# Actions: let your Agent do things, not just talk · Communicate > Communicate Actions let your AI Agent look up an order, check a subscription, update a ticket, or call your own API, mid-Conversation, inside the permissions and confirmation rules you set. Start with 100 credits. - **URL:** https://communicate.so/actions --- # Let your Agent finish the simple work you allow. Approved work, completed in the Conversation Add read-only lookups, support requests, forms, buttons, or your own API behind explicit permissions and review rules. [Start with 100 credits](https://app.communicate.so)One-time $1 activation · 100 credits included · No automatic subscription A support task is reviewed and added to the Agent’s allowed actions.Communicate Approved tasks Restricted Look up an account Read-only customer lookup Allowed Create a support request Send the question and Conversation to your support queue. Reviewing Change billing details Requires a Member Blocked The Agent can run only the tasks marked Allowed. How it works ## Define the task before the Agent can run it. Actions stay predictable because inputs, permissions, and the customer-facing result are configured together. - 1 01 · Choose ### Start from a clear support task Use a template or define one narrow Action such as looking up an account or creating a request. Action Draft Task Create support request Input Conversation and email Result Ticket reference - 2 02 · Restrict ### Set what the Agent may send and receive Keep the scope narrow, mark required fields, and leave sensitive mutations with a teammate. Permissions Restricted Account lookup Read only Create request Allowed Billing change Blocked - 3 03 · Observe ### Review the run and its result The Action state and result stay connected to the Conversation for your team to inspect. Run result Completed Customer Context included Support request Created Conversation Updated Action Draft Task Create support request Input Conversation and email Result Ticket reference Permissions Restricted Account lookup Read only Create request Allowed Billing change Blocked Run result Completed Customer Context included Support request Created Conversation Updated Approved tasks finish in the Conversation. Everything else stays with your team.Approved tasks finish in the Conversation. Everything else stays with your team. What changes ## Resolve the request where the customer already asked it. A useful Agent does more than acknowledge a task. It either completes the allowed work or brings in the right person. 01 ### Simple work ends in a real result Lookups and approved support tasks return their outcome in the Conversation instead of creating another queue. 02 ### Permission stays visible Allowed, blocked, and review-required work remain explicit for the team configuring the Agent. 03 ### Your systems remain the source of truth A custom API Action can use structured inputs and return a controlled response without giving the Agent open access. A support task is reviewed and added to the Agent’s allowed actions.Communicate Approved tasks Restricted Look up an account Read-only customer lookup Allowed Create a support request Send the question and Conversation to your support queue. Reviewing Change billing details Requires a Member Blocked The Agent can run only the tasks marked Allowed. Designed for the edge cases ## Narrow Actions are safer and easier to trust. Each Action should do one understandable job with a result your team can verify. ### Task templates Begin with common support work such as account lookups, request creation, lead capture, or a custom form. You stop dealing with: starting every support automation from an empty schema. ### Allow and block rules Keep read-only tasks separate from mutations and reserve sensitive changes for a teammate. You stop dealing with: granting broad access just to automate one small task. ### Structured API calls Define the endpoint, required inputs, and returned fields instead of passing free-form customer text downstream. You stop dealing with: letting unvalidated conversation text reach an internal system. ### Observable outcomes Keep the request, state, and result attached to the Conversation your team already reviews. You stop dealing with: chasing an automation log in a separate tool. Control before automation ## The allowed task is visible before it runs. Communicate keeps the permission boundary close to the customer request. Input ### Required information is explicit The Action defines the fields it needs instead of guessing from incomplete context. Access ### The permission is narrow Read, create, and blocked work stay separate. Result ### The outcome returns to the thread The customer and teammate see what happened without changing tools. Choose how work gets completed ## Approved Actions without an open-ended automation layer. Compare the control your support team keeps around each task. ### Task result CommunicateInside Conversation Macro or linkCustomer leaves flow Custom automationCustom UI ### Permission model CommunicatePer Action Macro or linkNo system access Custom automationYou build auth ### Structured inputs CommunicateRequired fields Macro or linkFree-form Custom automationYou validate ### Human fallback CommunicateSame Inbox Macro or linkNew ticket Custom automationYou build routing Decision Communicate Macro or link Custom automation Task result Inside Conversation Customer leaves flow Custom UI Permission model Per Action No system access You build auth Structured inputs Required fields Free-form You validate Human fallback Same Inbox New ticket You build routing [Start with 100 credits](https://app.communicate.so) Questions before you start ## Questions about what an Agent may do. Keep the permission model narrow and understandable. ### Does every Action run automatically? ### Can an Action call my API? ### Can I keep billing changes human-only? ### Where can my team see what happened? ## Turn a repeated support step into an approved Action. Choose a task your team already handles the same way, set its boundary, and test the result. [Start with 100 credits](https://app.communicate.so)One-time $1 activation · Grounded answers · Human takeover · Cancel anytime