🧪 Platform Testing Guide
Welcome to the Opensource AI Agent testing guide. Follow each task to explore every feature of the platform. Each task includes step-by-step instructions, expected outcomes, and tips.
✦ Easy tasks for beginners
✦ Medium — some configuration
✦ Advanced — full power
Before You Start
Get the platform ready in 60 seconds
0
Open the Platform & Choose AI Engine
⏱ 1 min
▾
The platform opens with a provider modal. You can use the free local Opensource AI 70B model or connect your own API key.
1
Open
http://localhost:5000/app in your browser2
The AI Provider modal appears — select Opensource AI Agent (free local model)
3
From the dropdown, choose Opensource AI Agent 70B — Default
4
Click Launch Platform → — the platform loads instantly
The modal auto-closes in 0.6 seconds when the AI engine is detected. If you see "Checking AI engine..." it means Ollama is starting up.
Expected Result
✓ Platform loads with a dark canvas showing Input → LLM → Output nodes
✓ Top-right pill shows "Opensource AI Agent"
✓ 6 tabs visible: Workflow, Agent, Workbench, AutoML, JupyterLab, API
Workflow Builder
Build visual automation pipelines with drag-and-drop
1.1
Build Your First AI Workflow
⏱ 3 min · Easy
▾
Create a simple workflow that takes a question, processes it with the AI engine, and outputs the answer.
1
Click the ⚡ Workflow tab
2
You should see 3 pre-connected nodes: Input 1 → LLM 2 → Output 3
3
Click Input 1 node → in right panel, change the text to:
What are the 3 most important trends in AI for 2026?4
Click ▶ Execute button in the toolbar
5
Watch the nodes highlight in sequence and logs appear in the right panel
Expected Result
✓ Nodes animate green one by one as they complete
✓ AI response appears in the logs panel on the right
✓ Output node shows a preview of the result
1.2
Add and Connect New Nodes
⏱ 5 min · Easy
▾
Learn to drag nodes from the palette and connect them using the port system.
1
Press
Ctrl+K to open the node search panel2
Type
transform → click Transform to add it to the canvas3
Drag the orange dot on the right of LLM node → release on the blue dot on the left of Transform node
4
Click Transform node → set code to:
return "SUMMARY: " + input.toUpperCase().substring(0,200)5
Connect Transform → Output, then click ▶ Execute
To delete a connection, click directly on the line between two nodes. To undo, press
Ctrl+Z.Expected Result
✓ All 4 nodes run in sequence
✓ Output shows "SUMMARY: " prefix with truncated text
1.3
Load a Template Workflow
⏱ 2 min · Easy
▾
Browse the templates library and load a pre-built workflow onto your canvas.
1
Click 📋 Templates in the toolbar
2
Browse the categories: AI & LLM, Data, Data Science, Integrations
3
Click "AI Research Agent" template → click Use Template →
4
The canvas loads with 4 connected nodes — click ▶ Execute
Expected Result
✓ Template loads instantly on canvas
✓ Workflow runs: Input → Think → AGI Agent → Output
✓ A research report appears in the output
1.4
Build the Daily Report Workflow
⏱ 10 min · Medium
▾
Build the example workflow: Schedule → Fetch data → AI analysis → Email + Slack.
1
Press
Ctrl+K → add a Schedule node. Set interval to daily and time to 08:002
Add an HTTP Request node. Set URL to
https://httpbin.org/json (test endpoint) and Method to GET3
Add an LLM node. Set system prompt to:
You are a business analyst. Summarise this data into a daily insights report with 3 key findings.4
Add an Output node. Connect: Schedule → HTTP → LLM → Output
5
Click ▶ Execute to test manually
For real Google Sheets and Slack, save your API keys in 🔑 Credentials first, then reference them in the HTTP headers.
Expected Result
✓ HTTP node fetches JSON from the test endpoint
✓ AI generates a formatted insights report
✓ Schedule trigger would run this automatically at 8am
AGI Agent
Autonomous reasoning agent with real-time sensors
2.1
Run Your First AGI Agent Task
⏱ 3 min · Easy
▾
Use the AGI Agent tab to run a multi-step reasoning task with live sensor data.
1
Click the 🤖 Agent tab in the platform
2
Enable sensors: toggle on Time and Weather
3
Type this task:
Based on the current time and weather, suggest 3 productive activities for today and explain why each is suitable right now.4
Set Reasoning Steps to
35
Click ▶ Run Agent and watch each step appear
Expected Result
✓ Agent shows 3 reasoning steps in sequence
✓ Response references actual current time
✓ Final answer includes 3 context-aware suggestions
2.2
Run a Research Agent Task
⏱ 5 min · Medium
▾
Test the agent's multi-step reasoning with a complex research task.
1
Set reasoning steps to
52
Enter task:
Research the current state of open-source AI models in 2026. Compare the top 5 models by capability, size, and use case. Provide a structured report with a recommendation.3
Enable all sensors for maximum context
4
Click ▶ Run Agent — watch each reasoning step
Expected Result
✓ 5 reasoning steps visible with progress
✓ Structured report with model comparisons
✓ Clear recommendation at the end
AI Engine 70B
Test the local Opensource AI model capabilities
3.1
Test AI Chat Completion
⏱ 2 min · Easy
▾
Test the AI engine directly via the API tab or workflow.
1
Click the 🔗 API tab in the platform
2
Find the POST /llm/chat endpoint and test it
3
Or run this in your terminal:
curl -X POST http://localhost:5000/llm/chat \
-H "Content-Type: application/json" \
-d '{"messages":[{"role":"user","content":"Explain quantum computing in 3 sentences"}],"model":"deepseek-r1:70b"}'
-H "Content-Type: application/json" \
-d '{"messages":[{"role":"user","content":"Explain quantum computing in 3 sentences"}],"model":"deepseek-r1:70b"}'
Expected Result
✓ Response in under 30 seconds
✓ Clear, accurate 3-sentence explanation
3.2
Test AI Code Assistant
⏱ 3 min · Easy
▾
Use the built-in AI code assistant in the Workbench tab.
1
Click the 🔬 Workbench tab
2
Click the ▲ button in the bottom-right to open the AI Assistant panel
3
Type:
Write a PySpark job that reads a CSV, calculates average salary by department, and saves results4
Click → Insert into Editor to paste the code
Expected Result
✓ AI generates complete PySpark code
✓ Code inserts directly into the editor
Integrations
Connect to external services and APIs
4.1
Test HTTP Integration — Fetch Live Data
⏱ 3 min · Easy
▾
Use the HTTP node to fetch live data from a public API and process it with AI.
1
Clear the canvas. Add Input → HTTP Request → LLM → Output
2
Click HTTP Request → set URL:
https://api.coindesk.com/v1/bpi/currentprice.json3
Click LLM → set prompt:
Analyse this Bitcoin price data and provide a brief market summary in 2 sentences.4
Click ▶ Execute
Expected Result
✓ HTTP node fetches live Bitcoin price data
✓ AI generates a market summary based on real data
4.2
Test Webhook Trigger
⏱ 5 min · Medium
▾
Set up a webhook trigger that receives external HTTP requests and processes them.
1
Add a Webhook node to the canvas. Note the webhook URL shown on the node.
2
Connect: Webhook → LLM → Output
3
Click ▶ Execute to start listening
4
In a terminal, send a test request to the webhook URL:
curl -X POST http://localhost:5000/webhook/YOUR_NODE_ID \
-H "Content-Type: application/json" \
-d '{"message":"Hello from webhook","data":"test payload"}'
-H "Content-Type: application/json" \
-d '{"message":"Hello from webhook","data":"test payload"}'
Expected Result
✓ Webhook node shows "Received" status
✓ Workflow executes automatically with the payload
4.3
Save Credentials for Slack / Email
⏱ 3 min · Easy
▾
Learn how to securely store API keys and tokens in the Credentials Manager.
1
Click 🔑 Credentials in the toolbar (or press
Ctrl+E)2
Add a test credential: Name:
My Slack Webhook, Type: Webhook URL, Value: https://hooks.slack.com/test3
Click + Save Credential
4
Confirm it appears in the Saved Credentials list with the value masked
Expected Result
✓ Credential saved and visible in the list
✓ Value is masked (shows sk-••••••••)
Data Science Workbench
PySpark, Hive SQL, Pig Latin, GPU Analytics
5.1
Run a PySpark Job
⏱ 5 min · Medium
▾
Run a distributed PySpark data processing job in your browser.
1
Click the 🔬 Workbench tab → select ⚡ PySpark
2
Click Basic DataFrame template to load example code
3
Click ▶ Run button
4
Or paste this custom code and run it:
from pyspark.sql import SparkSession
import pyspark.sql.functions as F
spark = SparkSession.builder.appName("Test").getOrCreate()
# Create sample data
data = [("Alice", "Engineering", 85000), ("Bob", "Marketing", 72000),
("Carol", "Engineering", 92000), ("Dave", "HR", 65000)]
df = spark.createDataFrame(data, ["name", "dept", "salary"])
result = df.groupBy("dept").agg(F.avg("salary").alias("avg_salary"))
result.show()
spark.stop()
import pyspark.sql.functions as F
spark = SparkSession.builder.appName("Test").getOrCreate()
# Create sample data
data = [("Alice", "Engineering", 85000), ("Bob", "Marketing", 72000),
("Carol", "Engineering", 92000), ("Dave", "HR", 65000)]
df = spark.createDataFrame(data, ["name", "dept", "salary"])
result = df.groupBy("dept").agg(F.avg("salary").alias("avg_salary"))
result.show()
spark.stop()
Expected Result
✓ PySpark initialises and runs the job
✓ Table shows avg salary by department
✓ Completion message shows execution time
5.2
Run a Hive SQL Query
⏱ 3 min · Easy
▾
Run SQL queries on big data using the Hive SQL environment.
1
In Workbench, select 🐝 Hive SQL tab
2
Click SELECT template or paste this query:
SELECT dept, COUNT(*) AS headcount,
AVG(salary) AS avg_salary,
MAX(salary) AS max_salary
FROM employees
GROUP BY dept
ORDER BY avg_salary DESC
AVG(salary) AS avg_salary,
MAX(salary) AS max_salary
FROM employees
GROUP BY dept
ORDER BY avg_salary DESC
Expected Result
✓ SQL executes via Spark SQL engine
✓ Results table appears with department stats
AutoML Pipeline
No-code machine learning from CSV to prediction
6.1
Train an ML Model on the Titanic Dataset
⏱ 5 min · Easy
▾
Use the pre-loaded Titanic dataset to train an ML model without writing any code.
1
Click the 🤖 AutoML tab in the platform
2
Click Use Pre-trained → select Titanic Dataset
3
Or upload your own CSV → click Clean Data → wait for cleaning report
4
Select target column → choose Classification → select models to race
5
Click Train Models → view the leaderboard
6
Click Download Best Model to save the .joblib file
Expected Result
✓ Model leaderboard shows accuracy scores
✓ XGBoost or Random Forest wins with ~85%+ accuracy
✓ Model file downloads successfully
6.2
Make Predictions with Trained Model
⏱ 3 min · Easy
▾
Upload new data and get instant predictions from your trained model.
1
In AutoML, scroll to the Predict section
2
Upload a new CSV file with the same columns (without the target column)
3
Click Predict and view results
Expected Result
✓ Predictions generated for each row
✓ Results downloadable as CSV
JupyterLab
Interactive GPU-accelerated notebooks
7.1
Open JupyterLab and Run a Notebook
⏱ 5 min · Easy
▾
Access the integrated JupyterLab environment and run Python code with GPU support.
1
Click the 📓 JupyterLab tab in the platform
2
JupyterLab opens in an iframe — login token:
opensourceai-jupyter-20263
Create a new Python notebook
4
Paste and run this code in a cell:
import pandas as pd
import numpy as np
# Generate sample dataset
df = pd.DataFrame({
'dept': np.random.choice(['Eng','Mkt','HR'], 1000),
'salary': np.random.randint(50000, 120000, 1000)
})
print(df.groupby('dept')['salary'].describe())
print("✓ Notebook running successfully!")
import numpy as np
# Generate sample dataset
df = pd.DataFrame({
'dept': np.random.choice(['Eng','Mkt','HR'], 1000),
'salary': np.random.randint(50000, 120000, 1000)
})
print(df.groupby('dept')['salary'].describe())
print("✓ Notebook running successfully!")
Expected Result
✓ JupyterLab loads in the iframe
✓ Code runs and shows salary statistics
✓ "✓ Notebook running successfully!" prints
GPU Analytics
cuDF RAPIDS — 50x faster than CPU pandas
8.1
Run GPU-Accelerated Analytics Benchmark
⏱ 5 min · Medium
▾
Run a GPU analytics benchmark to see the performance difference between CPU and GPU processing.
1
Go to 🔬 Workbench → 🚀 GPU Analytics
2
Click the Benchmark template
3
Click ▶ Run
# GPU Benchmark — paste in GPU Analytics tab
import cudf, pandas as pd, numpy as np, time
N = 5_000_000
data = {'a': np.random.rand(N), 'b': np.random.randint(0,100,N)}
# CPU benchmark
t0 = time.time()
pd.DataFrame(data).groupby('b')['a'].mean()
cpu_time = time.time() - t0
# GPU benchmark
t0 = time.time()
cudf.DataFrame(data).groupby('b')['a'].mean()
gpu_time = time.time() - t0
print(f"CPU: {cpu_time:.3f}s | GPU: {gpu_time:.3f}s | Speedup: {cpu_time/gpu_time:.1f}x")
import cudf, pandas as pd, numpy as np, time
N = 5_000_000
data = {'a': np.random.rand(N), 'b': np.random.randint(0,100,N)}
# CPU benchmark
t0 = time.time()
pd.DataFrame(data).groupby('b')['a'].mean()
cpu_time = time.time() - t0
# GPU benchmark
t0 = time.time()
cudf.DataFrame(data).groupby('b')['a'].mean()
gpu_time = time.time() - t0
print(f"CPU: {cpu_time:.3f}s | GPU: {gpu_time:.3f}s | Speedup: {cpu_time/gpu_time:.1f}x")
Expected Result
✓ Both CPU and GPU jobs complete
✓ GPU is 30-60x faster than CPU
✓ Speedup ratio shown in output
Sub-Workflows
Call saved workflows from within other workflows
9.1
Create and Call a Sub-Workflow
⏱ 8 min · Advanced
▾
Build a reusable sub-workflow and call it from a main workflow.
1
Create sub-workflow: Build Input → LLM → Output. Set Input to:
Summarise this text in one sentence: 2
Press
Ctrl+S → name it "AI Summariser" → save3
Clear canvas. Click ⎇ Sub-WF button in toolbar
4
Click Add to Canvas next to "AI Summariser"
5
Connect: Input → Sub-Workflow → Output. Execute.
Expected Result
✓ Sub-workflow runs inside the main workflow
✓ Output shows the summarised text
Error Handling
Route failures to fallback logic automatically
10.1
Set Up Error Handling Workflow
⏱ 5 min · Medium
▾
Load the Error Handling template and test automatic failure routing.
1
Click 📋 Templates → load "Error Handling Workflow"
2
You'll see: Input → LLM → Output, with Error Handler below LLM
3
Click Error Handler node → set Action to
fallback → set fallback value to: AI is temporarily unavailable. Please try again.4
Click ▶ Execute
5
Check the 🚨 Errors button in toolbar to view error log
Expected Result
✓ Workflow runs successfully end-to-end
✓ If any node fails, fallback value is used automatically
✓ Error log shows details of any failures
REST API
Access all platform features programmatically
11.1
Test All REST API Endpoints
⏱ 5 min · Medium
▾
Test the platform's REST API endpoints using curl commands.
1
Health check
curl http://localhost:5000/health
2
Run AGI Agent
curl -X POST http://localhost:5000/agent/run \
-H "Content-Type: application/json" \
-d '{"task":"Summarise the benefits of GPU computing in 3 points","steps":3}'
-H "Content-Type: application/json" \
-d '{"task":"Summarise the benefits of GPU computing in 3 points","steps":3}'
3
List available models
curl http://localhost:5000/api/models
4
Check workbench status
curl http://localhost:5000/workbench/status
Expected Result
✓ Health:
{"status":"ok","ollama":true}✓ Agent returns structured reasoning response
✓ Models lists available Ollama models
GitHub Integration
Create issues, list issues, and commit files from your workflows
1
Generate a GitHub Personal Access Token
⏱ 2 min · Easy
▾
You need a GitHub Personal Access Token (PAT) to authenticate API calls. This is like a password that lets your workflow interact with GitHub on your behalf.
1
2
Click Generate new token → Generate new token (classic)
3
Give it a name e.g.
Opensource AI Agent4
Set expiration: No expiration (or 90 days for security)
5
Select scopes — tick repo (gives full repository access)
6
Click Generate token → Copy the token
ghp_xxxxxxxxxxxxThe token is shown only once. Copy it immediately and store it somewhere safe before closing the page.
Expected Result
✓ Token starts with
ghp_ followed by random characters✓ Token is visible on screen — copy it now
2
Save Token in Credentials Manager
⏱ 1 min · Easy
▾
Save your GitHub token in the platform's Credentials Manager so it's available to all your GitHub nodes without re-entering it every time.
1
Open the platform at
http://localhost:5000/app2
Click 🔑 Credentials in the toolbar (or press
Ctrl+E)3
Fill in: Service Name =
GitHub, Type = API Key, Value = your ghp_xxx token4
Click + Save Credential
Credentials are stored locally in your browser. The GitHub node will automatically use the saved token — you don't need to paste it into every node.
Expected Result
✓ GitHub credential appears in the saved list
✓ Value is masked with ••••••••
3
Add a GitHub Node to Your Workflow
⏱ 3 min · Easy
▾
Add the GitHub node to your canvas and configure it to create issues, list issues, or commit files.
1
Press
Ctrl+K → type github → click GitHub to add it to canvas2
Click the GitHub node → fill in the properties panel on the right:
• GitHub Token:
• Repository:
• Action: choose from the list below
• GitHub Token:
ghp_xxxxxxxxxxxx (your token)
• Repository:
username/repo-name (e.g. johndoe/my-project)
• Action: choose from the list below
3
Connect an Input or LLM node to the GitHub node input port
4
Click ▶ Execute to run the workflow
Available Actions
create_issue
Create GitHub Issue
Takes the input text and creates a new issue in your repo. Perfect for auto-creating issues from error logs or AI analysis.
list_issues
List Open Issues
Fetches all open issues from your repository. Use with an LLM node to summarise or prioritise issues automatically.
create_file
Commit File to Repo
Creates or updates a file in your repository. Use to auto-generate reports, documentation, or config files.
Expected Result
✓ GitHub node runs and shows green tick
✓ Log shows: GitHub create_issue: https://github.com/user/repo/issues/1
✓ Issue appears in your GitHub repository
4
Build an Error → GitHub Issue Workflow
⏱ 5 min · Medium
▾
Build a complete workflow: paste an error log → AI analyses it → GitHub issue created automatically.
1
Add Input node → paste this error:
TypeError: Cannot read property 'id' of undefined at line 42 in UserService.js2
Add LLM node → set prompt:
Analyse this error and write a clear GitHub issue with: Title, Description, Steps to reproduce, Expected vs actual behaviour.3
Add GitHub node → set Token, Repo, Action =
create_issue4
Connect: Input → LLM → GitHub
5
Click ▶ Execute — check your GitHub repo for the new issue
# Test via API
curl -X POST http://localhost:5000/integrations/run -H "Content-Type: application/json" -d '{
"integration": "github",
"action": "create_issue",
"config": {"token": "ghp_xxx", "repo": "user/repo"},
"data": {"title": "Bug: API Error", "body": "Error found in production"}
}'
Expected Result
✓ All 3 nodes execute in sequence
✓ AI generates a well-formatted issue description
✓ Issue appears in your GitHub repository with AI-written content
5
Use the Pre-built GitHub Template
⏱ 1 min · Easy
▾
Skip the manual setup — load the pre-built GitHub Issue Creator template directly onto your canvas.
1
Click 📋 Templates in the toolbar
2
Filter by Integrations or DevOps & IT
3
Click GitHub Issue Creator → Use Template →
4
Click the GitHub node → enter your Token and Repository name
5
Click ▶ Execute
Other templates using GitHub: IT Ops Alert Handler (webhook → AI → GitHub + Slack) and Deploy Notifier (deploy hook → summary → team).
Slack Integration
Send messages and notifications to Slack channels from your workflows
1
Get a Slack Webhook URL
⏱ 3 min · Easy
▾
A Slack Incoming Webhook lets you post messages to any Slack channel with a single API call. No complex OAuth needed.
1
Go to api.slack.com/apps → click Create New App → From Scratch
2
Name it e.g.
AI Agent → select your workspace → click Create App3
Click Incoming Webhooks → toggle Activate Incoming Webhooks to ON
4
Click Add New Webhook to Workspace → select a channel → click Allow
5
Copy the Webhook URL:
https://hooks.slack.com/services/T.../B.../xxxSave the webhook URL in 🔑 Credentials with name "Slack" so all Slack nodes can use it automatically.
# Test your webhook
curl -X POST YOUR_WEBHOOK_URL -H "Content-Type: application/json" -d '{"text":"✓ Slack connected to Opensource AI Agent!"}'
Expected Result
✓ Message appears in your Slack channel
✓ Slack node shows green when executed
Google Drive Integration
Fetch, read and index Google Drive documents for RAG workflows
1
Connect Google Drive — Choose Your Method
⏱ 5 min · Medium
▾
There are two ways to connect Google Drive. OAuth is easier (one click), API Key gives more control.
🔐 Option A — OAuth (Recommended)
Go to
/credentials → click Connect Google Drive → sign in with Google → done in 30 seconds.🔑 Option B — Service Account
Go to Google Cloud Console → create Service Account → download JSON → paste at
/credentials.1
2
Click Google Drive card to expand it
3
Choose OAuth2 or API Key tab
4
Follow the connection steps → status shows ✓ Connected
5
Browse files → select documents → click 🧠 Index All Files for RAG
After indexing, load the RAG — Google Drive Documents template from 📋 Templates to start asking questions about your documents.
Expected Result
✓ Google Drive status shows Connected
✓ Files browser shows your Drive documents
✓ After indexing — RAG queries return answers from your docs