Agents Builder is a no-code / low-code platform for creating AI agents that connect to Access products and services. It is the engine room that powers intelligent automation across the Evo ecosystem.
An agent is a configured AI that does a specific job. Unlike a general chat tool, an agent has defined scope, access to relevant data sources, and instructions that govern how it behaves. You build it once, and it can be used repeatedly — by you, by your team, or surfaced to users directly within Evo.
You give an agent:
A persona - who it is and what it's for.
Instructions - what it should do, and what it should not do.
Tools - access to data sources, APIs, and services it can act on.
The agent can then have conversations, analyse information, make decisions, and take actions on behalf of the user - all without writing application code.
By combining agents with MFEs (Micro Frontend components) hosted in the Access platform, you can deliver users a rich agentic workflow with interactive UI components rendered dynamically based on conversation context.
Is it worth a look?
It is if any of these apply:
You have a repeated task that involves gathering information, making a judgement, and taking or recommending an action — and you want that task handled by AI rather than done manually each time.
You want to give a team or set of users access to an AI that works within a defined scope, rather than a general-purpose chat interface.
You need AI to connect to actual data (querying a product catalogue, calling an internal API, fetching records) rather than working only from what is typed into a prompt.
You want to build something that sits inside an existing workflow, rather than a standalone tool users have to go and find.
How it fits into the Evo ecosystem
Agents Builder does not sit in isolation. It is designed to work alongside the other Evo pillars:
Pillar | Role |
Evo Builder | Build apps, features, and MFE components |
Agent Builder | Create AI agents with tools and instructions |
Evo Workflow Automation (EWA) | Orchestrate multi-step automated workflows |
Evo Navigator | Where agents are surfaced and used day-to-day |
Evo Researcher (coming soon) | Claude-desktop-style experience available to all Evo users |
The real power comes from combining these tools. An agent created in Agent Builder can be called from a workflow in EWA, surfaced in Navigator, configured as an Evo CoPilot plugin, or invoked from Evo Researcher.
Key Features
Agent Creation and Management: Create, configure, and manage AI agents using guided flows, AI-assisted setup, or manual configuration.
Tool Integration: Connect agents to MCP servers, OpenAPI/REST APIs, GraphQL endpoints, and MFE components.
Multi-Agent Orchestration: Enable collaboration between multiple AI agents to tackle complex tasks.
Knowledge Sources (RAG): Connect agents to Evo AI Memory for retrieval-augmented generation — useful when the agent needs to reference documents, policies, or product information that isn't available via a live API.
Agent Memory: Allow agents to remember information from past conversations, scoped per user or per organisation, enabling continuity in long-running workflows or support contexts.
Publishing and Sharing: Publish agents to Evo Navigator, call them from Evo Workflow Automation, or publish to the pre-built partition for sharing across organisations.
Security: Access Control Lists, sensitive agent mode with encrypted conversations, and agent ownership controls.
Analytics and Observability: Built-in analytics dashboard and detailed execution logs via Langfuse integration.
Evo CoPilot Integration: Configure agents as Evo CoPilot plugins with metadata, icons, and suggested prompts.
API Access: Programmatically interact with your agents via REST APIs with streaming support.
Key Concepts
System Instructions
Every agent has a system instruction — a detailed prompt that defines its persona, scope, workflow steps, and constraints. Agent Builder generates this automatically based on the information you provide during setup, and you can refine it at any point.
AI is very good at prompt engineering. Use the "Improve with AI" button to refine your system instructions, but always review the output — AI suggestions are starting points, not finished articles. Use it on your full system instruction for the most structured refinement, or at individual steps for smaller, easier-to-review changes.
Tools and MCP
Agents need tools to do useful things. Tools give the agent access to external data and services — for example, querying a product catalogue, calling an internal API, or fetching records from a database.
Tools are connected via MCP (Model Context Protocol) — think of it as "like a USB driver, but for AI." MCP provides a standard way for agents to discover and call external services.
If a service has an OpenAPI/Swagger spec, it can be wrapped into an MCP and registered on the platform. This means most internal APIs can be made available to agents without significant engineering effort.
Services
Pre-configured services are available on the platform for agents to use. During agent creation, the system will auto-suggest relevant services based on your agent's name and description — so the more specific you are, the better the matching.
User and Admin roles
Agents Builder has two apps:
App | Audience | Capabilities |
Agents Builder | Users / Co-worker Creator | Create agents and test them, focused on simpler scenarios |
Admin Agents Builder | Admins (internal Access teams) | Full feature set including agent publishing across organisations |
The main difference between User and Administrator roles is that Administrators can publish agents so they are visible to other organisations. In production, only Access Org admins can publish. The intention is that Admin Agents Builder will be an internal Access app, while Agents Builder will be a customer-facing app.
EVO Product Finder
A great example of replacing a research task with a conversational interface backed by live data. The problem: sales teams manually reviewing 50+ products to find the right fit for a customer — extremely time-consuming.
The solution was an agent that:
Gathers customer requirements through conversation.
Queries the EVO product catalogue via MCP.
Returns a ranked list of recommendations with justification.
People XD Smart Roster
AI agents that provide intelligent recommendations for employee scheduling and workforce management.
Paycircle
AI agents that assist in handling payroll tasks in a chat-style interface.
More Ideas
Customer Support Automation: Deploy AI agents with access to internal systems that handle customer inquiries, provide instant responses and free up human agents for more complex tasks.
Data Analysis and Reporting: Use AI agents to process large datasets, generate insights, and create reports.
In-Product AI Experiences: Surface AI capabilities directly within Access products using MFE components and agent tools.
Tips for Building Agents
Be specific when naming and describing your agent — the platform uses this to auto-suggest relevant services and MCPs.
Review generated system instructions carefully before publishing — AI suggestions can occasionally drift from your intent.
Test early and often — catch hallucinations and scope issues before users do.
Save regularly — Agent Builder does not auto-save during configuration.