Make

Agent tools are modules, scenarios, MCP server tools and other agents, drawing on Make's app and module catalog, and the documentation states tools connect agents to third party services such as email, database and CRM systems to perform tasks; scenarios become tools with defined inputs and outputs, and MCP toolboxes group tools.

Sourcehelp.make.com Introduction to Make AI Agent (New) and Scenarios for AI agentsread 2026-09-30

Agents call other agents, scenarios and modules as tools, and agents run inside scenarios, whose engine routes and branches between modules and carries a documented error handling tree (error handlers, exponential backoff, throw), so deterministic steps and agent reasoning combine in one flow.

Sourcehelp.make.com Introduction to Make AI Agent (New) and error handling treeread 2026-09-30

Knowledge files are stored in a retrieval augmented generation vector database: Make splits each file into chunks, converts them to vectors and returns the most relevant chunks to the agent at request time rather than the whole file. Files of up to 20 MB, 20 per agent, in TXT, PDF or DOCX format or as CSV, MD or JSON, are managed at organization level, can be downloaded, replaced and deleted, and are deleted after 180 days without a query.

Sourcehelp.make.com Knowledge files for AI agentsread 2026-09-30

The customer controls which modules, scenarios and MCP tools an agent may call, and the best practices page presents giving the agent minimal access as the most reliable constraint, alongside prompt level instructions to have a human validate outputs, which the page warns agents may not always follow. No approval, checkpoint or pause for a person appears in the Make AI Agent (New) documentation.

Sourcehelp.make.com Make AI Agent (New) app and AI agent best practicesread 2026-09-30

The vendor's security page states it operates under an information security program that is ISO 27001 certified and that it has completed SOC 2 Type II and SOC 3 audits alongside GDPR adherence, explaining that the SOC 2 report details internal controls while the SOC 3 report is a public summary for external audiences, and linking a product-specific public SOC 3 report at trust.celonis.com/product/make/soc-3.

Named customer-facing controls include single sign-on with Google, Facebook and GitHub with Enterprise customers able to use their own SSO implementation, advanced role and team management within an organization profile, configurable log retention beyond the 30-day default on Enterprise, and a hosting environment accessible only from the private network via VPN with no direct public internet access.

The program includes regular independent penetration testing, vulnerability management, OWASP-aligned coding standards with static application security testing, AES-256 full-disk encryption with AWS KMS, TLS 1.2 and 1.3 in transit, a public bug bounty and an enterprise security whitepaper.

Sourcemake.com security pageread 2026-09-01

Each agent run returns metadata with its execution steps and token usage and a Reasoning tab showing processing steps, speed, context window use and any fallback connection, and reasoning shows the agent's step by step decisions for debugging; agents run as modules inside scenarios, which carry the platform's run records and error handling.

Sourcehelp.make.com Make AI Agent (New) app and Introduction to Make AI Agent (New)read 2026-09-30

The Run an agent module takes a Conversation ID that keeps a user's interactions in the same thread across runs, with a maximum conversation history setting. No scope beyond the thread, retention or delete path is documented for that history; data stores persist between scenario runs, but a database the builder reads and writes is application data rather than a memory layer.

Sourcehelp.make.com Make AI Agent (New) app and data storesread 2026-09-30

The vendor publishes an on-prem agent page shipping downloadable installers for Windows Server and for Debian, Ubuntu, Fedora and Red Hat, requiring Java 11 or above and configured with a client ID, secret and base URL, with a documented install sequence and a supporting help center page, allowing a customer to run an execution component inside their own infrastructure without a sales conversation; the page is marked beta.

The Enterprise platform runs in a separately managed AWS environment isolated from self-service cloud customers, and log data is retained 30 days by default with extended retention available on Enterprise. The security page states the cluster is deployed over two zones on AWS EC2 private instances within an Amazon VPC. No region selection or residency option is documented, and the platform itself is not self-hostable.

Sourcemake.com on-prem agents and security pagesread 2026-09-01

The AI Agents Library offers pre-made AI agent scenarios as a starting point, among them Account Researcher, Lost Deals Analysis, Resume Analysis, Sales Coach, Lead Qualification, Personalized Outreach, Interview Scheduler, Customer Feedback, HR Onboarding Q&A and Competitor Price Monitor agents, each doing its own job, and the agent introduction points builders to the library.

Sourcemake.com AI agents libraryread 2026-09-30

The triggers page documents agents started by the first module of their scenario, with guided examples for chat messages, Slack and Telegram messages, emails and mailhooks, form submissions and webhooks, and states that agents can run autonomously on a schedule.

Sourcehelp.make.com Create AI agents for different triggersread 2026-09-30

An AI provider connects the agent to a model, with OpenAI, Gemini and Anthropic Claude named as common providers; Make's own provider needs no provider account, custom AI provider connections are available on paid plans, and the Run an agent module takes a connection and a model selection.

Sourcehelp.make.com Introduction to Make AI Agent (New) and Make AI Agent (New) appread 2026-09-30

The Make MCP server lets MCP clients such as Claude and ChatGPT run scenarios and manage the Make account, including checking data stores and inviting organization members, with MCP toolboxes to limit the scenarios exposed, and the Make Developer Hub carries the API documentation and the path for building custom apps.

Sourcehelp.make.com Make MCP server and developers.make.com API documentationread 2026-09-30

Debugging surfaces are documented: the reasoning view for understanding each decision and debugging when something goes wrong, the error handling tree and the common errors page, and the best practices page's advice to monitor agent behavior through output bundles. No evaluation, scoring, test dataset or regression surface appears in the AI agent, best practices or error handling documentation.

Sourcehelp.make.com Introduction to Make AI Agent (New), AI agent best practices and error handling treeread 2026-09-30

Agent tools are modules, scenarios, MCP server tools and other agents, and no browser, desktop or computer use capability appears in the Make AI Agent (New) documentation or the Make AI Agent (New) app reference; web content reaches scenarios through app modules and HTTP requests rather than a browser Make runs for the agent.

Sourcehelp.make.com Introduction to Make AI Agent (New) and Make AI Agent (New) appread 2026-09-30