What Is Dataiku Agentic AI: Agents for Large Dataflows

Dataiku AI handles large dataflows, processing data in minutes instead of days. You get infrastructure maintenance, time and money savings, and actionable insights that drive competitive advantage. As you’re here, you might already guess that almost all agentic AI workflows require a proxy: namely, a residential rotating proxy. 

This is a vast topic, and we’re at the beginning. I’ve already gathered the information for you: from user success stories online to firsthand data from my interviews with AI experts. Get ready, and let’s dive into it!

TL;DR

What is Dataiku AI and when you need it

Dataiku is an enterprise AI platform that unifies analytics, machine learning, and AI agents into a single governed environment. Its agents are autonomous systems powered by large language models (LLMs) that plan, retrieve data, invoke external tools, and execute multi-step workflows without requiring human intervention.

A company needs Dataiku AI agents when it has large, complex data operations that cannot scale with manual effort. The platform's ROI becomes particularly clear when workflows involve multi-source data, expert knowledge that needs to be packaged for wider teams, or recurring processes that currently cost analyst time.

A bit of Dataiku numbers for clarity:

• ZS Associates saw 60% faster root cause analysis and 25% fewer post-deployment errors

• Euronext saved analysts up to 20% of the time previously spent on recurring market queries. 

• Mitsubishi Electric accelerated analytics delivery by 60% by deploying Dataiku agents across their reporting stack. 

• John Lewis Partnership reported £40 million in ROI, with 25–30% higher conversion rates and 2x faster campaign launches. 

My favorite part about Dataiku is the easy access to the tool — the no-code kind of way. Our data analysts and citizen guys can enter very quickly and rapidly build a use case.

— Stéphane Callamand, digital transformer at Michelin

Dataiku AI usage: Data-heavy workflows

Dataiku AI agents work best in environments where data size and complexity are the most critical metrics. They're optimized to handle customer, financial, or scientific data and process it according to instructions. If your workflows fall into the categories below, they may help.

Automated financial flows

What Dataiku financial agents do:

• Validate invoices against contracted terms and flag mismatches automatically

• Analyze market pricing data across thousands of SKUs or securities

• Monitor payment anomalies and trigger alerts or escalations

• Route compliance-sensitive transactions for human review with AI-generated summaries

• Generate recurring financial reports by querying structured databases with natural language

Support and business operations

What Dataiku support agents do:

• Classify incoming tickets by category, urgency, and product area

• Retrieve answers from structured knowledge bases and send validated responses

• Open or update tickets in Jira, ServiceNow, or Freshdesk based on trigger conditions

• Escalate to specialists with an AI-generated summary of history and recommended action

• Measure resolution time and quality across large support backlogs

Supply chains and compliance

What Dataiku supply chain and compliance agents do:

• Monitor supplier risk by aggregating news, sanctions lists, and PEP databases

• Trigger reordering workflows based on inventory-level thresholds and lead-time predictions

• Run AML pattern analysis across transaction clusters and flag suspicious activity

• Prepare investigation summaries with recommended escalations for compliance officers

• Correlate delivery data with demand forecasts to surface bottlenecks proactively

Science and research usage

What Dataiku research agents do:

• Search global trial registries and rank potential sites by patient pool, geography, and performance history

• Extract and compare investigator performance metrics across trials

• Aggregate academic literature and return structured summaries for researchers

• Identify patterns across experimental datasets and flag anomalies for expert review

• Automate market research: gather competitive intelligence, extract key data, and produce analysis reports

When you need a proxy for Dataiku

Most Dataiku agentic workflows don't operate in a clean, controlled internal environment. Instead, they reach out to:

• scrape competitor data

• monitor external registries

• pull localized pricing

• query global compliance databases

Without a proxy layer, these agents get blocked, serve inaccurate geo-specific results, or expose the company's infrastructure IPs to external systems.

Rotating residential proxies solve each of these problems by maintaining a large pool of real-user IPs, automatically cycling them per request, and routing agent traffic through geographies that match the target data source. They serve as the operational backbone that makes the agent's data layer reliable and consistently clean.

Deploying and troubleshooting Dataiku AI agents 

To truly know something is to be able to deploy and use it. Here are basic algorithms for using Dataiku agentic AIs that apply to most workflows.

How to deploy the Dataiku AI agent

1. Log in to your Dataiku instance and navigate to the Projects dashboard.

2. Create a new project or open an existing one where you want to deploy the agent.

3. Go to the LLM Mesh via Administration → Connections → New Connection, and configure your preferred LLM (OpenAI, Anthropic, Azure OpenAI, or a custom/proxy endpoint).

4. Open the Agent Designer (available in GenAI flows or via the visual recipe builder) and define your agent's goal, tools (APIs, datasets, Dataiku flows), and memory settings.

5. Add tools like datasets, SQL endpoints, external REST APIs, or Python/R recipes it can call to accomplish tasks.

6. Test the agent in the interactive studio by reviewing chain-of-thought logs to verify that it calls the right tools and produces correct outputs.

7. Publish to Agent Hub for team-wide access. Set access permissions and governance rules (output review, human-in-the-loop triggers, escalation logic).

8. Monitor via Agent Management. Track uptime, response time, error rate, requests per minute, and business impact metrics (quality of outputs, policy alignment).

How to set up a proxy for AI agents in Dataiku

Step 1: Get your proxy credentials from CyberYozh

• Log in to your CyberYozh account.

• Navigate to Residential Rotating Proxies and generate your credential list

• Note your proxy host (IP), port, username, and password, and the rotation strategy setup

• Go to your API keys and generate an API endpoint which you’ll use in automation workflows

Step 2a: Configure the proxy in the Dataiku dashboard (global method)

• Go to Administration → Settings → Misc in your DSS web interface.

• Fill in HTTP Proxy Host (your CyberYozh gateway), Port, and authentication credentials.

• Save. Then, on any connection you want to proxy (S3, HTTP datasets, API Connect plugin), check Use global proxy.qiita+1

• All agent requests routed through those connections will now go through CyberYozh automatically.

Step 2b: Define the proxy directly in agent code (per-task method)

For Python tools or recipes called by your agent, add the proxy at the request level using your API key.  Here is a small example with basic rotating proxy setup:

import requests

# Get rotating proxy credentials

credentials = requests.post(
    'https://app.cyberyozh.com/api/v1/proxies/rotating-credentials/',

    headers={'X-Api-Key': “your_API_key”},

    json={

        'connection_login': 'your_login',
        'connection_password': 'your_password',
        'connection_host': 'your_IP',
        'connection_port': ‘your_port’,
        'session_type': 'your_session_type', # short_session, etc.
        'country_code': 'your_country_code', # US, UK, GE, etc.
        'amount': 5  # How many credentials you need
    }

)

# Get credentials in a code-readable JSON format
creds = credentials.json()['credentials']

# Use the first credentials set
proxy = {
    'http': f'http://{creds[0]}',
    'https': f'http://{creds[0]}'
}

# Use the proxy

response = requests.get("https://target-data-source.com", proxies=proxy)

Troubleshooting and known Dataiku issues

Practitioners on LinkedIn and in the Dataiku Community flag several recurring pain points. Here are the most common issues and how to address them.

1. Agent sprawl — too many agents with unclear ownership

Symptom: Multiple teams build overlapping agents; IT loses visibility; duplicated costs emerge.

Fix:

• Require all agents to be registered and published through Agent Hub before use.

• Appoint an agent owner for each deployed agent.

• Use Dataiku's Agent Management control tower to audit active agents, usage, and policy compliance.

2. External API and data source IP blocks

Symptom: Agent fails mid-run with HTTP 403 or 429 errors when collecting external data.

Fix:

• Configure a rotating residential proxy (e.g., CyberYozh) in Administration → Settings → Misc or directly in the agent's Python tool code.

• For agents accessing geo-restricted data, use CyberYozh's country-targeting feature via the API.

• Test the proxy connection before deploying: validate with a single request first, then run bulk tasks.

3. LLM and tool calls leaking sensitive data

Symptom: Agents send PII or confidential business data to external LLM APIs, creating compliance exposure.

Fix:

• Deploy a proxy as a local gateway between your agents and external LLMs.

• Configure LLM Mesh to point to the proxy endpoint rather than directly to OpenAI or Anthropic.

• It automatically detects and masks PII before prompts leave your environment, restoring original values in responses.

4. Agent reasoning failures and hallucinations

Symptom: Agent calls the wrong tool, takes an incorrect action, or produces fabricated outputs.

Fix:

• Review agent chain-of-thought logs in the Dataiku agent studio to trace which tool call caused the failure.

• Add explicit tool descriptions and parameter constraints to reduce ambiguity.

• Implement human-in-the-loop review steps for high-stakes actions (financial transactions, customer-facing outputs).

• Use Dataiku Reasoning Systems for multi-step workflows that require extended planning before acting.

5. Difficult-to-prove ROI on agent projects

Symptom: Agents are built, but the business doesn't see measurable impact, and funding for scaling is rejected.

Fix:

• Link every agent to a specific, measurable KPI before building (e.g., "reduce invoice processing time by X hours per week").

• Use the Agent Management → Business Impact tab to track KPI performance over time.

• Start with narrow, high-frequency use cases (support ticket routing, invoice validation) where volume is large enough to generate visible savings quickly.

Conclusion: Reduce time spent and get benefits from data

Dataiku AI agents turn weeks of expert effort into governed, repeatable workflows that run in minutes, with proven results in different conditions. Proxies make these agents reliable in the real world: residential rotating IPs bypass geo-blocks and rate limits, while keeping sensitive data within your perimeter. It ensures consistent data quality, compliance, and ROI.