> ## Documentation Index > Fetch the complete documentation index at: https://docs.every.ai/llms.txt > Use this file to discover all available pages before exploring further. # Every Agent > Your AI assistant that reads and acts across your whole workspace — clients, invoices, proposals, payments, bookings, and more — with confirm-before-send safety and one-click undo ## Overview Every Agent is Every's conversational assistant. Instead of clicking through menus, you describe what you need in plain language and the agent creates, updates, analyzes, and drafts across your workspace. It is not a chatbot bolted onto the side of the app — it operates the same data you do, through a large and continually growing toolset, with the same permissions and the same records. **What makes it useful:** * It works directly with your live business data in Every * It uses your saved [Business Context and Voice & Tone](/ai/context-and-voice) so its work sounds like you * It [remembers facts and applies skills](/ai/memory-and-skills) you've taught it across every chat * It connects to external systems like Google Workspace, Xero, HubSpot, and Granola * It handles multi-step workflows without you mapping every click Start chatting with Every Agent in the app ## What the Agent Can Reach Every Agent works directly with your data in Every. It can read, create, update, and (with confirmation) delete across: | Area | What the agent can do | | ---------------------- | ------------------------------------------------------------------------------------ | | **Clients** | Create and update clients, enrich company records, analyze concentration and revenue | | **Contacts** | Create and link contacts; import from Google, LinkedIn, or CSV | | **Invoices** | Draft, edit, apply tax and discounts, schedule sends, send to clients | | **Proposals** | Draft structured proposals, track approval, convert approved proposals into invoices | | **Payments** | Record manual payments, send reminders and confirmations, check Stripe-linked state | | **Expenses** | Parse receipt uploads into expenses, add unbilled expenses to invoices | | **Time** | Log time entries, add unbilled time to invoices | | **Bookings** | Check upcoming bookings, create booking records | | **Recurring invoices** | Create, pause, and update recurring invoice automations; review run history | It also runs reporting and analysis on top of all of this — revenue summaries, overdue and cash-flow reviews, client concentration, and service or product trends. The agent's toolset is large and grows continuously as new capabilities ship, so think of it less as a fixed menu and more as: *if you can do it in Every, you can usually ask the agent to do it.* **Ask Every Agent:** * "Create an invoice for ACME Corp for \$5,000 for consulting services" * "Show me all overdue invoices and draft reminders" * "Find the HubSpot deal for Acme and summarize what stage it's in" * "Pull the action items from my Granola meeting with Sarah last week" ## How Every Agent Works Every Agent operates in **direct execution mode**: you describe the task, the agent figures out the steps, performs them, and shows or confirms the result. You keep refining conversationally. ```text theme={"system"} You: "Create an invoice for ACME Corp for $5,000" Agent: [Shows draft] You: "Actually make it $5,500 and add a thank-you note" Agent: [Updates the draft] You: "Perfect, send it" ``` ### Confirm Before It Sends or Deletes The agent acts on your data, so the actions that **leave your workspace or remove data are gated**. The agent shows you what it's about to do and waits for your go-ahead before: * **Sending** anything to a client — invoices, proposals, payment reminders, confirmations, and emails * **Deleting** records — clients, contacts, invoices, proposals, payments, services, and more Reads, drafts, and internal edits happen directly so the conversation stays fast; the irreversible, client-facing steps are the ones that ask first. Always double-check invoice and email details before confirming a send. Once an email is delivered to a client, it cannot be unsent. ## Navigation You don't have to know where a page lives. Ask the agent to take you somewhere and it can navigate the app for you — "open the payments page," "take me to my overdue invoices," "show me ACME's client record." This pairs well with analysis: ask a question, then jump straight to the records behind the answer. ## Voice Input Every Agent supports voice input. Click the microphone icon in the chat input, speak your request, and your speech is transcribed and sent as a normal message. **Useful for:** * Logging time while working * Capturing expenses quickly * Sending short requests from mobile ## Context, Memory & Skills The agent gets better the more it knows about your business. Three layers shape its work: Teach the agent what your business does and how it should sound when writing on your behalf The agent remembers facts across chats and applies the reusable skills you create ### Learned Preferences Beyond freeform memory, the agent can save specific, standing business rules. **Examples:** * "ACME Corp always gets Net 15 terms" * "Never apply tax to consulting services" * "Enterprise clients default to 45-day terms" * "Always include project reference numbers on TechStart invoices" When the agent proposes saving a rule, it asks you to confirm. Manage saved rules in [**Agent → Preferences**](https://every.ai/agent/preferences). Learned Preferences are standing rules ("always do X"). [Agent Memory](/ai/memory-and-skills) is freeform recall (running lists, observations, open todos). [Business Context](/ai/context-and-voice) is the narrative of what your business is. They work together. ## File Processing Every Agent can read supported uploads and help you work with them: ```text theme={"system"} "Extract the client list from this spreadsheet" "Read this contract and tell me the payment terms" "Import this CSV of historical invoices" "Log this receipt" ``` **Supported file types include:** PDF, Word documents, Excel spreadsheets, CSV files, images, and plain text files. See supported import formats, file handling, and batch workflow details ## Batch & Multi-Step Workflows The agent can apply changes across many records at once, and break a goal into ordered steps. ```text theme={"system"} "Update all enterprise clients to Net 45" "Archive clients with no invoice activity in the last 2 years" "Change tax rate to 13% on all consulting services" ``` ```text theme={"system"} "Set up my consulting business for 2026: 1. Create a new service for strategy work at $175/hour 2. Update my recurring automations to use the new rate 3. Draft a rate change note for retainer clients" ``` ## Undoing Agent Actions Every Agent includes a built-in undo system. When the agent creates or modifies data, an **Undo** button appears below its message so you can reverse the action if something is wrong. ### How It Works * A yellow **Undo** button appears inline after any data-modifying action * Hover to preview which items will be affected (they highlight in yellow) * Click to reverse **all actions from that agent turn** (it's all-or-nothing to prevent inconsistent states) * You can also undo from the **Activity Timeline** on the Home dashboard ### What Can Be Undone Most data operations can be reversed: * **Invoices & proposals** — creation, updates, status changes, line item additions * **Payments & time tracking** — payment records, time entries, expense records * **Clients & contacts** — creation, updates, contact linking * **Complex operations** — multi-step workflows, bulk operations, CSV import results ### What Cannot Be Undone * **Emails already sent** to clients (invoices, notifications) * **Third-party actions** already processed (Stripe payments, Google Calendar events, HubSpot syncs) * **Manually edited data** — if you edited the record after the agent action, undo is blocked to protect your changes ### Limitations * Only the **most recent agent turn** can be undone — once the agent performs a new action, the previous undo buffer is replaced * After undoing, you cannot "redo" — ask the agent again instead * If a conflict is detected (manual edits, external processing), the system blocks undo and shows an explanation If you notice an error, undo immediately before the agent performs more actions. It is easier to undo right away than after additional turns. **You:** "Create a \$5,000 invoice for ACME Corp for consulting services" **Agent:** "I've created invoice INV-001 for \$5,000..." *Oops, should be \$6,000.* Click the **Undo** button. The invoice is deleted. Then ask: **You:** "Create a \$6,000 invoice for ACME Corp for consulting services" **You:** "Record a \$5,000 check payment for invoice INV-001" **Agent:** "I've recorded the payment and marked invoice INV-001 as paid" *Wrong amount.* Click **Undo**. The payment record is deleted and the invoice status reverts to its previous state. **You:** "Record a \$5,500 check payment for invoice INV-001" **You:** "Archive all paid invoices from 2023" **Agent:** "I've archived 47 invoices from 2023" *Wrong year!* Click **Undo (47 actions)**. All 47 invoices are unarchived. **You:** "Archive all paid invoices from 2022" ## Activity History Every action the agent takes is recorded in the **Activity Timeline** on your Home dashboard. Use it to review what the agent did, see which records changed, and undo a past turn even after you've moved on in the conversation. ## A Note on Credits AI usage consumes credits — agent conversations, file processing, and scheduled runs all draw from your balance. See [What is a credit?](/plans/what-is-a-credit) and [Credit usage & errors](/plans/credit-usage-and-errors) for how that works. ## Getting Started 1. Ask a simple question like "Show me my last 5 invoices" 2. Try an action like "Create a draft invoice for \$100" 3. Add [Business Context and Voice & Tone](/ai/context-and-voice) 4. Create a [skill](/ai/memory-and-skills) for a workflow you repeat 5. Connect Google Workspace and any agent integrations you use ## Next Steps Teach the agent how your business works and sounds Persistent memory and reusable, auto-applied skills Extend the agent's reach with external connections and MCP See example workflows and prompts The better your Business Context, Voice & Tone, skills, and integrations are set up, the more useful Every Agent becomes.