Skip to main content
Tutorials

Build a New Project

Create a complete AI system from scratch

Overview

In this tutorial, you'll build a News Aggregator Project that collects articles from multiple sources, summarizes them, and outputs a daily digest. You'll learn to:

  • Create a Project from scratch
  • Build Actions for data collection
  • Design Flows for orchestration
  • Configure an Agent for summarization
  • Test and deploy your system

Time required: 20-30 minutes

Step 1: Create the Project

  1. Click the Home icon in the Top Bar
  2. Click New Project
  3. Name it News Aggregator
  4. Add description: Collects and summarizes news from multiple sources
  5. Click Create

Your Canvas in the Builder workspace now shows an empty Project node.

Step 2: Build the fetch Action

Let's create an Action to fetch news from an API:

  1. Open the Chat Panel (Triton) at the bottom

  2. Say: "Create an Action called fetch_news that takes a source URL and returns a list of articles"

  3. Triton will generate the Action with:

    • Input: source_url (string)
    • Output: articles (list of dicts)
    • Basic HTTP request logic
  4. Review the generated Action.py in the Properties Panel

  5. Adjust if needed (add error handling, rate limiting, etc.)

Step 3: Create the aggregation Flow

Now build a Flow to call the Action for multiple sources:

  1. Ask Triton: "Create a Flow that takes a list of news sources and calls fetch_news for each one"

  2. Triton will create:

    • Input node with sources parameter
    • Loop or parallel execution of fetch_news
    • Output node with combined articles
  3. Double-click the Flow node to inspect the structure

  4. Verify the connections are correct

Step 4: Add a summarization Agent

Create an Agent to summarize the collected articles:

  1. Say: "Create an Agent called news_summarizer that takes articles and creates a concise daily digest"

  2. Triton will configure:

    • System prompt for summarization
    • Input/output schema
    • Model selection (GPT-4 or Claude recommended)
  3. Review and refine the system prompt:

    You are a news summarization expert. Given a list of news articles,create a concise, informative daily digest. Group by topic, highlightkey trends, and maintain neutral tone. Output as structured markdown.

Step 5: Connect the components

Wire everything together:

  1. Create a main Flow called daily_news_digest

  2. Structure it as:

    Input (sources list)  → fetch_news_flow (aggregation)  → news_summarizer (Agent)  → Output (formatted digest)
  3. You can do this via Triton or manually:

    • Drag components from the Agent Toolbox
    • Connect them with edges (click and drag from output to input ports)
    • Configure each node's parameters

Step 6: Test with sample data

Execute your system:

  1. Select the daily_news_digest Flow

  2. Open Properties → Execute

  3. Enter test payload:

    {  "sources": [    "https://newsapi.org/v2/top-headlines?country=us",    "https://api.example.com/tech-news",    "https://api.example.com/world-news"  ]}
  4. Click Execute

  5. Monitor execution in real-time

  6. Review the output digest

Step 7: Debug and iterate

If something goes wrong:

  • Check execution logs in Properties → Executions
  • Verify each component's input/output schema
  • Test Actions individually before running the full Flow
  • Ask Triton: "Why did the execution fail?"

Common issues:

  • API authentication missing → Add API keys to Project Variables
  • Timeout errors → Adjust Action timeout settings
  • Empty results → Verify API endpoints are accessible

Step 8: Deploy as API

Make your Project accessible:

  1. Click Deploy in the Top Bar

  2. Select deployment target (e.g., production)

  3. Configure:

    • Endpoint path: /daily-digest
    • HTTP method: POST
    • Authentication: API key required
  4. Click Deploy Now

  5. Copy the generated endpoint URL

  6. Test with curl:

This is an API-call that would trigger a certain flow with the {"sources": ["..."]} payload:

curl -X POST 'https://app.triform.ai/api/in/31460071-17aa-43e8-a0e6-fb22984c0bdc/26420f15-2143-443f-bca9-45c5b0720784' \  --header 'Authorization: 88ffe8398f972837ccc6e0a6cd31c5c92689040c' \  --header 'Content-Type: application/json' \  --data '{"sources": ["..."]}'

Step 9: Schedule daily execution

Set up automatic runs:

  1. In the Project Properties, go to Triggers

  2. Click Add Trigger

  3. Select Schedule

  4. Configure:

    • Schedule: Daily at 6:00 AM UTC
    • Payload: Your default sources list
    • Notification: Email on success/failure
  5. Save the trigger

Next steps

Now that you've built a complete Project:

  • Enhance it — Add sentiment analysis, language detection, custom formatting
  • Monitor it — Set up alerts for failures or anomalies
  • Share it — Invite team members to collaborate
  • Extend it — Create variations for different news categories

Tips for success

Start simple — Build and test each component individually before connecting them

Use Project Variables — Store API keys, base URLs, and configuration centrally

Save test payloads — Keep example inputs for regression testing

Version your work — Use meaningful names and descriptions for components

Iterate with Triton — Don't hesitate to ask for modifications or explanations