AgentX

What is AgentX?

AgentX is a platform used for building, orchestrating, evaluating, deploying, and monitoring AI Agents. Individual developers can create agents for their own products or internal processes, while agencies can deliver multi-agent solutions to clients under their own brand.

The platform offers both self-service Builder and enterprise-based construction services, covering knowledge bases, tool invocation, workflows, APIs, channel deployment, and evaluation. Different models and actions all consume Credits; the enterprise solutions also provide dedicated infrastructure, SSO, compliance support, and options for local deployment.

Core functions

• Create AI Agents with roles and instructions using the visual Builder.
• Construct multi-agent workflows by combining management agents with specialized agents.
• Connect websites, documents, Google Drive, and manual knowledge
• Use RAG and rearrangement to generate evidence-based responses from a knowledge base.
• Calling Python tools, MCP services, and external business systems
• Deployment via API, website, Slack, Discord, and other channels
• Use datasets and multiple runs to evaluate the reliability of the Agent.
• Track sessions, tasks, tool calls, and performance metrics
• Provide agents with white-labeling, customer workspaces, and team collaboration tools.
• Provides enterprises with a dedicated environment, SSO, SLAs, and on-premises deployment.

Three ways of use

Visual Agent Builder

Users can create Agents in the workspace and configure their names, avatars, creativity levels, models, role descriptions, and response rules. Building an Agent is possible with the free plan; no credit card is required.

• Choose from OpenAI, Claude, Gemini, or other available models.
• Define the agent’s identity, objectives, tone, and boundaries of behavior.
• Set the topics that require a response and those that must be declined.
• Add brief fixed knowledge and common FAQs.
• Link documents, web pages, and cloud storage to create long-term knowledge.
• Enable document reading, lead collection, or custom tools
• Test repeatedly with real samples before release.

Multi-agent workflow

AgentX Workforce consists of multiple specialized Agents coordinated by a Manager Agent, which decides whom to assign a task to based on its requirements. Different Agents can utilize various models, knowledge bases, and tools to carry out tasks such as research, validation, writing, or business execution.

• The Manager splits the tasks and selects appropriate Agents.
• Have different agents be responsible for retrieval, analysis, review, and execution respectively.
• Aggregate the results of multiple Agents within the same session
• Use different models to balance cost, speed, and quality.
• Publish the workflow as a reusable business capability
• Record task assignment and the final answer
• Add manual approval for high-risk external actions

Knowledge base and RAG

Each Agent can have its own independent knowledge base; the platform extracts content from these bases to create a vector retrieval system. The sources of this content can be individual pages, site maps, sub-pages discovered on the main website, local documents, Google Drive, or data entered manually.

• Capture content from authorized websites to create knowledge sources.
• Upload Word, TXT, digital PDFs, and other supported files
• Connect to Google Drive and sync subsequent changes.
• Extract the text from PDFs and create a searchable index.
• Use RAG to select relevant segments and then generate answers.
• Improve the relevance of the returned context through rearrangement.
• Isolate customers and business knowledge by Agent

The official old documentation specifies that the maximum number of pages in a PDF is 200; larger files require contact with the platform or need to be split into smaller parts. The actual format, file size, and synchronization limits are determined by the current workspace settings.

Document and multimodal processing

• Extract fields from contracts, invoices, orders, and forms.
• Read the documents and images uploaded in the chat sessions.
• Run advanced OCR when needed.
• Process CSV, Excel, and other structured files
• Generate text, image, or audio-related outputs.
• Verify the extracted content and carry out subsequent business processing.
• Fields with low confidence or missing values are routed for manual review.

Tools and MCP

AgentX supports custom Python tools and MCP Servers, which enable agents to read external data or carry out business-related actions. Tool permissions should be set to the minimum necessary; models must not be granted unrestricted system privileges.

• Package the enterprise API into tools with defined inputs and outputs.
• Connect to approved MCP services to gain professional capabilities.
• Perform calculations, conversions, or internal logic using Python.
• Types of customers, records, and operations to which the tool has access limited to
• Verify all tool parameters and structured output.
• Set approval thresholds for writing, sending, deleting, and making payments.
• Record each call, the results, any exceptions, and human decisions.

Agent API

AgentX provides API keys and an official SDK, which enable interaction with, querying, tracking, or orchestrating Agents that have already been created on the platform. The older version of the documentation indicated that Agents were primarily created through the interface, while the API capabilities should be based on the current reference documents.

Deployment channels

AgentX enables the deployment of agents on proprietary products, websites, and collaboration platforms; the agent packages can also be delivered under a custom brand. The capabilities of these platforms may be at different levels of maturity, and it is necessary to verify each one before they are put into use.

• Embed own products and backend processes via API
• Publish chatbots or embedded components on a website
• Deploy to collaboration platforms such as Slack and Discord
• Connect to messaging apps such as WhatsApp
• Trigger tasks using Webhooks, Zapier, and workflow systems
• Create a separate white-label space for customers.
• Connect internal systems through the enterprise environment

Clue collection

Agents can collect names, email addresses, and custom lead fields during conversations, and they can connect to HubSpot, Wix CRM, or Webhooks. Before collecting this data, it is necessary to explain its purpose, obtain consent, and verify the data format.

• Define required and optional lead information
• Collect fields gradually during natural conversation
• Avoid asking again for information that has already been provided.
• Verify the format of email, phone number, and business field values.
• Synchronize to CRM or custom Webhook
• Restrict sensitive information and set data retention periods.
• Manual follow-up is initiated for high-value leads.

Agent evaluation

AgentX offers a structured evaluation framework that uses datasets, multiple runs, and various metrics to assess reliability, security, and performance. The official Python SDK also supports LLM-as-a-judge, cosine similarity, and Jaccard similarity.

• Establish input, expectation, and scoring rules using real-world examples.
• Check the stability of the output by running multiple tests on the same case.
• Compare different models, prompt configurations, and tool settings
• Use LLMs to assess the quality of answers, while retaining manual random checks.
• Measure semantic similarity and word set overlap
• Check samples of rejected responses, privilege violations, and prompt injection.
• Continuous monitoring of regressions and business outcomes after going live

Evaluation level

White-label and customer workspace

Professional and Business are designed for teams that create agents for clients, and they include options for white-label deployment as well as client workspaces. Agents can manage multiple projects and decide for themselves how to charge their clients.

• Use a proxy brand to showcase the Agent experience.
• Create an isolated workspace for each customer.
• Centralized management of customer Agents, knowledge, and Credits
• Invite customers to view or use the specified Agent.
• Avoid mixing data and tool permissions among customers.
• Specify the responsibilities regarding AI, data, support, and outputs in the contract.

Applicable scenarios

• Website customer service, sales FAQs, and lead collection
• Processing of invoices, contracts, orders, and forms
• Corporate knowledge assistant and employee self-service
• Multi-Agent research, writing, review, and reporting processes
• Execute internal operational tasks via Slack or API
• The agent delivers white-label AI Agents to multiple clients.
• Compare the proprietary Agent with third-party frameworks using a evaluation framework.
• Companies convert repetitive manual processes into controlled Agent workflows.

Which users are it suitable for

• Individual developers who learn about and test the concept of Agents
• Startup teams that need to integrate agents into their products
• Establish internal automated business and IT teams
• Agencies and consulting firms that implement AI solutions for clients
• Engineering teams that require evaluation, tracking, and multi-Agent orchestration
• Enterprises that require SSO, a dedicated environment, or on-premises deployment

AgentX price

AgentX offers versions such as Free, Solo Builder, Professional, Business, and Enterprise, depending on the way it is used. There are significant differences in terms of monthly/annual billing, the number of agents, licenses, credits, and the ability to use a custom brand.

Comparison of package features

Both Professional and Business plans include 2 seats; additional seats cost $10 per month. Additional Credits for the public plan can be purchased at $10 for every 1,000 credits.

Credits billing

Credits are the unified unit for measuring AI usage in AgentX; actions such as generating responses, extracting information from documents, and running tools all consume credits. Different models require different amounts of credits – for example, GPT-4o uses 10 credits per use, Claude Sonnet uses 12 credits per use, and Gemini Pro uses 8 credits per use.

• The amount of Credits consumed depends on the model and the specific action.
• Complex toolchains may be used multiple times within a single task.
• Document processing and multi-Agent delegation increase the overall usage.
• The console should allow users to view the items that have been consumed and the reasons for it.
• The public price for additional Credits is $10 per 1,000.
• Before going live, it should be estimated based on successful business tasks rather than individual conversations.

Notes for the price page

The official package cards on the pricing page indicate that the Professional plan costs $199 per month, but individual cost examples below the page show $249. Purchases should be based on the settlement page, the order details, and the current package card, with attention paid to any promotions, annual payment discounts, and renewal prices.

• Check whether the monthly and annual plans offer the same benefits.
• Clarify the difference between one-time credits and monthly credits
• Determine whether to stop operation or automatically purchase more when there is a surplus.
• Confirm whether to carry over if no Credits have been used.
• Include additional seats, models, and professional services in the budget.
• Enterprise customers should clarify the infrastructure, SLAs, and scope of support.

Tutorial on Creating Agents

1. Register for a free account and access the workspace.
2. Create a new Agent and enter its name, role, and target.
3. Select a model that meets the cost and quality requirements.
4. Define response rules, prohibited items, and output format.
5. Add brief information, website, or document sources.
6. Only enable the tools and channels that are truly necessary for the task.
7. Use normal, abnormal, and attack samples to test the responses.
8. It is deployed to real user channels only after evaluation.

Tutorial on Building Multi-Agent Processes

1. Break down the entire business process into verifiable, specialized tasks.
2. Create a specialized Agent with a single responsibility for each task.
3. Set the delegation and aggregation rules for the Manager Agent.
4. Configure separate knowledge, models, and tools for each Agent.
5. Define the structured input and output transmitted between Agents.
6. Manual approval is required for actions that write to external systems.
7. Use multiple rounds of evaluation to check the stability and cost of delegation.
8. After going live, it is determined which Agent or tool is responsible for the tracking failure.

API Deployment Tutorial

1. Create an API Key and store it in a secure key management system.
2. Install the official Python SDK or use the API reference.
3. Create and test the target Agent on the platform interface.
4. Create a separate Conversation context for end users.
5. Use streaming output to improve the experience when dealing with long responses.
6. Handling limits, timeouts, retries, and tool failures.
7. Log requests, models, Credits, references, and user feedback.
8. Regularly rotate the keys and conduct regression tests.

Security and AI Governance

• Isolate knowledge, sessions, and tools by workspace and customer.
• Use API keys, Webhooks, and MCP connections with the minimum required permissions.
• Perform checks for malicious content and permissions on the uploaded files.
• Prevent prompt injection in web pages and documents
• Do not expose model inference or system prompts to end users.
• Manual confirmation and audit records are reserved for high-impact actions.
• Define the purposes for collecting personal data, obtain consent, and specify the retention period.
• Complete risk mapping, red team testing, and failure handling planning prior to going live.

Effect evaluation

Product advantages

• Covers the entire Agent lifecycle, including construction, evaluation, deployment, and monitoring.
• The visual Builder is suitable for non-professional developers to get started quickly.
• Supports the formation of multi-agent teams consisting of Managers and professional Agents.
• The knowledge base can connect to websites, documents, and Google Drive.
• API, MCP, A2A, and official SDKs facilitate development and integration.
• White-label and customer workspaces are suitable for agents’ commercial delivery.
• The free plan provides 5 agents and API access for testing.
• The prices are transparent, and Credits are used uniformly to measure the amount of AI usage.
• The company offers SSO, dedicated infrastructure, and on-premises deployment options.

Usage restrictions and precautions

• Free offers only 200 Credits once, which is not suitable for continuous production workloads.
• Complex multi-Agent processes lead to a rapid increase in Credits consumption.
• The quality, speed, and unit consumption of different models vary.
• There may be differences in the naming of functions between the old help documentation and the new price page.
• The examples of individual costs on the price page do not match the amounts listed on the official package cards.
• Capabilities such as creating Agents via API need to be verified according to the current interface documentation.
• Website and document knowledge may contain outdated or malicious content.
• AI evaluation metrics cannot replace business experts and security audits.
• Third-party channels, models, and MCP services introduce additional dependencies.
• The core hosting platform is not a fully open-source project.

GitHub and open source

The official AgentX GitHub repository provides a Python SDK, SDKs for JavaScript or TypeScript, as well as several MCP Servers. The official Python SDK is licensed under the MIT license and can be used for interacting with hosting platforms, orchestrating multiple agents, and conducting evaluations.

The fact that the open-source SDKs and MCP connections are available does not mean that the AgentX platform itself is open-source; the visual builder, the managed knowledge base, the white-label workspaces, billing features, the enterprise control panel, and the full set of services it offers remain commercial products.

Basic information

Recommendation score

4.6 / 5. AgentX is suitable for Agent developers and agencies that need to move from prototyping to evaluation, deployment, and delivery to multiple clients; the rules regarding packages and Credits are clear. However, the costs associated with complex processes, differences between old and new documents, and corporate governance aspects still need to be verified before going live.

Frequently Asked Questions

What is AgentX mainly used for?

It is used to create, orchestrate, evaluate, and deploy AI Agents, and it also supports the delivery of client projects under the agent’s own brand.

Can AgentX be used for free?

Yes, the Free plan is permanently free; it includes 1 workspace, up to 5 Agents, and 200 Credits for one-time use.

How much is Solo Builder?

49 dollars per month or 490 dollars per year, including up to 25 Agents and 5,000 Credits per month.

What is the difference between Professional and Business?

Both support white-labeling and customer workspaces; Business offers an unlimited number of agents, 20,000 Credits, as well as enhanced support and SLAs.

How much are the additional Credits?

The public plan shows that 10 dollars are charged for every 1,000 additional Credits.

Which models is AgentX compatible with?

The document lists options such as OpenAI, Claude, Gemini, and Llama; the specific models available depend on the current list in the workspace.

Does AgentX support MCP?

Yes, the official SDK can connect to the MCP Server, and GitHub also makes some MCP services available publicly.

Can AgentX be deployed locally?

The Enterprise option includes local deployment; the specific architecture, data boundaries, and costs need to be determined upon customization.

Is AgentX open source?

The core platform is not open-source, but the official Python SDK and certain MCP projects are made available under the MIT license.