Agent Tasks, Chat & Artifacts

🛠️ Agent Hub SEO Task Stalling or Stopping Early

🔍 Overview Some customers have reported that the Fix Critical On-Page SEO Issues task in Agent Hub runs for approximately an hour, stops after completing only the first step, and generates an unusually large amount of text output. If this has happened to you, the most likely cause is a temporary outage with one of the AI providers that powers the platform — not an error in your project or settings. ⚙️ Why This Happens Agent Hub SEO tasks rely on third-party AI providers to process and complete each step in the workflow. When one of those providers experiences instability or an outage, the task can stall mid-run, time out, or behave unexpectedly — including stopping after the first step and producing excessive text output. These outages are temporary and are resolved by our engineering team as quickly as possible. Our team has confirmed that this type of stalling behaviour is a known issue during AI provider outages and has been addressed in previous engineering updates to improve how the agent handles multi-step workflows. ✅ What To Do 1. Wait a short time before retrying. AI provider outages are usually temporary. Waiting 30–60 minutes before trying again gives providers time to recover. 2. Re-run the task from scratch. Once you believe enough time has passed, navigate back to Agent Hub and start the Fix Critical On-Page SEO Issues task again. Do not attempt to resume a stalled run — start a new one. 3. Monitor the task closely on your next attempt. If the task completes all steps successfully this time, the provider issue has been resolved. If it stalls again in the same way, the outage may still be ongoing. 4. Check whether the issue persists across multiple attempts. If the task consistently fails over several hours or the next day, it is no longer likely to be a temporary outage and you should contact support. 🚩 Signs That This Is an AI Provider Issue - The task runs for a long time (around an hour) before stopping. - Only the first step completes; subsequent steps do not run. - The output contains an unusually large or repetitive amount of text. - The same task has worked correctly for you in the past. - The issue resolves itself after waiting and retrying. 💡 Tips to Avoid Confusion - Do not run the task multiple times simultaneously. Starting duplicate runs during a provider issue will not help and may compound the problem. - Do not change your project settings in response to this error — the stalling is not caused by your configuration. - Give it at least a day before escalating if retries continue to fail, as some provider outages can last several hours. 📞 Still Need Help? If you have waited, retried, and the task is still stalling or stopping early, our team is ready to investigate further. If you need further assistance, open the chat widget in the bottom-right corner of the platform and type human teammate to be connected with a member of our team.

🛠️ Troubleshooting Atlas Agent 400 Error on Hero Image Prompts

This article addresses the request failed: 400 error that can appear in Atlas Agent when submitting a hero image prompt. A 400 error means the request was rejected before it was processed. Before reaching out to our team, try the self-serve steps below — many users are able to resolve this by adjusting their prompt. 🔧 Self-Serve Steps to Try First 1. Retry the prompt: Occasionally a transient issue can cause a 400 error. Submit the exact same prompt again to see if it succeeds on a second attempt. 2. Simplify the prompt: Shorten your hero image prompt to the core description, removing any extra instructions, punctuation, or formatting. A very long or complex prompt can sometimes trigger a rejection. 3. Check for unsupported characters: Remove any special characters (e.g., /, &, <, >, quotes within quotes) from the prompt text and try again. 4. Try a different prompt entirely: Write a new, plain-language description of the hero image you need. If this succeeds, the issue is specific to something in your original prompt wording. 5. Reload the page and try again: Refresh your browser, navigate back to the Atlas Agent project, and resubmit the prompt to rule out a session-level issue. 📋 If the Error Persists — What to Prepare Before Reaching Out If none of the steps above resolve the error, our team will need to investigate on a case-by-case basis. Please have the following ready: - Your project name: The exact name of the project where the error occurred. - The exact error message: Copy the full error text as it appears on screen (e.g., request failed: 400). - The hero image prompt you submitted: Include the full text of the prompt that triggered the error. - Timestamp: Note the date and approximate time when the error occurred. - Steps you already tried: Any actions you took before or after the error appeared, including which self-serve steps above you attempted. ✅ Next Steps Having the information above ready will allow our team to identify the cause and apply the correct fix without unnecessary back-and-forth. If you need further assistance, open the chat widget in the bottom-right corner of the platform and type human teammate to be connected with a member of our team.

💬 Managing Live Chat Sessions and Workflow Messages

🗺️ Overview If you are setting up or testing a new workflow for a client, it is important to know how to properly close an active chat session and validate that your messages are working as expected. This article walks you through both actions so your workflow testing runs smoothly. 💬 How to Close an Active Chat Session When you have finished a conversation in the live chat widget — whether it was a real support interaction or a test session — you can close it at any time. Follow these steps: 1. Locate the live chat widget in the bottom-right corner of the Search Atlas platform. 2. Open the active chat session you want to close. 3. Look for the close or end conversation option within the chat window (usually a small X icon or a menu option inside the widget). 4. Confirm the closure when prompted. The session will be marked as resolved and removed from your active queue. Note: Closing a chat session does not delete the conversation history. Previous messages remain accessible for reference. 🛠️ Testing a New Workflow Message for a Client Before launching a workflow message to a client, it is good practice to test it internally to make sure the content, formatting, and triggers behave as expected. Here is how to do that: 1. Navigate to Coworker using the left sidebar, or go directly to /coworker in your browser URL. 2. Select the workflow or message sequence you are building or editing for the client. 3. Use the built-in preview or test option within the workflow editor to simulate how the message will appear to the recipient. 4. Review the message content, variables, and any conditional logic to confirm everything displays correctly. 5. If the workflow includes automated triggers, verify that the trigger conditions match the intended client scenario before activating. 6. Once validated, save and activate the workflow for the client. ✅ Best Practices for Workflow Validation - Use a test account or sandbox client profile whenever possible to avoid sending unfinished messages to real clients. - Label test sessions clearly so team members know a conversation is part of a workflow test and not a live client issue. - Close test chat sessions promptly after validation is complete to keep your support queue clean and accurate. - Document your workflow changes with a brief internal note so your team stays aligned on what was updated and why. ❓ Frequently Asked Questions Can another team member close a chat session I started? Yes, any team member with the appropriate permissions can close an open session from within the platform. Will closing a test chat affect the client's live workflow? No. Closing a chat session only resolves that specific conversation. It does not modify or deactivate any workflow you have set up for the client. What if the workflow message does not trigger correctly during testing? Double-check the trigger conditions and any dynamic variables inside the Coworker workflow editor. Make sure all required fields are filled in and that the client profile used in the test matches the expected trigger criteria. 🙋 Need More Help? If you need further assistance, open the chat widget in the bottom-right corner of the platform and type human teammate to be connected with a member of our team.

📁 How Atlas Brain Delivers Files and Assets

🤖 How Atlas Brain Actually Shares Files Atlas Brain is an AI assistant built into Search Atlas. It can help you research, plan, and create content — but it does not generate downloadable file attachments or display file cards inside the chat interface. If Atlas Brain described a file attachment card or a download link inside the chat window (for example, a file labeled modern_heating_square_logo.svg), that element does not exist. No attachment cards, file panels, or out-of-band download mechanisms are present in the Search Atlas chat interface. This was an error in the AI response, and we apologise for any confusion it caused. 📤 Where Exports and Downloads Actually Live Depending on what you need, downloadable assets and exports are available in specific sections of the platform — not inside the Atlas Brain chat. The exact location of download links or export options will vary based on the type of content you are working with. If you are unsure where to find a specific file or export within the platform, our support team can help identify the correct location for your use case. 🖼️ What To Do If You Need a Specific Logo or SVG File Atlas Brain does not store, generate, or deliver image files such as SVG logos. If you are looking for a logo or brand asset, here are your options: 1. Check the original source where the file was created or saved — for example, your design tool, email, or cloud storage. 2. If the file was part of a Search Atlas export, navigate to the section of the platform where that content was produced and look for any available download or export option within that section. 3. If you are unsure whether a file should have been generated as part of a Search Atlas workflow, or you cannot locate a download link, have ready your project name, the exact file name or type you expected, and a description of the Atlas Brain response you received (including any timestamp if possible), then contact our support team for help. ✅ What Atlas Brain Can Help You With Atlas Brain is designed to assist with text-based tasks inside the platform. Here is what it reliably supports: - Drafting and refining written content such as press releases, blog posts, and outreach copy - Answering questions about your SEO strategy and campaign performance - Guiding you to the right section of Search Atlas for a specific task - Summarising data and making recommendations based on your project context Atlas Brain does not generate image files, create downloadable attachments, or simulate UI elements that are not part of the platform. 💬 Need More Help? If you need further assistance, open the chat widget in the bottom-right corner of the platform and type human teammate to be connected with a member of our team.

🔧 Fix Atlas Agent 'I Need a Message' Error

What Is This Error? When you type a message into Atlas Agent and see the response "I need a message to respond" instead of a normal reply, the agent is not recognising your input. This is a known issue where Atlas Agent's text input becomes blocked, preventing normal interactions. What Causes This Error? This error has been identified as a persistent blocker affecting text input in Atlas Agent. When it occurs, every message you type triggers the same error response, with no successful interactions possible. This is not something that can be resolved through browser-level adjustments on your end — it requires review and action by our team. What to Do When You See This Error If Atlas Agent is returning "I need a message to respond" to every message you send, contact our support team right away. To help us resolve this as quickly as possible, please have the following information ready before you reach out: - Your project name or workspace name - The exact error message you are seeing - The date and time the error first appeared - Confirmation of whether the error occurs on every message you send If you need further assistance, open the chat widget in the bottom-right corner of the platform and type human teammate to be connected with a member of our team.

📥 Find and Download Atlas Brain File Links

🔍 Overview Atlas Brain can generate or reference files — such as SVG logos, images, or documents — during a chat session. In some cases, the download link for these files may not appear clearly in the chat, or clicking it may not trigger a download. This article explains how to locate and access those files. 💬 How Atlas Brain Shares Files When Atlas Brain produces a file as part of a response, it typically shares a download link directly in the chat window. The link may appear as a clickable file name, a button, or a URL embedded in the message. However, the link is not always visually prominent, and in some sessions it may not render correctly. Common reasons a file link may seem missing or inaccessible: - The link blends into the surrounding text and is easy to overlook. - The chat session was refreshed or navigated away from before the file rendered fully. - A rendering issue prevented the link from displaying in that session. - The file was generated in an earlier message that has scrolled out of view. 🛠️ How to Find and Download the File 1. Open the Coworker page. Go to the left sidebar and select Coworker (URL: /coworker). This is where Atlas Brain chat sessions are hosted. 2. Locate the correct chat session. If you have multiple sessions, make sure you are in the session where Atlas Brain referenced the file. 3. Scroll up through the conversation. File links are sometimes embedded earlier in the response thread. Scroll up carefully to check all messages. 4. Look for subtle link formatting. The file name (for example, modern_heating_square_logo.svg) may appear as a hyperlink. Try hovering your cursor over the file name text to check whether it is clickable. 5. Click the file name or link. If the link is active, your browser will either open the file directly or prompt you to save it. If your browser opens it in a new tab instead of downloading, right-click the file and select Save As to save it to your device. 6. Try regenerating the file. If you cannot find the link at all, send a new message in the same Atlas Brain session asking it to regenerate or re-share the file. For example: "Can you share the download link for the SVG file again?" Atlas Brain will attempt to reproduce and re-share the file in a new message. ⚠️ Known Limitations There is a known issue where file previews and download links may not render correctly in certain Atlas Brain sessions, particularly for image and SVG file types. If you have tried the steps above and still cannot access the file, this may be caused by a temporary rendering issue in your current session. As a workaround, try the following: - Refresh the page and return to the same session to see if the link re-renders. - Start a new Atlas Brain session and ask it to recreate the file from scratch. - Use a different browser or clear your browser cache, then reload the session. ✅ Tips for Smoother File Downloads - Do not navigate away from the chat while Atlas Brain is still generating a file — wait for the full response to appear before scrolling or switching pages. - If Atlas Brain references a file by name but provides no visible link, always try asking it to re-share the link in the same session before starting over. - Google Chrome and Microsoft Edge tend to handle in-browser file downloads most reliably. 🙋 Still Need Help? If you have followed all the steps above and still cannot access or download your file, If you need further assistance, open the chat widget in the bottom-right corner of the platform and type human teammate to be connected with a member of our team.

🧠 Finding Atlas Brain File Download Links

🔍 Overview When Atlas Brain, the Search Atlas AI Agent, references or generates a file during a conversation — such as a logo, image, or document — customers sometimes find it unclear where to access the actual download link. This article explains exactly where to look so you can retrieve any file Atlas Brain mentions. 💡 How Atlas Brain Shares Files Atlas Brain delivers files directly inside the chat conversation window. When a file is ready, Atlas Brain will display a download link or a clickable file reference inline within its response message. There is no separate downloads page or external storage location — everything stays within your active chat session. 📥 How to Find Your File Download Link 1. Open Atlas Brain: Navigate to the Atlas Brain chat interface inside your Search Atlas platform. 2. Locate the relevant message: Scroll through the conversation to the message where Atlas Brain confirmed the file was created or referenced. For example, if Atlas Brain mentioned a file called modern_heating_square_logo.svg, find the message where it was discussed. 3. Look for the inline link: The download link appears as a clickable link or button directly inside that message — it is not sent separately. It may appear as the file name itself (e.g., modern_heating_square_logo.svg) or as a labelled download button. 4. Click the link to download: Click the file name or download link to save the file to your device. Depending on your browser settings, the file will either download automatically or prompt you to choose a save location. ⚠️ If You Cannot Find the Link If you have scrolled through the conversation and cannot locate the download link, try the following: - Check your chat history: Past sessions are saved in the chat history sidebar. Click the session name where the file was generated to reopen it and search for the link. - Reload the page: Occasionally a page refresh restores links that did not render correctly on first load. - Re-request the file: In a new or existing chat message, ask Atlas Brain to regenerate or re-share the file. Atlas Brain can reproduce the output if given the same context or request. - Check your browser's download history: If you already clicked the link but are unsure where the file saved, check your browser's built-in download history (usually accessible via Ctrl+J on Windows or Cmd+Shift+J on Mac). 🛠️ Tips for Managing Atlas Brain Files - Download files promptly after Atlas Brain generates them — chat sessions may expire or be cleared over time. - Rename sessions using the chat history sidebar so you can easily find past conversations that contain important files. - If Atlas Brain is processing a large request, wait for a completion message before expecting the download link to appear. 🆘 Still Need Help? If you need further assistance, open the chat widget in the bottom-right corner of the platform and type human teammate to be connected with a member of our team. Additional Notes Navigate to the left sidebar and select Coworker (URL: /coworker). If browser opens file in a new tab instead of downloading, right-click the file and select Save As. Also: link may blend into surrounding text and is easy to overlook as a subtle hyperlink — hover cursor over file name text to check if clickable.

🎯 The Role of Prompting in Atlas Agent

Atlas Agent (formerly called Search Atlas Brain) is Search Atlas's chat-driven system for digital marketing execution. Within Atlas Agent, prompting, issuing commands through natural language, is the central mechanism that drives specialized agents to execute complex work, apply your standards, and maintain strategic control at scale. The entire system operates through a chat-driven interface, where you interact with Atlas Agent using a prompt box instead of navigating dashboards, reports, and manual workflows. ⚡ Simplicity and Efficiency of Input Prompting in Atlas Agent is designed to prioritize speed, accessibility, and strategic focus, letting you concentrate on outcomes rather than mechanics. 🚫 No Prompt Engineering Required You do not need to be a prompt engineer to work effectively with the system. Atlas Agent is built to work with basic prompts, simple conversational commands, and even partially submitted inputs. The system translates these into orchestrated execution tasks. More structured prompts may produce more refined results, but functionality is preserved even with minimal input. 🧠 Transparency of Thought Process and Context Prompting activates a built-in transparency layer that helps make AI decisions understandable and contextualized. Every chat interaction can display the agent's thought process, showing how a natural language prompt is translated into a structured work order, and giving you visibility into the logic being applied, the scope of execution, and the assumptions behind decisions. The agent retains context from previous chats. As you provide feedback, preferences, and direction over time, it applies learned preferences automatically and maintains continuity across sessions. 🛠️ Prompting as a Control and QA Mechanism Prompting is not just how work is started, it is how your expertise, judgment, and standards are injected into AI execution. The agent is semantic enough to follow detailed instructions within both prompts and Playbooks. If your agency has defined standards for how content should be written, structured, or reviewed, that guidance can be embedded into Playbooks and followed consistently. You can also specify conditions such as whether content should publish automatically and how often. You can directly instruct the agent to self-review and improve its work, for example by asking it to evaluate the quality of its suggestions and present new ones if the originals are not strong. The agent will present revised suggestions and explicitly state what changed from the previous generation. For complex executions, prompting lets you control how much autonomy the agent receives; you can run a full optimization end-to-end, request samples before deployment, or guide the agent through a custom execution path. 🚀 Prompting as the Fulfillment Orchestrator Once you approve a plan, the agent acts as a fulfillment orchestrator, communicating directly with the system to produce real outcomes rather than task lists. A final command results in changes being deployed to your live website or account. 🔽 Prompt Examples You Can Use Right Away These examples are written in plain language, no special formatting or prompt engineering required. - Show me the biggest SEO opportunities for this site right now. - Auto, show me title tag opportunities. - Suggest new blog topics based on keyword gaps. - Show me optimizations for this Google Business Profile. - Run a fast authority boost. - Generate a performance summary for this project. - Pause and ask for approval before making changes. 🧭 Helpful Tip If you are ever unsure which agent to use, just ask. You can say Which agent should handle this, and the system will guide you. In Atlas Agent, prompting is not just input, it is control, orchestration, and execution combined. It lets you manage systems instead of tasks, apply judgment at scale, maintain transparency and safety, and execute strategy continuously at the speed of AI. If you need further assistance, open the chat widget in the bottom-right corner of the platform and type human teammate to be connected with a member of our team.

🛠️ Fixing the 400 Error in Agent Hub Custom Prompts

🔍 What This Error Means A 400 error in Agent Hub means the request your custom prompt sent could not be processed by the system. It is not a problem with your account or with Agent Hub itself. Instead, it usually points to something in the way the custom prompt is written or configured. This is why playbooks keep working while your custom prompt fails. Playbooks are pre-built and validated by Search Atlas, so their structure, variables, and formatting are already correct. A custom prompt is created from scratch, which means a small formatting or input issue can cause the request to be rejected. ⚙️ Common Causes of the 400 Error Most 400 errors with custom prompts come from one of the following: - Prompt is too long: Very long prompts can exceed the allowed input size and be rejected. Prompts that go over the character limit are a confirmed cause of this error, especially when they contain long or detailed instructions. - Unsupported characters: Pasting from another program can add hidden symbols, smart quotes, or formatting that the system cannot read. - Empty or missing fields: A required field, such as a target URL or keyword, was left blank. - Broken variables: A placeholder or variable was edited incorrectly, so the system cannot fill it in. - Invalid links: A URL in the prompt is misspelled, incomplete, or missing the https:// prefix. 🛠️ How to Fix It Work through these steps in order. Most customers resolve the error within the first few checks. 1. Shorten your prompt: Remove any text that is not essential and keep your prompt within the character limit. Keep instructions clear and direct. 2. Retype instead of pasting: If you copied the prompt from a document or email, retype it directly into Agent Hub to strip out hidden formatting. 3. Check all required fields: Make sure every field the prompt needs is filled in correctly and nothing is left blank. 4. Review your variables: Confirm that any placeholders match the format used in working playbooks. Do not change the surrounding symbols. 5. Verify your links: Make sure each URL is complete, correctly spelled, and begins with https://. 6. Test with a simple prompt: Enter a short, basic prompt to confirm the error is tied to your original wording and not a wider issue. 💡 Tips to Prevent Future Errors A few habits will help your custom prompts run smoothly every time: - Start from a playbook: Use a working playbook as a template, then adjust it gradually to fit your needs. - Build in small steps: Add one instruction at a time and test as you go, so you can quickly spot what causes a problem. - Keep prompts focused: Clear, concise prompts are less likely to trigger errors and produce better results. - Avoid special formatting: Stick to plain text and avoid pasting styled content from other tools. 💬 Still Seeing the Error? If you have tried these steps and the 400 error continues, our team is ready to help. If you need further assistance, click the chat icon in the bottom-right corner of the platform to start a live chat with a member of our team. Sharing the exact prompt you used will help us resolve the issue faster.

💬 Using Search Atlas Coworker in Channels vs. Direct Messages

You can interact with Search Atlas Coworker in two ways inside Slack: via a direct message (DM) to the bot, or by @mentioning it in a channel. Each mode works slightly differently. Direct Messages (DMs) DMs are the simplest way to use Search Atlas Coworker. Open a conversation with the Search Atlas Coworker bot and type your question or request directly. Search Atlas Coworker replies in the same DM thread. Best for: personal research, sensitive data, one-on-one tasks, setting up routines. @Mentioning in a Channel In any channel where Search Atlas Coworker has been added, you can trigger it by starting your message with @Search Atlas Coworker: "@Search Atlas Coworker What were our top 5 ranking pages last week?" Search Atlas Coworker will reply in a thread on your message — keeping the channel clean and the conversation organized. Best for: sharing reports with your team, collaborative analysis, posting results visible to the whole channel. Adding Search Atlas Coworker to a Channel 1. Open the channel you want to add Search Atlas Coworker to. 2. Type /invite @Search Atlas Coworker in the message bar and press Enter. 3. Search Atlas Coworker is now active in that channel and will respond to @mentions. How Replies Work in Channels - Search Atlas Coworker always replies in a thread to avoid flooding the main channel feed. - When multiple team members @mention Search Atlas Coworker in the same channel, each request starts its own thread. - Search Atlas Coworker can only see messages in threads it is part of — it does not read general channel history. Keeping Sensitive Data Private If your request involves sensitive account data you do not want visible to everyone in a channel, use a DM instead of a public channel mention. Search Atlas Coworker's response in a DM is only visible to you.

💡 Example Prompts & Use Cases for Search Atlas Coworker

Here are ready-to-use prompts organized by role to help you get the most out of Search Atlas Coworker. Copy any prompt directly into Slack. SEO Manager - "What are my top 10 ranking keywords this week and how did they change from last week?" - "Which pages on my site lost the most impressions in the last 30 days?" - "Run a site audit on [domain] and give me the top 5 critical issues." - "What is the keyword difficulty for 'best CRM software' and who are the top competitors?" - "Show me OTTO's pending recommendations for [domain]." - "What content pieces on my site are underperforming and why?" Agency Owner / Account Manager - "Give me a performance summary for [client domain] covering rankings, traffic, and top opportunities." - "Every Monday at 9am, post a GSC summary for [client domain] to #client-reports." - "What are the top 3 quick wins I can present to my client this week?" - "Compare [client domain] vs [competitor domain] on keyword overlap and backlink profile." - "What is the white-label status for [client] and are there any issues?" Local Business / Multi-Location - "What is the local ranking heatmap for [location] for the keyword [service]?" - "Are there any issues with my Google Business Profile for [location]?" - "What GBP posts have performed best in the last 30 days?" - "Show me citation inconsistencies for [business name] in [city]." - "Which of my locations has the lowest local SEO score and what should I fix first?" PPC / Ads Manager - "How are my Google Ads campaigns performing this week vs. last week?" - "Which keywords in my campaigns have the highest cost per conversion?" - "Show me underperforming ad groups and suggest optimizations." - "What is my current ROAS across all active campaigns?" Content Team - "Grade this article against the top 5 results for [keyword]: [paste URL]." - "What are the top content gaps on my site compared to [competitor]?" - "Generate 10 title ideas for a blog post targeting 'best local SEO tools'." - "Which of my published articles have dropped more than 20 positions in the last 60 days?" Automations (Routines) - "Every Friday at 4pm, send me a summary of this week's top keyword movers to #seo-weekly." - "On the 1st of every month, run a full site audit for [domain] and post the results to #website-health." - "Every morning at 8am, send me any new OTTO recommendations that were auto-applied yesterday."