--- title: "Agent Builder" canonical: https://harmony.io/platform/ai-agent-builder --- # Agent Builder Turn business processes into AI agents in minutes. Connect enterprise systems, define guardrails, and deploy autonomous agents across every employee channel - no code required. ## Agent Builder Turn business processes into AI agents in minutes. Connect enterprise systems, define guardrails, and deploy autonomous agents across every employee channel, no code required. ## What is an AI agent builder? An AI agent builder is where a process your team runs by hand becomes an agent that runs it for you. You describe the job in plain language - who can ask for this, what has to be checked, what happens next - and the builder produces an agent that plans the steps, calls the systems involved, and knows when to stop and ask a person. Harmony covers both ends of that. A library of pre-built agents handles the processes every IT team shares: onboarding and offboarding, password and MFA resets, application access requests, license expiry, low storage and end-of-life hardware. The visual Custom Workflow Builder handles everything specific to you, with manual, scheduled, event and webhook triggers, If/Else branching, For Each loops over up to 20 parallel items, and durable sleep timers that survive a restart. Every agent inherits the platform’s guardrails - your [approval policies](https://harmony.io/platform/compliance-and-policies), identity-provider-backed access, scoped permissions and a step-by-step audit log of each run - and can be simulated in a sandbox, released to one team first, versioned and rolled back in a click. [Harmony Genie](https://harmony.io/platform/harmony-genie), our AI assistant, can create and change agents for you in plain English, and open the Agent Builder when you want finer control. ## How the Agent Builder helps your teams Enterprise workflow agents: build specialized agents for every department, from IT support, access approval, and knowledge assistance to finance requests and procurement Cross-system agents: orchestrate actions across ITSM, HRIS, identity providers, SaaS applications, databases, and internal tools Autonomous execution: build agents that plan, reason, and execute multi-step workflows, not just answer questions Enterprise integrations: connect agents to any API, MCP server, webhook, database, SaaS application, or custom action Controlled rollout: test new agent capabilities with specific teams before company-wide deployment Human escalation: seamlessly hand off conversations or actions to people when needed ## From process to production agent in minutes ### No Code Simple building experience: business teams build and iterate on AI agents without engineering ### Deploy Everywhere Build once, deploy everywhere: agents work across every communication tool, or autonomously in the background ### Shared Memory Shared enterprise memory: every agent operates with a consistent understanding of your organization’s knowledge, policies, and context ### Guardrails Built In Enterprise-grade guardrails: every agent inherits your approval policies, access controls, and audit requirements ### Live In Minutes Production-ready in minutes: from first build to production ## Powerful capabilities built for enterprise IT - Intent-driven LLM builder: describe the agent in natural language - Enterprise knowledge grounding: combine documentation, APIs, databases, SaaS apps, and MCP servers into a single reasoning layer - Guardrail configuration: define permissions, escalation rules, approval policies, and operational boundaries - Template library: pre-built agents for common use cases across departments - Testing sandbox and simulations: run simulated conversations before going live - Agent analytics: measure usage, success rates, resolution rates, and business impact - Versioning and rollback: safely iterate on agents with version control and one-click rollback ## How the Agent Builder works **Trigger:** You have a process that runs on tribal knowledge: a runbook, a spreadsheet, or a person everyone messages. 1. **Describe the agent** - Write what it should do in natural language, or start from a template in the library. The builder turns the description into a configured agent with its trigger, inputs and steps laid out for review. 2. **Connect the systems it needs** - Point it at the systems that hold the truth and take the action: identity providers, HRIS, MDM, ITSM, databases, SaaS applications, any HTTP-accessible API, MCP servers and webhooks. 3. **Set the guardrails** - Define who can trigger it, which approvals apply, what it may and may not touch, and when it escalates. Guardrails come from the policies you already maintain instead of being rewritten per agent. 4. **Test before it goes live** - Run simulated conversations and dry runs in the sandbox, then release the agent to one team or department to prove it against real requests before company-wide rollout. 5. **Deploy, measure, iterate** - The agent runs in Slack, Teams, email and the portal, or silently in the background on a schedule. Analytics show usage, success and resolution rates, and version control lets you change it safely. **Outcome:** A process that used to depend on someone remembering it now runs on its own, with every run logged and reversible. ## Describe the flow once. Access just happens. Every AI tool has its own access policy. Here, IT writes the whole process in one paragraph, and the Agent Builder builds the agent: the trigger, the approvals, the groups, the expiry, and even a token-limit flow. **In about two and a half minutes, you'll see:** - **One paragraph, a complete agent**: the policy, the approval, the group, and the expiry each become a step - **Gaps filled for you**: free seats first, and access that cleans itself up - **The right policy for each person**: a designer, a contractor, and a token limit raised before it is hit - **Zero tickets**: every request resolved in Slack in minutes ## What running AI agents actually looks like Marcè Ayala, Senior Manager of IT Infrastructure at Mammoth Brands, opens up their live Harmony workspace and shows how the team audits its agents. That means checking which runs succeeded, which failed and why, and the run log that proves a request was handled today. **In under 2 minutes, you'll see:** - **64 agents in production** across service, autonomous and custom - **Custom agents built by several teams**, not only IT - **One license-request agent**, from workflow canvas to today's run history > I go back and look at the success rate of each agent. > > - Marcè Ayala, Senior Manager of IT Infrastructure, Mammoth Brands ## Fireblocks built a production access agent in 2 days Fireblocks runs 24 agents on Harmony, many of them built by its own IT and security team. One went from a spec written in Slack to a production application-access agent in two days and has since completed 269 runs. Across the platform, agents have processed 915 application access requests, and a third of 4,800+ employee conversations resolve with no ticket at all. > Harmony took over all of our simple tickets. My team doesn’t see them anymore, and the engineers finally spend their day on the work that actually needs them. > > - Meir Grinberg, Global IT Manager, Fireblocks ## Building agents in Harmony vs. scripts and point automations Most IT teams already automate. The question is what it costs to build, change and trust each automation. | | Scripts and point automations | Harmony Agent Builder | | --- | --- | --- | | Who can build | Engineers, or whoever wrote the original script | IT, HR and finance teams, in natural language, with no code | | Time to production | Weeks of scoping, coding and review | Minutes to a first version; Fireblocks shipped a production agent in two days | | Guardrails | Written into each script, and different in every one | Inherited from your approval policies, access controls and audit requirements | | Reach | One system per script, glued together with cron | Any API, MCP server, webhook, database or SaaS app, orchestrated in one run | | Testing | In production, on real employees | Sandbox simulations, then a controlled rollout to a single team | | Change management | Someone edits the script and hopes | Versioning, one-click rollback and analytics per agent | | Where it runs | A backend job nobody sees | In Slack, Teams, email and the portal, or headless in the background | ## Start from an agent that already works: - [Employee Onboarding](https://harmony.io/agents/new-hire-onboarding) - [Employee Offboarding](https://harmony.io/agents/employee-offboarding) - [Account Lockouts & MFA Reset](https://harmony.io/agents/account-lockouts-mfa) - [Application Access Request](https://harmony.io/agents/application-access-request) - [Temporary Admin Access](https://harmony.io/agents/temporary-admin-access) - [Low Storage Device Detection](https://harmony.io/agents/low-storage-device-detection) ## Give your agents the systems they need to act in - [Slack](https://harmony.io/integrations/slack) - [Microsoft Teams](https://harmony.io/integrations/microsoft-teams) - [Okta](https://harmony.io/integrations/okta) - [Microsoft Entra ID](https://harmony.io/integrations/microsoft-entra-id) - [Google Workspace](https://harmony.io/integrations/google-workspace) - [JumpCloud (IdP)](https://harmony.io/integrations/jumpcloud-idp) - [Jira](https://harmony.io/integrations/jira) - [GitHub](https://harmony.io/integrations/github) ## FAQ ### What is Harmony's Agent Builder? Harmony’s Agent Builder lets IT teams build, configure, and deploy autonomous AI agents and custom workflow automations without writing code, in natural language, combining a library of pre-built agents for common IT use cases with a visual workflow builder for any connected system. ### What kinds of built-in AI agents does Harmony offer? Harmony’s built-in agent library spans Employee Lifecycle agents like onboarding, offboarding, and account termination, Access Control agents like password reset and MFA reset, Software Management agents like application access requests and license expiration monitoring, and Hardware Management agents like low storage and end-of-life asset detection. ### What systems can Harmony's workflow builder take actions in? Harmony’s Custom Workflow Builder can trigger actions in any HTTP-accessible API. Popular examples are Slack, Microsoft Teams, Okta, Entra ID, Google Workspace, JumpCloud, Active Directory, ServiceNow, Jira/JSM, GitHub, Kandji, Jamf, and Intune. ### How do triggers and logic work in Harmony's Custom Workflow Builder? The Custom Workflow Builder supports Manual, scheduled cron-based, event-based, and webhook triggers, combined with If/Else branching, For Each loops over up to 20 parallel items, and durable Sleep timers that survive server restarts. ### Can Harmony support multi-tenant environments for MSPs? Yes. Harmony’s AI agents can be configured for multi-tenant environments, allowing MSPs and large enterprises to run the same agent across multiple tenants from a single configuration. ### What are the main benefits of Harmony's Agent Builder? Harmony’s Agent Builder lets IT teams replace manual, repetitive processes with pre-built agents for onboarding, access control, and hardware monitoring, without writing code. The visual Custom Workflow Builder extends this to any process across integrated systems, and multi-tenant support lets MSPs scale the same automation across every client from a single configuration. ### Can we test an agent before employees see it? Yes. Every agent can run in a sandbox with simulated conversations and dry runs, then be released to a single team or department before company-wide deployment. Versioning and one-click rollback mean a change that misbehaves is undone rather than rebuilt, and agent analytics show success and resolution rates per version. ### How is this different from a general-purpose automation tool? General automation tools connect APIs. They have no idea who is asking or what your policy allows, so every workflow re-implements eligibility, approvals and logging. Agents built in Harmony run on the same organizational context as the rest of the platform: the employee record from your HRIS, devices from MDM, access from your identity provider and the rules in Compliance & Policies. An access agent can therefore decide eligibility, request the right approval and log the decision, and it inherits the platform’s audit log instead of needing its own.