How to Set Up a Multi-Agent Research Pipeline with Claude Code (2026)

Research takes time. A proper competitive analysis requires gathering data from dozens of sources, synthesizing findings, and producing an actionable report. Doing this manually takes 8-15 hours per project. With a multi-agent research pipeline, you can reduce that to 30 minutes of setup and review.

This tutorial walks through building a research pipeline where one agent gathers raw data, another analyzes it, and a third produces a polished report. All coordinated through Ivern AI.

Related guides:

• How to Automate Research with AI Agents
• How to Build an AI Research Agent That Works
• How to Build an AI Research Pipeline That Actually Works

What You Need Before Starting

Requirement	Details
Ivern AI account	Free tier works -- sign up here
Claude Code installed	npm install -g @anthropic-ai/claude-code
API key	Anthropic API key with Claude access
Node.js 18+	Required for the agent daemon

Step 1: Create a Research Squad

A squad is a team of agents that work together. For a research pipeline, you want three agents with distinct roles:

1. Data Gatherer -- searches and collects raw information
2. Analyst -- processes data and identifies patterns
3. Report Writer -- formats findings into a structured report

In the Ivern dashboard, navigate to Squads > New Squad and name it "Research Pipeline." Add three agents with these system prompts:

Data Gatherer prompt:

You are a research data collector. Your job is to gather comprehensive
raw data on the given topic. Focus on:
- Recent statistics and metrics
- Expert opinions and quotes
- Industry trends and patterns
- Key competitors and their positioning

Return all findings in a structured markdown format with sources cited.

Analyst prompt:

You are a research analyst. Review the raw data provided and:
- Identify the top 5 key findings
- Highlight patterns and contradictions
- Rate the strength of each finding (strong/moderate/weak)
- Flag any data gaps that need further research

Output a structured analysis with clear sections.

Report Writer prompt:

You are a professional report writer. Take the analysis provided and
create an executive-ready report with:
- Executive summary (3-5 sentences)
- Key findings with supporting data
- Recommendations based on findings
- Sources appendix

Use clear headers, bullet points, and tables where appropriate.

Step 2: Connect Your Claude Code Agent

For BYOA (Bring Your Own Agent) mode, connect your local Claude Code:

npm install -g @ivern-ai/agent
ivern-agent connect --platform-api-key YOUR_KEY

The daemon polls the Ivern task board for assignments and spawns your local Claude Code instance to execute them.

For BYOK (Bring Your Own Key) mode, add your Anthropic API key in Settings > API Keys and Ivern calls the API directly.

Step 3: Configure the Pipeline Execution Order

In the squad settings, set the execution mode to Pipeline and order the agents:

1. Data Gatherer
2. Analyst
3. Report Writer

Pipeline mode ensures each agent completes before the next starts, with the output flowing from one to the next.

Step 4: Run Your First Research Task

Assign a task to your research squad:

Research the current state of AI code review tools in 2026.
Focus on pricing, features, and market adoption rates.

Each agent processes in sequence:

1. The Data Gatherer produces raw findings (2-3 minutes)
2. The Analyst processes those findings (1-2 minutes)
3. The Report Writer produces the final output (1-2 minutes)

Total time: approximately 5-7 minutes for a complete research report.

Step 5: Review and Iterate

The streamed output appears in real-time in the Ivern dashboard. Review the final report and if needed:

• Re-run with modified instructions
• Add a follow-up task for deeper research on specific findings
• Adjust agent prompts based on output quality

Cost Estimates

Agent	Estimated Tokens	Cost (Claude Sonnet)
Data Gatherer	~4,000 in / 3,000 out	$0.04
Analyst	~4,500 in / 2,500 out	$0.04
Report Writer	~3,500 in / 2,000 out	$0.03
Total per run		~$0.11

Compare this to $200-500 for a human research analyst doing the same work.

FAQ

Can I use GPT-4 instead of Claude?

Yes. Ivern supports both Anthropic and OpenAI providers. Add your OpenAI API key in Settings and assign models per agent.

How do I improve research quality?

Add specific constraints to your Data Gatherer prompt: required sources, date ranges, geographic focus, or data types. The more specific the instructions, the better the output.

Can I run multiple research tasks in parallel?

Yes. Create multiple tasks and assign them to the same squad. Ivern queues them and processes each through the full pipeline.

Is my data private?

With BYOA mode, your agents run locally on your machine. With BYOK mode, API calls go directly to the provider. Ivern never stores or trains on your data.