business-marketing

Relevance AI Review: Tested 45 Cells (2026)

Name: Relevance AI
Availability: InStock
Author: AI Demos

A no-code builder for business agents that can reason, retrieve documents, search the web, and wait for approval.

Visit Relevance AI

No-CodeHuman-in-the-LoopWeb SearchFree Tier Available

TL;DR — our verdictUpdated July 2026 · 10 test artifacts

Strongest no-code agent builder we tested for plain-English business workflows.

Where it wins

You want to build and test AI agents without writing code.
You need agents that can reason through business logic and produce structured outputs.
You work with internal documents and want grounded answers plus draft actions.

Main limitation

You need CRM, email, HRMS, or ticketing integrations on the free tier.

Pricing (verified plans)

Free $0/monthPro $19/monthTeam $234/monthEnterprise Custom — billed annually

Strongest test artifacts

OUTPUT →

Our take

Relevance AI was the strongest no-code agent builder we tested for turning plain-English instructions into working business workflows. It handled lead qualification, policy-grounded drafting, customer routing, live web research, and approval-gated email drafting without code; the main trade-off on the free tier is that CRM, email, HRMS, and ticketing handoffs still stay manual.

Walkthrough of Relevance AI in use across agent setup and run outputs.

In-Depth Review

Our detailed analysis of Relevance AI — features, performance, and real-world testing.

AI Demos Team

Expert Reviewer

Verified Review

Feature-by-Feature Breakdown

Prompt-Based Agent Configuration

The builder is genuinely no-code and easy to configure from plain-English instructions.

▾

Test Summary

Feature tested: Prompt-Based Agent Configuration

Result: Passed — The builder is genuinely no-code and easy to configure from plain-English instructions.

Relevance AI lets you define an agent in plain English by specifying role, business rules, required outputs, and guardrails. In this test set it was used to configure lead qualification, leave-policy, customer routing, company research, and approval workflows without code, nodes, or API work.

text

Lead Qualification agent instructions requiring High-Fit, Medium-Fit, or Low-Fit classification, a short reason, a next step, a follow-up email draft, and a structured CRM note for the QuickCart India lead.

↓→

image

Output artifact for "Prompt-Based Agent Configuration" test: The Lead Qualification agent was configured entirely in the prompt editor with plain-English rules; no code, nodes, or API wiring were needed., image-30.png

The Lead Qualification agent was configured entirely in the prompt editor with plain-English rules; no code, nodes, or API wiring were needed.

text

No HRMS or HR tool integration on the free tier. The Tools section is completely empty — no Darwinbox, Keka, or SAP is connected. The actual submission to any HR system remains a fully manual step.

↓→

image

Output artifact for "Prompt-Based Agent Configuration" test: The Leave Policy agent was also set up through the builder in plain English, and the free tier showed no connected HRMS tool., image-10.png

The Leave Policy agent was also set up through the builder in plain English, and the free tier showed no connected HRMS tool.

Bottom Line

Strong no-code setup for business users; the free tier still does not include downstream integrations or persistent action storage.

Grounded Retrieval and Research

The agent grounded its policy answer in the uploaded PDF and did not hallucinate policy facts.

▾

Test Summary

Feature tested: Grounded Retrieval and Research

Result: Partial — The agent grounded its policy answer in the uploaded PDF and did not hallucinate policy facts.

Feature tested: Grounded Retrieval and Research

Result: Partial

Verdict: The agent grounded its policy answer in the uploaded PDF and did not hallucinate policy facts.

Expected behavior: Relevance AI can search source material, pull relevant facts, and synthesize them into a grounded response. In these tests it worked both on an uploaded leave-policy PDF and on live web research for Lenskart, including citations and action-ready summaries.

Test case: PDF document → Image

Input type: PDF document

Input used: Input artifact (PDF document): INPUT — FutureSmart-AI-Leave-Policy.pdf

Observed output: Output artifact (Image): The run trace showed the agent using Search on the uploaded leave-policy PDF before responding. — image-9.png

Input artifact: Input artifact (PDF document): INPUT — FutureSmart-AI-Leave-Policy.pdf

Output artifact: Output artifact (Image): The run trace showed the agent using Search on the uploaded leave-policy PDF before responding. — image-9.png

What changed: PDF document transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): INPUT

Observed output: Output artifact (Image): The response used the policy facts correctly, drafted the leave request, and asked for full-day versus half-day confirmation, but it came after drafting and did not mention leave balance. — image-11.png

Input artifact: Input artifact (Text prompt): INPUT

Output artifact: Output artifact (Image): The response used the policy facts correctly, drafted the leave request, and asked for full-day versus half-day confirmation, but it came after drafting and did not mention leave balance. — image-11.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): INPUT

Observed output: Output artifact (Image): The agent produced a grounded company summary for Lenskart, listed AI automation opportunities, named likely decision-makers, and cited real sources. — image-15.png

Input artifact: Input artifact (Text prompt): INPUT

Output artifact: Output artifact (Image): The agent produced a grounded company summary for Lenskart, listed AI automation opportunities, named likely decision-makers, and cited real sources. — image-15.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): INPUT

Observed output: Output artifact (Image): A separate research run returned a general company overview but omitted the requested AI opportunities, outreach angle, and decision-makers, showing that instruction following can slip on multi-part research tasks. — image-18.png

Input artifact: Input artifact (Text prompt): INPUT

Output artifact: Output artifact (Image): A separate research run returned a general company overview but omitted the requested AI opportunities, outreach angle, and decision-makers, showing that instruction following can slip on multi-part research tasks. — image-18.png

What changed: Text prompt transformed into Image

Why it matters / Conclusion: Reliable grounding with visible retrieval; the factual answer was solid, but the clarification timing was not ideal.

Relevance AI can search source material, pull relevant facts, and synthesize them into a grounded response. In these tests it worked both on an uploaded leave-policy PDF and on live web research for Lenskart, including citations and action-ready summaries.

pdf

FutureSmart-AI-Leave-Policy.pdf

↓→

image

Output artifact for "Grounded Retrieval and Research" test: The run trace showed the agent using Search on the uploaded leave-policy PDF before responding., image-9.png

The run trace showed the agent using Search on the uploaded leave-policy PDF before responding.

text

I need to take leave next Friday (27th June 2025) for a medical appointment with my doctor. My manager is Priya Sharma. Can you check the leave policy and create a leave request for me? My name is Arjun Desai.

↓→

image

Output artifact for "Grounded Retrieval and Research" test: The response used the policy facts correctly, drafted the leave request, and asked for full-day versus half-day confirmation, but it came after drafting and did not mention leave balance., image-11.png

The response used the policy facts correctly, drafted the leave request, and asked for full-day versus half-day confirmation, but it came after drafting and did not mention leave balance.

text

Company Name: Lenskart; Website: lenskart.com

↓→

image

Output artifact for "Grounded Retrieval and Research" test: The agent produced a grounded company summary for Lenskart, listed AI automation opportunities, named likely decision-makers, and cited real sources., image-15.png

The agent produced a grounded company summary for Lenskart, listed AI automation opportunities, named likely decision-makers, and cited real sources.

text

Company Name: Lenskart; Website: lenskart.com

↓→

image

Output artifact for "Grounded Retrieval and Research" test: A separate research run returned a general company overview but omitted the requested AI opportunities, outreach angle, and decision-makers, showing that instruction following can slip on multi-part research tasks., image-18.png

A separate research run returned a general company overview but omitted the requested AI opportunities, outreach angle, and decision-makers, showing that instruction following can slip on multi-part research tasks.

Bottom Line

Reliable grounding with visible retrieval; the factual answer was solid, but the clarification timing was not ideal.

Structured Fielded Output

The platform can produce multi-field business outputs in one pass.

▾

Test Summary

Feature tested: Structured Fielded Output

Result: Passed — The platform can produce multi-field business outputs in one pass.

Relevance AI can return labels, reasons, next steps, drafted emails, and structured notes in a copy-ready format. In this research it was exercised on lead qualification and customer routing outputs that were already organized for downstream business use.

text

Lead Name: Rahul Mehta; Company: QuickCart India; Website: quickcartindia.com; Role: Head of Operations; Budget: $3,000/month; Requirement: automate customer complaint handling and route tickets to the right team automatically; around 500 complaints per day; 10 agents currently handle them manually.

↓→

image

Output artifact for "Structured Fielded Output" test: The lead-qualification run returned a Medium-Fit classification, business reasoning, next step, follow-up email draft, and CRM note in one pass., image-3.png

The lead-qualification run returned a Medium-Fit classification, business reasoning, next step, follow-up email draft, and CRM note in one pass.

text

Customer message: a subscriber was charged twice, had been waiting 5 days, wanted an immediate refund, and threatened to cancel and leave a public review.

↓→

image

Output artifact for "Structured Fielded Output" test: The routing run returned category, priority, escalation decision, assigned team, and an empathetic response draft for the duplicate-charge complaint., image-25.png

The routing run returned category, priority, escalation decision, assigned team, and an empathetic response draft for the duplicate-charge complaint.

Bottom Line

Strong fielded outputs across sales and support workflows, but the free tier still lacks export/download and some metadata fields.

Human Approval Gate

The approval loop is genuine and persists across turns.

▾

Test Summary

Feature tested: Human Approval Gate

Result: Passed — The approval loop is genuine and persists across turns.

Relevance AI can pause after drafting an action, ask for explicit YES/NO approval, and continue the loop based on the response. In the approval workflow it revised the email after a NO reply and kept waiting for confirmation instead of finalizing automatically.

text

I need to send a follow-up email to a potential client named Vikram Singh at TechNova Solutions. We met at a conference last week and discussed our AI automation services. He seemed interested in automating their HR onboarding process. Please draft a follow-up email for me.

↓→

image

The agent drafted the email and explicitly asked for YES or NO approval before treating it as final.

text

NO — make it shorter and more casual

↓→

image

After a NO reply, the agent produced a shorter, more casual redraft and asked for approval again.

Bottom Line

The approval gate is persistent and usable, but the free tier does not provide a durable approval audit trail or actual email sending.

Pricing & Access

Free tier works for building and testing; paid plans unlock broader integrations and audit features.

TESTED

Free

$0/month

200 Actions/month, $2 bonus Vendor Credits, unlimited agents and tools, 1 workforce, 1 user, 1 project, 30-day task history, marketplace access; sufficient for testing and prototyping.

Pro

$19/month (billed annually)

30,000 Actions/year, $240 Vendor Credits/year, unlimited workforces, 2 build users, scheduled tasks, chat mode, smart escalations, bring-your-own LLM.

Team

$234/month (billed annually)

84,000 Actions/year, $840 Vendor Credits/year, unused credits rollover, 5 build users, 45 end users, calling and meeting agents, A/B testing, analytics dashboard, priority support.

Enterprise

Custom — billed annually

Custom Actions and Vendor Credits, unlimited users and projects, 2,000+ integrations, agent evaluations, SSO/RBAC/audit logs, dedicated account manager.

Pricing checked June 2026. We re-check quarterly.

✓ Use This If

●You want to build and test AI agents without writing code.

●You need agents that can reason through business logic and produce structured outputs.

●You work with internal documents and want grounded answers plus draft actions.

●You need a human approval step before a risky action is finalized.

✕ Skip This If

●You need CRM, email, HRMS, or ticketing integrations on the free tier.

●You require persistent approval audit trails or session storage on the free tier.

●You need polished export/download features for every output panel.

●You need every multi-part research response to satisfy all requested sections without follow-up editing.

business-marketingagent-platformstext

No. In this research, all five agents were configured with plain-English instructions in the prompt editor rather than code or workflow nodes.

Yes. In the leave-policy test, it searched the uploaded PDF and the policy facts in the response matched the document exactly.

Yes. The leave-policy workflow both answered the policy question and drafted a usable leave request in the same response.

Not in the tests here. The Tools sections for those workflows were empty on the free tier, so the final handoff stayed manual.

Yes. It drafted the email, asked for YES or NO approval, and after a NO reply it produced a revised draft and asked again.

Yes. In the successful Lenskart run, it searched the web, identified what the company does, suggested automation opportunities, named likely decision-makers, and cited sources.

The report lists Free, Pro, Team, and Enterprise plans, with Free tested in this review.