Relevance AI

What is Relevance AI?

Relevance AI is a code-free AI Agent and AI Workforce platform designed for business automation. Users can define roles, goals, and instructions for these agents, and connect them to knowledge bases, tools, and enterprise applications, allowing them to automatically plan and carry out tasks related to sales, marketing, customer support, research, and operations.

A single agent can operate independently or carry out sensitive tasks after human approval; multiple agents, on the other hand, can form a workforce that assigns tasks based on roles and handles tasks among themselves.

It is more suitable for business processes that require “using tools to complete tasks”, rather than just generating chat responses.

Core functions of Relevance AI

Code-free AI Agent Builder

Users can set the agent’s name, role description, system instructions, models, knowledge, and tools through the interface, without the need to build a backend from scratch. The agent analyzes tasks, breaks them down into steps, selects the appropriate tools, and makes further decisions based on the results of its actions.

The platform also offers a Marketplace where existing agents and tools can be cloned and modified to suit specific business needs.

Tools, tools, and processes

A Tool is a specific capability that an Agent can execute, such as searching websites, sending emails, updating CRM systems, extracting documents, calling APIs, or running multi-step processes. Tools can be composed of code-free steps, model prompt chains, and integrated actions.

Even if a complex tool comprises multiple steps, each execution is counted as one Action; however, third-party services and models generate Vendor Credits.

AI Workforce multi-agent collaboration

Workforce organizes multiple specialized Agents into a virtual team. Upon receiving a task, the system assigns it to the most suitable Agent and can pass it on to another role when necessary.

Typical combinations include market researchers, lead scoring specialists, email writers, CRM update specialists, and agents responsible for follow-up meetings. To achieve consistent results, it is necessary to define the responsibilities, inputs and outputs, as well as the criteria for promotion for each role.

Knowledge base and long-term context

Agents can draw on documents, web pages, and structured data as knowledge to enhance the generation of responses through retrieval. Companies can use product information, sales manuals, policies, and frequently asked questions to guide the output.

A knowledge base cannot automatically ensure accuracy; it is still necessary to manage versioning of the data, access permissions, search parameters, and sensitive fields.

Over 2,000 integrations and triggers

The platform offers a wide range of application integrations, enabling connections to CRM systems, email services, messaging tools, spreadsheets, databases, and task management tools. Tasks can be triggered from calendars, Webhooks, WhatsApp, LinkedIn, Telegram, or enterprise-level platforms such as Salesforce, Snowflake, and Zendesk.

Some advanced triggers are subject to package restrictions, and third-party applications may also incur additional fees.

Manual approval and Smart Escalations

When an anomaly is detected, or when an operation has a low level of confidence or is associated with high risk, the Agent can request manual input and approval via email or Slack. Teams can require approval for actions such as sending messages, modifying customer data, making price commitments, or deleting data, thereby avoiding the business risks that come with complete automation.

Calling, Meeting, and Evaluation Analysis

Team and Enterprise offer features such as phone and meeting agents, A/B testing, and analytics for handling calls and meetings or for comparing different agent configurations. Enterprise additionally provides agent evaluations, working hour controls, multi-organization management, and auditing capabilities.

The evaluation should cover task success rate, tool misuse, delays, costs, and the rate of manual intervention.

API and official SDK

Developers can use APIs to initiate Agent, Tool, and Workforce tasks within their own products. The official JavaScript/TypeScript SDK is compatible with Node.js, Deno, Bun, Cloudflare Workers, and browsers, and it provides task events as well as real-time updates.

The official Python SDK allows for the management of Agents, tools, knowledge, and tasks.

The relevant SDKs are licensed under open-source licenses such as MIT or Apache 2.0, but the Relevance AI cloud platform and its core services are not open-source products.

Comparison of Relevance AI prices and plans

The current billing is divided into two parts:Actions It indicates the number of times the agent has executed the tool.Vendor Credits Payment model and costs related to third-party AI services: The vendor credits are charged at the actual cost incurred by the supplier, without any additional markup.

The following are the packages currently listed in the official documentation.

The annual pricing for Pro and Team plans saves approximately 33% compared to the monthly pricing; the exact monthly costs and taxes are indicated on the settlement page. The Business plan, which is no longer available, is reserved only for some existing users and should not be considered as an option for new users.

How are Actions and Vendor Credits calculated?

A single tool can encompass multiple steps, so proper design can help reduce the number of actions required; however, the more complex the tool, the greater the challenges in terms of debugging, recovery from failures, and resource consumption.

Running Tools or Agents on the platform also incurs usage of basic credits; before creating a production process, it is necessary to use the task cost details to determine the actual amount of resources that will be used.

Relevance AI Usage Guide

1. Select scenario:First, define a measurable task, such as researching leads, updating the CRM, or creating and approving follow-up emails.
2. Create Agent:Specify the role, objectives, prohibited actions, completion criteria, and situations in which an upgrade is required.
3. Add knowledge:Import authorized and up-to-date product, policy, and business materials, and test the search results.
4. Configure Tools:Connect to search, email, CRM, or internal APIs; use a test account with minimal permissions first.
5. Set up triggers:Select Manual, Schedule, Webhook, or Application Event to add approval for external write operations.
6. Run tests:It covers normal cases, missing data, duplicate tasks, and tool failures, allowing you to view task steps and costs.
7. Components of Workforce:Once the process is stable, the roles can be separated, allowing multiple agents to hand off tasks according to clear protocols.
8. Online monitoring:Track success rate, action consumption, vendor credits, manual intervention, and errors.

Relevance AI Usage Guide

Create reusable professional workflows

1. Select a low-risk, real-world project as a template;
2. Fix the order of use for the code-free AI Agent Builder, various Tools, processes, and multi-agent collaboration within the AI Workforce;
3. Record the environment, model, prompts, and failure conditions;
4. Set up manual approval for write, deploy, and delete actions;
5. Compare speed, cost, test pass rate, and the amount of rework;
6. Expand to a team or production environment only after stability has been verified;

Which users are it suitable for

• We hope to establish a GTM team responsible for researching sales leads, scoring them, and following up with the relevant agents.
• Operators and automators who need to connect various enterprise applications without using any coding;
• Product and innovation teams that aim to use multi-agent systems to handle complex tasks;
• Developers who need to integrate Agents or Workforces into their own applications;
• Enterprise organizations that place emphasis on manual approval, analysis, permissions, and auditing.

Advantages and precautions

• The advantage of RelevanceAI is that it brings together Agents, tools, knowledge, integration capabilities, triggers, and multi-agent systems within a code-free platform, while also providing for human intervention and enterprise management functions.
• Actions are separated from model costs, which makes them easier to estimate than single fuzzy credits;
• However, once an Agent is able to connect to emails, CRM systems, and databases, it can also have an impact on actual business operations.
• Minimum permissions, testing environments, approval processes, idempotency, and operation auditing must be implemented to prevent duplicate messages from being sent or data from being updated incorrectly.
• The carryover of VendorCredits is conditional on maintaining the subscription; it is necessary to check the balance rules before canceling the subscription.
• The platform is a cloud service, and sensitive data requires evaluation with regard to privacy, data residency, and corporate contracts.

Frequently Asked Questions

Is Relevance AI free?

There is a Free plan that includes 200 Actions per month, as well as one-time Vendor Credits, making it suitable for creating and testing small agents.

What is an Action?

Each time an Agent or user runs a Tool, one Action is counted. A Tool can contain multiple steps, but a failed execution still results in a deduction of Action points.

Can I use my own OpenAI or other model keys?

The paid plans support Bring Your Own LLM; the cost of the models is covered by the respective suppliers, thereby allowing avoidance of Relevance AI’s Vendor Credits.

Does Relevance AI have an API?

Yes, official JavaScript/TypeScript and Python SDKs are provided to trigger Agents, tasks, tools, and Workforce, as well as to receive real-time updates on tasks.

Is Relevance AI open source?

The platform itself is not open source. Although the official SDKs and some development tools have their source code made available, this does not mean that it is possible to deploy the entire Agent platform on a private basis.