## 🛠️ How Atlas Brain Routes AI Requests — and What Causes Errors in Atlas Agent

Every task you send to Atlas Agent passes through **Atlas Brain** — the AI routing engine that selects, calls, and returns results from external AI model providers. When Atlas Brain encounters a problem at the routing layer, Atlas Agent displays an error message. While many of these errors are transient and resolve within minutes, some may persist for 24–48 hours during platform incidents — if an error continues for more than 30–60 minutes, contact support rather than waiting. Understanding how this routing works helps you interpret errors faster, choose the right next step, and know when to wait versus when to contact support.

## 🧠 What Is Atlas Brain?

Atlas Brain is the core AI engine that powers Atlas Agent. It acts as an intelligent dispatcher sitting between your request and the external AI models that fulfill it. **Every Atlas Agent response you receive has been routed, processed, and returned by Atlas Brain.** Without a functioning Atlas Brain connection, Atlas Agent cannot complete any task.

Atlas Brain is not a single AI model — it is a routing and orchestration layer that manages connections to multiple external AI model providers simultaneously, selecting the best available model for each request type.

## 🔀 How Request Routing Works Step by Step

➡️ **Step 1 — Task received:** You submit a task to Atlas Agent. Atlas Agent captures your input and forwards it to Atlas Brain.

➡️ **Step 2 — Model selection:** Atlas Brain evaluates the task type and selects the most appropriate external AI model provider for that specific request.

➡️ **Step 3 — Provider API call:** Atlas Brain sends the request to the selected provider's API endpoint and waits for a response. **This is the step where failures occur.**

➡️ **Step 4 — Response delivered:** The provider's output is processed by Atlas Brain and returned to Atlas Agent as your completed result.

Under normal conditions this entire flow completes in seconds. An error surfaces when Step 3 fails — Atlas Brain cannot reach the provider, the provider returns an error, or the target model no longer exists at the expected endpoint.

## ⚠️ The Three Root Causes of Atlas Brain Routing Failures

🔧 **Provider-side maintenance:** External AI providers occasionally take their APIs offline for planned or emergency maintenance windows. Atlas Brain detects the unavailability immediately and returns an error rather than leaving your request in a permanent pending state. These outages are typically brief — most resolve within two to five minutes — and Atlas Brain resumes automatically once the provider is back online.

🚦 **Rate limits:** AI providers enforce usage ceilings on their APIs. If Atlas Brain reaches a provider's rate limit during a period of high demand, requests temporarily fail until the rate window resets. Like maintenance outages, this is a transient condition that self-resolves without any action on your part.

❌ **Provider infrastructure failures:** Unexpected faults — regional outages, degraded APIs, or transient network issues on the provider's infrastructure — can interrupt the connection between Atlas Brain and the provider. These are outside Search Atlas's direct control and resolve when the provider restores service. They typically appear as intermittent failures rather than a complete outage.

🔴 **Model removal or renaming (the most impactful cause):** When an AI provider retires, renames, or restructures a model, Atlas Brain's routing configuration still points to the old endpoint — which no longer exists. **This causes every request to fail immediately and consistently**, not just occasional requests, because the model is permanently gone. This failure type does not self-resolve. It requires the Search Atlas engineering team to update the routing configuration and deploy a replacement model.

## 🔁 What the Failure Pattern Tells You

⏱️ **Intermittent pattern** — the error appears on some requests but not all, and clears within two to ten minutes. This almost always indicates a transient provider issue: maintenance, a rate limit reset, or a brief infrastructure fault. Waiting and retrying is the correct response. Self-service troubleshooting steps resolve these cases.

🔴 **Persistent pattern** — the error fires on every single request from the start and does not recover after ten or more minutes. This strongly indicates a routing misconfiguration, most commonly a retired or renamed model on the provider side. **No browser-side troubleshooting step will resolve this.** The fix must be applied on the routing layer by the engineering team.

Identifying which pattern you are experiencing is the single most useful piece of information you can provide when contacting the Search Atlas support team.

## 👷 How the Search Atlas Engineering Team Monitors and Responds

🔍 **For transient provider failures:** Atlas Brain's monitoring detects elevated error rates automatically. The engineering team watches provider status pages and internal signals. No code change is needed — Atlas Brain resumes routing automatically once the provider recovers.

🔧 **For routing misconfigurations:** When a model is retired or renamed, engineers update the routing configuration to replace the broken endpoint with an active equivalent. This fix is packaged and deployed as part of a regular engineering release. **Until the fix is live, all requests routed to the affected model will continue to fail.**

📢 **Active incidents affecting Atlas Brain are published on the Search Atlas status page at status.searchatlas.com.** Checking this page is the fastest way to confirm whether a known issue is already under active investigation before contacting support.

## 💡 Key Takeaways for Atlas Agent Users

✅ **Intermittent errors are almost always provider-side and self-resolve** — wait two to three minutes and retry before taking any other action.

✅ **Persistent errors that fire on every request require an engineering fix** — clearing your cache or switching browsers will not help. Contact the Search Atlas support team directly so they can identify the affected model and escalate to engineering.

✅ **Your account, your data, and your Search Atlas settings are never affected by Atlas Brain routing failures.** These errors are fully isolated to the request execution layer and carry no risk of data loss or account impact.