Error messages
Look up the exact message you see in Halo: interrupted replies, model errors and rate limits, context limits, images, Windows, Digital Humans, bots and teams.
Search this page for the message you see with ⌘ F / Ctrl F. If it isn't here, copy the exact text to Halo and ask about it.
In a conversation
| Message | Meaning | What to do |
|---|---|---|
| "Model response interrupted" | The reply didn't finish normally | Click Continue and Halo carries on |
| "Model response interrupted" with "Reached the maximum turn limit. Send a message to continue." | One message used up its step limit | Send a message to continue; if it happens a lot, raise Settings › Advanced › Max Turns per Message |
| "Something went wrong" | This turn failed | The red box says why. For the full error, find the failed step in the thought process above the reply: click Copy (tooltip "Copy error message") and send it to Halo or support; for a long error, click Expand full error to read it all |
| "Approaching context limit", "Context will be automatically compressed soon" | The conversation is nearing what the model can hold | Nothing to do; Halo compresses it automatically. If you've moved on to a new topic, start a new conversation |
| "Context has been intelligently compressed" | Earlier content was condensed into a summary | Normal. Repeat any important requirement if needed |
| "OAuth token expired or invalid. Please login again." | Your account sign-in expired | Sign in again as in Connect an AI model |
The model service failed (retries automatically)
When a model request fails and Halo is about to retry, a countdown card appears in the reply with "Attempt N of M" and a Stop button. Its title says why:
| Card title | Meaning | What to do |
|---|---|---|
| "Could not reach the model service" | The model service can't be reached | Check your network and proxy |
| "The model service is overloaded" | The provider is too busy | Let it retry, or send again later |
| "The model service is limiting requests" | Too many requests, or your quota is used up | Wait a little; if it keeps happening, change your plan or model |
| "The model service rejected the credentials" | The key or sign-in isn't valid | Check the key under Settings › AI Model, or sign in again |
| "The model provider reported a billing problem" | The account is unpaid or out of credit | Top up in the provider's console |
| "The model service returned an error", "The model request failed" | Something else went wrong | Read the status code and text on the card, and see Model API error codes |
Connecting a model or fetching its model list
| Message | Meaning | What to do |
|---|---|---|
| "Invalid API Key or no permission" | The key is wrong, or lacks access to that model | Copy the key again; check that your account can use the model |
| "Endpoint not found, check the URL" | The API URL is wrong | Check the address in the provider's documentation |
| "Rate limited or out of quota" | Rate limited, or quota used up | Try again later, or top up |
| "Request timed out, check the server or proxy" | The server is slow or the proxy has a problem | Check your proxy settings and try again later |
| "Cannot reach server, check network or proxy" | It can't connect | Check the network, the proxy and the API URL |
| "Failed to fetch models" | Another reason | Read the provider's explanation after the colon |
Test connection results are in English, such as "Invalid API key", "Rate limited - try again later" and "Cannot connect to API server - check URL"; they mean the same as the table above. See Get an API key.
Images and attachments
| Message | Meaning | What to do |
|---|---|---|
| "Image too large (…), max 20MB" | One image is over 20 MB | Compress or crop it, then send again |
| "Unsupported image format: …" | Only JPEG, PNG, GIF and WebP are supported | Convert it to one of those |
| "Current model has no vision — images will be read via local OCR (text only)" | The current model can't see images, so Halo passes it only the text in them | To have it understand the picture itself, switch to a model that can see images |
| "N item(s) could not be attached: not a file or folder on this computer" | What you dropped isn't a file on this computer | Save it to the computer first, then attach it |
Windows command environment
| Message | Meaning | What to do |
|---|---|---|
| "Command execution environment is not installed, AI cannot run system commands" | Git Bash is missing, so the AI can't run commands | Click Install now; see Install and launch problems |
| "Initialization failed" | Installing Git Bash on first launch failed | See It says "Initialization failed" |
Digital Humans
| Message | Meaning | What to do |
|---|---|---|
| "Not started" | A connection it needs isn't ready | See A run shows "Not started" |
| "Stopped itself after repeated failures" | 5 runs in a row failed, so its automatic tasks stopped | Fix the cause, then click Resume work |
| "An independent execution is already running" | The previous run hasn't finished | Wait for it, or click Stop this execution |
| "The holiday calendar does not cover today, so this scheduled run was skipped. Run it manually if needed." | The holiday calendar didn't download today's date | Click Run now if you need it |
| "Remote Access is off. Enable it to trigger this over HTTP from other devices." | The HTTP trigger needs Remote Access | Turn on Settings › Remote Access › Enable Remote Access |
Bots
| Message | Meaning | What to do |
|---|---|---|
| "⚠️ This bot has no owner yet, so it cannot execute tasks." | Permission Control is on but there's no owner | Message the bot directly once; see Owner and guest permissions |
| "⚠️ This bot only responds in group chats. Please contact the bot administrator for access." | Reply Scope is Group chats only | @mention it in a group, or change the scope |
| "⚠️ This bot only responds to direct messages." | Reply Scope is Direct messages only | Message it directly, or change the scope |
| "This bot is already in use by another instance. Each bot can only be bound to one digital human." | The same bot was connected twice | Use the existing entry |
| "Connected · no messages received yet" (Feishu) | Connected, but nothing has arrived | See Feishu: no messages arrive |
| "File export denied: … Only files within the current space or temporary directory can be sent." | The file isn't in the Digital Human's workspace | Put the file in its workspace first |
More in Bot not replying.
Teams
| Message | Meaning | What to do |
|---|---|---|
| "Add at least one member, or let AI build the team" | The new team has no members | Add a Digital Human, or click Let AI build the team from the goal |
| "Turn on Remote Access in Settings so others can join this team." | Invites need Remote Access or a Federation Gateway | See Teams across computers |
| "Could not join the team. Check the link and that the person who invited you is online." | The inviter's computer can't be reached | See Teams across computers |
| "This invite is no longer valid — it may have expired, been revoked, or the team was closed. Ask for a new invite link." | The link no longer works | Ask the inviter for a new link |
| "This team requires everyone to update Halo to the latest version" | Versions don't match | Click Check for updates |
| "‹member› did not receive this message. ‹owner›'s computer is offline right now; …" | The receiving member's computer is offline | Resend once it's back online, or assign the work to another member |
| "This message was not delivered." | The receiving member is no longer in the team | Assign the work to another member |
Halo is stuck or keeps failing
After repeated errors, Halo resets its AI engine by itself; if that doesn't help, it shows "Halo is having trouble" with Restart App and Try to Fix.
You can also do it yourself: open Settings › System › System Diagnostics, click Run Diagnostics, then Reset AI Engine; if that fails, click Restart App. See Reset the AI engine or restart. When you ask someone for help, click Copy Report or Export Report. The report doesn't contain the key itself. In error messages, Bearer tokens, JWTs and some keys starting with sk- are masked, but formats such as Anthropic (sk-ant-…) and OpenAI project keys (sk-proj-…) are not; user names in macOS and Linux paths are hidden, but not in Windows paths. Read it through before sending and delete any key you see.
Next steps
- Model API error codes: what status codes like 401 and 429 mean.
- Help center: find troubleshooting pages by problem type.