How to Research with Perplexity in a Jinba Workflow

Summary

Unstructured AI research creates compliance gaps because there is no record of what was searched, which sources were used, or where the result went.
Perplexity adds real-time web search with numbered citations and Deep Research reports in under four minutes, making claims more traceable.
A governed research pipeline follows three auditable steps: retrieve with Perplexity, summarize with an AI model, and route the result to Slack, Sheets, or a webhook.
Key practices: write narrow queries, verify citations before high-stakes use, monitor API limits, and iterate on workflow configuration rather than re-running ad hoc queries.
Jinba Flow turns these steps into a repeatable, auditable workflow and publishes a no-code form in Jinba App for non-engineers.

Unstructured AI research creates a specific operational problem. A query returns an answer, but there is no record of what was searched, which sources were used, or what happened to the result afterward. When that research feeds a business decision, the absence of a traceable process is a compliance gap.

Jinba addresses this by giving teams a structured environment to run AI tools inside governed, repeatable workflows. Combined with Perplexity as the research engine, the result is an automated pipeline: query in, summarized and routed output out, with every step logged.

What Jinba provides

Jinba is a platform for building, running, and deploying AI-powered workflows using natural language. It has two primary interfaces.

Jinba Flow is the workflow builder. Teams describe what they want in plain English, and Jinba generates a draft workflow. That draft can then be refined in a visual flowchart editor, where individual steps are configured, conditional branches are added, and exception paths are defined.

Jinba App is the execution interface. It exposes auto-generated input forms, so non-engineers can invoke a workflow without touching the underlying configuration.

The distinction matters for research workflows specifically. The person commissioning the research does not need to understand the tool chain. They submit a query through a form. The workflow handles the rest.

What Perplexity provides

Perplexity runs real-time web searches and compiles answers from indexed sources using models including GPT-5 and Claude 4.6 Sonnet. It is purpose-built for research queries rather than generative tasks.

Two features are directly relevant to workflow use:

Deep Research performs automated, multi-step searches for complex queries and synthesizes a report, typically in under four minutes.
Source citations appear as numbered references alongside every answer, so the origin of each claim is traceable.

The citation layer matters. A known issue with AI-generated research is that source quality is not always apparent from the answer itself. Perplexity surfaces the sources explicitly, which allows downstream steps in a workflow to flag or route results based on source type.

Building the research workflow

The following steps cover the full workflow: from environment setup through to a routed, summarized output. The Jinba tools documentation covers tool configuration in detail.

Step 1: Configure the Perplexity integration

Log into Jinba Flow and navigate to the Secrets tab. Add a Perplexity API key, obtained from the Perplexity API platform. Storing the key in Secrets rather than inside the workflow configuration keeps it out of execution logs and prevents accidental exposure when workflows are shared or exported.

Step 2: Create the workflow with natural language

Use the Jinba Flow chat interface to describe the workflow. A concrete starting instruction from Jinba's own published examples:

"Create a flow that uses Perplexity to search for the latest research on AI in education."

Jinba generates a draft workflow from this description. The draft can be accepted as-is or opened in the visual editor for refinement before any step executes.

Step 3: Add the Perplexity tool step

In the visual flowchart editor, add a step from the Search and Data Retrieval Tools category. Select Perplexity from the available tool list.

Configure the query field. This can be:

A static string, for recurring searches on a fixed topic
A dynamic value drawn from a preceding step or a workflow input field, using the variable syntax {{steps.<step_id>.result}}

Dynamic configuration is the more useful pattern for research workflows, because it allows the query to be driven by a form input submitted at run time rather than hardcoded in the tool configuration.

Step 4: Summarize and route the result

The raw Perplexity output contains the research findings alongside source citations. Add a subsequent AI Tool step, selecting an Anthropic or OpenAI model, and pass the Perplexity output into it using {{steps.<step_id>.result}}. Instruct the model to produce a structured summary at a specified length or format.

From there, add an Output Tool step to route the summary to its destination. Common destinations include Slack for immediate team notification, Google Sheets for structured record-keeping, or a downstream webhook.

The workflow now has three active steps: retrieve, summarize, route. Each step is a discrete, auditable unit.

Step 5: Execute via Jinba App

Publish the workflow and open it in Jinba App. The platform generates an input form from the workflow's defined inputs. The person running the research submits the query through that form. No access to the underlying flow configuration is required.

Review the output at each step to confirm retrieval and summarization are producing expected results. Adjust query phrasing or model instructions in the visual editor if the outputs require refinement.

Governance: the audit log

Ungoverned AI agent setups produce a well-documented operational problem. When research is performed across multiple tools with separate logging, it becomes difficult to produce a coherent record of what the agent retrieved, what it did with the output, and on what basis decisions were made. Fragmented logs across different interfaces cannot demonstrate a clear chain of accountability to an auditor.

Jinba's audit log addresses this at the workflow level. Every execution captures:

The trigger that initiated the run
The inputs passed to each step
The outputs returned by each step
Any approval decisions recorded during execution

This applies to the perplexity integration step the same as every other tool in the workflow. The record shows which query was submitted, what Perplexity returned, and what happened to that result in subsequent steps. The log is centralized, not fragmented across the tool chain.

For enterprise teams operating under compliance requirements, this structure answers the question auditors consistently ask: not just what the agent produced, but what it received, what it acted on, and in what sequence.

Practices that improve reliability

Query specificity directly affects output quality. Vague queries return broad answers that require significant downstream filtering. Specific queries, scoped to a named topic, time frame, or domain, return answers that require less manual review before routing.

A practical set of operating practices:

Write narrow queries. Specify the domain, the time frame, and the type of source required. "Latest peer-reviewed research on AI in clinical diagnostics, 2023 to present" performs better than "AI in medicine."
Verify citations before downstream use. Perplexity provides numbered citations, but source authority varies. Build a manual review step for high-stakes research before the summarized output is routed to production systems.
Monitor API usage. Perplexity's API platform enforces usage limits. Workflows that run at high frequency or on large query sets should include rate-limit handling in the flow configuration to prevent step failures.
Iterate on the workflow, not the query. When output quality is inconsistent, adjust the step configuration in the visual editor. Changing the AI Tool's summarization instruction produces more reliable results than rephrasing the research query at run time.

What to build next

A single-query workflow is a starting point. Once the retrieve-summarize-route pattern is functioning, the same structure supports more complex research operations: parallel queries across multiple topics, conditional routing based on source type or keyword presence in the summary, or scheduled runs that deliver research digests to a Slack channel on a defined cadence.

The Jinba tools documentation covers the full set of available integrations for extending the workflow beyond the perplexity integration step. The audit log applies across all of them, which means governance scales with the workflow rather than requiring separate configuration as new tools are added.

Start by publishing a minimal three-step workflow: Perplexity retrieval, AI summarization, one output destination. Review the audit log after the first run to confirm the execution record meets the requirements of the teams or auditors who need visibility into the process.

Frequently Asked Questions

What is Jinba and how does it work with Perplexity?

Jinba is an AI workflow orchestration platform that lets teams build, run, and deploy governed multi-step processes using natural language. When paired with Perplexity, Jinba handles the retrieval, summarization, routing, and audit logging of real-time web research in a single structured workflow instead of leaving agents to run ungoverned.

How do I connect Perplexity to Jinba?

You connect Perplexity to Jinba by storing a Perplexity API key in the Jinba Flow Secrets tab, then adding a Perplexity tool step from the Search and Data Retrieval Tools category. Storing the key in Secrets keeps it out of execution logs and protects it when workflows are shared or exported.

Why should I use Perplexity instead of ChatGPT for research workflows?

Perplexity is purpose-built for research queries. It runs real-time web searches, compiles answers from indexed sources, and returns numbered citations alongside each claim. This citation layer is critical for governed research workflows because it lets downstream steps flag or route results based on source type and gives auditors a traceable path back to the original source.

How do I create a Perplexity research workflow in Jinba Flow?

You can create a workflow by describing your goal in plain English in Jinba Flow, such as: “Create a flow that uses Perplexity to search for the latest research on AI in education.” Jinba generates a draft workflow that you can refine in the visual editor, add a Perplexity tool step, configure a dynamic query, and then add summarization and output steps.

Who can run a Jinba workflow after it is built?

Non-engineers can run a published Jinba workflow through Jinba App, which auto-generates an input form from the workflow’s defined inputs. The person commissioning the research only needs to submit a query through that form; they do not need access to the underlying flow configuration or engineering tools.

What does the Jinba audit log record?

The Jinba audit log records the trigger that initiated a run, the inputs passed to each step, the outputs returned by each step, and any approval decisions recorded during execution. For Perplexity integrations, this shows which query was submitted, what Perplexity returned, and what happened to that result in subsequent routing steps.

How do I summarize and route Perplexity results in Jinba?

After the Perplexity step returns research findings and citations, add an AI Tool step that accepts the Perplexity output using the variable syntax {{steps.<step_id>.result}} and instructs the model to produce a structured summary. Then add an Output Tool step to route the summary to a destination such as Slack, Google Sheets, or a downstream webhook.

What are the best practices for improving Perplexity research quality in Jinba?

Write narrow, specific queries that specify the domain, time frame, and type of source required. Verify citations before high-stakes use, monitor Perplexity API usage and rate limits, and iterate on the workflow configuration in the visual editor rather than repeatedly rephrasing the query at run time when output quality is inconsistent.

Can I use Perplexity Deep Research in Jinba workflows?

Yes, you can use Perplexity’s Deep Research capability inside Jinba by configuring the Perplexity tool step for complex, multi-step search queries. The workflow can then pass the synthesized report into summarization and routing steps just like any other Perplexity output, with the audit log capturing the full execution record.

What should I build after my first Perplexity workflow?

Start with a minimal three-step workflow: Perplexity retrieval, AI summarization, and one output destination. Once that pattern is stable, extend it with parallel queries across multiple topics, conditional routing based on source type or keyword presence, or scheduled runs that deliver research digests to a Slack channel on a defined cadence.