> For the complete documentation index, see [llms.txt](https://docs.blockbrain.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.blockbrain.ai/for-users/durable-agents.md). # Durable Agents An answer in Blockbrain is no longer tied to the browser tab that asked for it. Refresh the page, close the tab, switch to another chat or lose your connection: the agent keeps working on the server. When you come back, the answer is waiting for you or still being written in front of you, and if you were in another chat you are told the moment it is ready. Leaving a chat never ends an answer. Only **Stop** does. Blockbrain switches this on for your organisation, one organisation at a time. If your answers still stop when you leave a chat, it is not switched on for yours yet — there is nothing to change in your own settings, and your Blockbrain Customer Success Manager can tell you when it is planned. ### What changes for you Nothing new to set up and nothing new to click. The difference shows the moment you leave a chat while an agent is still answering: | What you do | What happens | | ----------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Refresh the page mid-answer | Your question is there straight away. A few seconds later the answer so far fills in, and it carries on writing in the same chat. | | Close the tab, come back later | The finished answer is in the chat. If the agent is still working, you rejoin the live answer. | | Switch to another chat | The agent keeps working, and the chat carries a dot in the sidebar from the moment you leave until you open it again. You get a notification when the answer is ready, or when the agent needs your input. | | Lose your connection, close the laptop, go through a tunnel | The answer keeps running on the server. Open the chat again when you are back: the answer so far loads and, if it is still being written, carries on. | | Press **Stop** | The answer ends immediately. What was written before the cut is kept. | This applies to any chat in the Blockbrain web application where an Agent is answering — in Company GPT and in a Data Room. A chat that talks to a model alone, with no Agent selected, is not affected. The Blockbrain assistant embedded in another website, and the Word, Outlook and PowerPoint add-ins, do not show you a running answer again and do not notify you; an answer one of them started still keeps running on the server until it finishes or you stop it. ### Leaving does not cancel the answer Before, an answer belonged to the tab that asked for it. Closing the tab, refreshing or a dropped connection lost it — the text you had watched stream never reached the chat — and a long answer could leave the chat stuck. Now the agent run lives on the server. It keeps going until it finishes, until it needs something from you, or until you press **Stop**. Nothing you do in the browser ends it: not a refresh, not a closed tab, not a lost connection, not a switch to another chat. Very occasionally an answer is interrupted on our side, for example while we update the platform; then no answer arrives and you can simply ask again. Your question is written into the chat the moment you send it, so a chat you reload mid-answer shows what you asked rather than an answer with no question above it. {% embed url="" %} ### Coming back to a running answer Open the chat again and Blockbrain checks whether an answer is still being generated there. If it is, the chat first shows your question on its own, then the answer so far fills in — a few seconds, usually — and carries on in front of you. You do not need to reload, and you do not need to ask again. The **Stop** button is available as before; an answer you rejoined is still yours to stop. {% embed url="" %} ### Being told when an answer is ready If you leave a chat while the agent is still working, Blockbrain keeps an eye on that chat for you: * **When the answer is finished**, a notification appears wherever you are in Blockbrain: *Response ready in \[chat name]*, with an **Open** link that takes you straight there. The notification closes itself after a few seconds — the dot in the sidebar stays until you open the chat. * **When the agent needs you**, for example to approve an action or answer a question, you get *The agent is waiting for your input in \[chat name]* instead, also with **Open**. * **In the sidebar**, the chat carries a small dot from the moment you leave it until you open it again: * **a grey dot that pulses** — the agent is still working; * **a blue dot** — the answer is finished; * **an orange dot** — the agent is waiting for your input. Opening the chat clears the dot. {% embed url="" %} *The agent needs an answer: the notification arrives, the chat stays marked in the sidebar, and opening it shows the question waiting for you* {% embed url="" %} Good to know: * The chats Blockbrain is watching for you survive a page refresh, and so do the dots in the sidebar. A notification that was already on screen is not shown again after a reload — the dot is how you find the chat. * A Blockbrain tab in the background does not notify you. The notification appears in the tab you are looking at; if you are in another application altogether, it appears when you come back to Blockbrain. * If you left a chat while an approval or question card was already on screen, Blockbrain does not watch that chat: there is no notification and no dot. Open the chat when you are ready — the card is still there. * If an answer finished while you were completely offline, Blockbrain catches up when you reconnect, so you find the finished chat rather than a stuck one. * Notifications are shown in the app only. There is no e-mail, push or desktop notification. If you close the browser entirely, you find the finished answer in the chat on your next visit. * A chat is named after its first answer finishes, so a notification about a chat’s very first answer says *Response ready in New Chat*. ### Stopping an answer **Stop** is the only way to end an answer before it finishes. **Stop** ends the answer straight away. What was written before the cut stays in the chat, appears once, and survives a reload. The message box is ready again straight away, so you can type your next message. Stop works after a refresh too. An answer that keeps running after you leave uses AI processing like any other answer. If you no longer need it, press **Stop** before leaving the chat. ### Approvals and questions When an agent asks you to approve an action or asks you a question, the card appears in the chat right away, without a refresh. Answering it continues the same run. Refreshing while a card is on screen leaves the card where it is. Refreshing after you answered brings you back to the answer continuing, not to a card you already dealt with. A paused answer waits for you for up to 7 days. After that it can no longer be continued. Send your request again as a new message rather than answering the card that is still on screen. ### Tips * Give an agent a long task, switch to another chat and carry on working. The notification tells you when it is done. * Check the sidebar for dots when you come back from a break: blue means an answer is ready, orange means the agent is waiting for your input, grey means it is still working. * Stop what you do not need. An answer you walk away from keeps running until it finishes or you stop it. * If you were away for more than about an hour, you may not see a notification, but the finished answer is in the chat. ### Troubleshooting | Symptom | What to do | | --------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | I refreshed and the answer was still writing | This is expected. After a reload the chat may show only your question for a few seconds, then the answer so far appears and continues writing. | | The answer kept going after I closed the tab and I did not want it to | Only **Stop** ends an answer. Press Stop before leaving the chat. | | I did not get a notification | Notifications only appear for chats that were still generating when you left, and only inside the Blockbrain app. If an approval or question card was already on screen when you left, no notification is sent. Open the chat: the answer or the card is there. | | The dot in the sidebar will not go away | Open the chat once; that clears it. If it stays, report it to Blockbrain support. | | I got the notification but the chat looks unchanged | Open the chat. The finished answer loads when you open it. | | My question is in the chat but there is no answer | Reload once. If the chat still shows no answer and nothing is being generated, report it to Blockbrain support with the chat name. | | My answers still stop when I leave the chat | This has not been switched on for your organisation yet. There is nothing to switch on in your own settings; your Blockbrain Customer Success Manager can tell you when it is planned. | --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://docs.blockbrain.ai/for-users/durable-agents.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.