ThreoAI Server Connection Error
Diagnosing ThreoAI send failures: provider outages, context limit reached, and rate limits, with the Keep going, Free up room and new chat resolutions.
Product: ThreoAI
Ticket Type: Common Error / User-Facing Issue
Symptom
Section titled “Symptom”A user reports that a message would not send, or that the response failed part-way through.
ThreoAI does not show a single generic “Server Connection Error”. It names the cause, so the first diagnostic step is to get the exact wording of the message the user saw. The wording is what tells you which of the causes below applies.
Messages you will be quoted, and what each one means:
| What the user saw | Cause |
|---|---|
| ”The model provider is temporarily unavailable. Try again in a moment.” | The upstream AI provider is down or degraded |
| ”The model provider rejected our credentials. Try again in a moment or pick a different model.” | A provider credential problem on the Synthreo side |
| ”The selected model is not available right now. Pick a different one from the model selector.” | The chosen model is not currently servable |
| ”Too many requests. Give it a moment and try again.” | Rate limiting |
| ”This conversation is too long for <model>. Start a new conversation, or switch to a model with a larger context window.” | The conversation no longer fits the model’s context window |
| ”This conversation reached the model’s memory limit. Start a new chat or shorten the input.” | The same cause, reported by the model mid-turn |
| ”Couldn’t reach the server. Check your connection and try again.” | A genuine network or connectivity failure |
| ”Something went wrong on our end. Please try again.” | An unclassified server-side error |
Root Causes
Section titled “Root Causes”Cause 1: Provider Outage or Degraded Provider
Section titled “Cause 1: Provider Outage or Degraded Provider”ThreoAI serves models hosted by third parties. The vendors in the platform’s model catalog are Azure OpenAI, OpenAI, Anthropic on Azure, Anthropic on AWS Bedrock, and Google, and the hosting platforms that appear on the wire include Bedrock, Azure AI and Fireworks. If one of those is unavailable, ThreoAI reports it as a provider error.
Signs this is the cause:
- The user was shown one of the three “model provider” messages in the table above
- The error affects multiple users at the same time
- The error is limited to one model, and other models still answer
- Provider status pages show an active incident
Cause 2: Rate Limiting
Section titled “Cause 2: Rate Limiting”A burst of requests, from one user or across the tenant, produces “Too many requests. Give it a moment and try again.” This is transient and clears on its own.
Cause 3: The Conversation No Longer Fits
Section titled “Cause 3: The Conversation No Longer Fits”Every model has a context window - a cap on the tokens it can process in one turn, counting the conversation history as well as the reply. A long conversation, or one with large pasted content, eventually stops fitting.
ThreoAI warns before it fails. The composer shows a context indicator with the share of the window used, and a warning icon appears once the conversation passes 75 percent. Two advisory dialogs also fire:
- “Approaching context limit” at 75 percent, offering Keep chatting or Start new conversation
- “Context limit nearly full” at 85 percent, offering Send anyway, Free up room or Start new conversation
The token indicator sits under the message box, to the right of the disclaimer line:
Signs this is the cause:
- The error names the conversation being too long, or the model’s memory limit
- The error only occurs in one specific, long-running conversation
- Earlier messages in that same conversation were answered successfully
- Starting a new conversation resolves it immediately
- The user has been pasting large documents, or the exchange has run very long
Cause 4: A Real Connection Failure
Section titled “Cause 4: A Real Connection Failure”“Couldn’t reach the server. Check your connection and try again.” is the only message that actually means a network problem. Treat the others as product-side causes, not connectivity, no matter what the user calls the ticket.
Resolution Steps
Section titled “Resolution Steps”For a Provider Error
Section titled “For a Provider Error”-
Check the provider’s status page for the affected model’s vendor. For example, Anthropic publishes at status.anthropic.com, and Azure at azure.status.microsoft. Models hosted on AWS Bedrock or Google follow those providers’ own status pages.
-
Have the user start a new chat and pick a different model. This is the step help desks most often get wrong: the model selector locks as soon as a conversation has any messages. The disabled control reads “Model is locked once a conversation starts.” A user cannot switch models inside the failing conversation. They must open New chat in the sidebar and choose the model before sending the first message.
-
Wait and retry. Provider errors are transient by design and the copy says so.
-
Report to your Synthreo MSP contact if the outage is prolonged or is affecting business operations.
For Rate Limiting
Section titled “For Rate Limiting”Wait and retry. If a tenant hits it repeatedly, raise it with your Synthreo MSP contact rather than changing anything in the conversation.
For a Context Limit
Section titled “For a Context Limit”When ThreoAI predicts the next message will not fit, it interrupts the send with a dialog titled “This chat is getting long”, which offers up to three actions:
- Keep going - resends the message using extra capacity, keeping the whole conversation intact. Nothing is lost.
- Free up room - condenses earlier messages into a short summary and then sends. Detail from those messages is lost, and they stay summarized for the rest of the chat.
- Start new chat - leaves the conversation as it is and opens an empty one.
Keep going and Free up room are not always offered. Keep going requires the customer’s large-context feature to be enabled and a larger model the user is entitled to; Free up room requires durable compaction to be enabled for that customer. If neither applies, the dialog offers only Start new chat, and that is the fix.
Beyond the dialog:
- Start a new conversation - if the history is what overflowed. An empty chat clears the accumulated thread, so a conversation that grew too long will send again. It does not help when a single oversized input is the cause: pasting the same document into a fresh chat overflows the same window. For that, shorten or split the input, or start the new chat on a larger-context model.
- Break up large inputs. Advise against pasting very large documents into a single message.
- Summarize and continue. Ask the model for the key points, start a new conversation, and paste the summary as the opening context.
- Choose a larger-context model at the start of the next chat. Each row in the model picker shows that model’s context window next to its vendor, so the user can compare before starting. Which models are offered is controlled by your MSP under Platform Admin in Canopy.
For a Connection Failure
Section titled “For a Connection Failure”Have the user check their network, VPN and any proxy, then retry. If other Synthreo surfaces also fail to load for that user, it is not a ThreoAI issue.
Prevention Tips
Section titled “Prevention Tips”- For work involving large documents, recommend a model with a larger context window and starting a dedicated conversation for it.
- Remind users that the model choice is made at the start of a conversation and cannot be changed later, so it is worth choosing deliberately.
- Users who regularly hit context limits should raise it with your MSP, who can review which models the customer has enabled.
Escalation
Section titled “Escalation”If none of the causes above applies, or the failure persists across multiple models, new conversations and multiple users, escalate to your Synthreo MSP administrator with the exact error wording, the model the user selected, and the approximate age of the conversation.