
force give me three headline ideas,team what should I post this week, or any plain message for the default agent.
Longer notes on the design are in Multi Ai Agent Descriptions and Multi Ai Agent Feature.
scripts/validate-workflows.js checks every workflow JSON in this repo (valid JSON, well-formed nodes, connections to existing nodes, no orphaned nodes, no embedded secrets). It runs in GitHub Actions on every push and pull request and locally with:
node scripts/validate-workflows.jsSaizul Amin: web developer, SEO and automation specialist.