🚨 amberAI has issue

What it means
A broad “something went wrong on our side or we couldn’t finish.” Typical cases:

• Temporary overload or capacity limits.

• Bad/missing setup (e.g., document chat load failure, empty message with no files).

• Empty final answer after streaming.

• API errors with no partial response.

• Uncaught agentic errors.

Your data and permissions are not affected; the action was not executed or was cleanly aborted.

Common causes

• Temporary overload in amber or a connected service (LLM, tool, connector).

• Rate limits due to many rapid successive requests.

• Prompt was empty or missing required context/files.

• Short-term network/session issues.

Immediate actions (Quick Fix)

• Wait 1–5 minutes and resend the request.

• Resend the input (“Resend” button or submit the prompt again).

• Switch the AI model:

  • Open the settings/tool menu in the input field.

  • Expand “AI Model” or “Best Model.”

  • Choose an alternative (e.g., GPT 5, Mistral Large 3, GPT 5.4, Qwen 3.5‑27B, or an org‑approved model).

  • Send the request again.

• Temporarily bypass the feature:

  • If web search is blocked, first use only internal sources or split research into smaller steps.

• Simplify the prompt:

  • Break very extensive tasks into multiple sub-prompts.

• Refresh the page or restart the session; if necessary, check your network.

If the message appears repeatedly

• Avoid peak times (e.g., on the hour) and space out requests with short pauses (2–5 s).

• Contact support (support@amberoad.ai) if:

  • The message persists for >30 minutes,

  • Only a specific feature remains blocked while others work,

  • All team members are affected simultaneously,

  • The error appears immediately on every request—even after refresh/model switch.
    Please include:time(s), affected features/models, exact prompt/step, screenshots, and whether the behavior is reproducible in other browsers/networks.

FAQ

• Are any data lost? No. The action was not executed or was cleanly aborted. You can resend.

• Does this affect permissions/compliance? No. It’s about availability/capacity.

• May I retry as often as I want? Yes—use short pauses to avoid rate limits.

🔒 Restricted lexicon

What it means
Your query (or an API message in a chain) triggered amber’s internal forbidden‑word filter. These are amber’s own compliance/content rules and differ from policies of upstream providers.

Common causes

• Inclusion of sensitive or disallowed terms defined by your organization’s compliance settings.

• Pasted content (e.g., from emails, PDFs) that contains forbidden phrases.

• Tool or agent messages internally producing restricted terms.

Immediate actions (Quick Fix)

• Rephrase the prompt to remove flagged terms while keeping your intent.

• Replace sensitive specifics with neutral placeholders (e.g., “[customer name]”, “[ID]”).

• If you pasted long text, summarize or redact sensitive lines before sending.

• If needed, ask amber to “rewrite this text to comply with company lexicon rules.”

If the message appears repeatedly

• Check if your organization’s guidelines allow the requested content in another phrasing.

• If you believe the term is incorrectly flagged for your use case, contact your admin or support (support@amberoad.ai) with:

  • The redacted prompt, time, feature, and use case justification.

FAQ

• Can I see which word was restricted? For security reasons, we do not echo restricted terms.

• Does this block all usage? Only the specific request or internal step that violates the lexicon is blocked.

• Is there a way to whitelist? Speak with your compliance/admin team; organization settings may be adjusted.

🛡️ OpenAI policy restriction

What it means
An upstream model provider (e.g., OpenAI) rejected the request due to its Responsible AI policy or a bad request.

Common causes

• Content violates the upstream provider’s safety policy (even if compliant with amber’s).

• Model‑specific safety triggers (e.g., sensitive topics, misuse).

• Invalid or malformed request payloads to the provider.

Immediate actions (Quick Fix)

• Rephrase to align with provider policies (focus on informational, non‑harmful intents).

• Remove disallowed content or ask for high‑level, policy‑safe guidance.

• Switch to another approved model with different policy scope:

  • Open settings/tool menu → “AI Model”/“Best Model” → choose an alternative (e.g., Mistral Large 3, Qwen 3.5‑27B, GPT 5.4) and resend.

• Validate inputs (e.g., ensure files/URLs are valid and the prompt isn’t empty).

If the message appears repeatedly

• Try a different model family with policies more suitable for your use case.

• Contact support with redacted examples if you believe it’s a false positive or a malformed request issue.

FAQ

• Is this amber’s policy? No. This error is from an upstream provider.

• Does switching models help? Often yes; policy boundaries differ across providers.

• Are my data safe? Yes. The request was rejected before completion.

🌐 Network errors

What it means
A generic connectivity or unknown failure occurred that doesn’t map to a specific reply token. Examples:HTTP/SSE/client failures, timeouts, or unmapped server errors.

Common causes

• Temporary internet disruptions, VPN/proxy issues, or unstable Wi‑Fi.

• Browser session hiccups, expired auth, or blocked third‑party endpoints.

• Transient outages in connected services.

Immediate actions (Quick Fix)

• Check your connection (switch network/Wi‑Fi, disable VPN if allowed).

• Refresh the page or sign out and back in.

• Retry the request; if possible, split the task into smaller steps.

• Switch the AI model or temporarily disable the optional feature (e.g., web search) and try again.

• Try another supported browser to rule out local issues.

If the message appears repeatedly

• Confirm whether teammates are affected; if yes, it may be a broader connectivity issue.

• Contact support (support@amberoad.ai) with:

  • Timestamp(s), the exact prompt/step, browser/network details, and screenshots.

FAQ

• Did my action partially run? Generally no; most network failures abort safely.

• Is it safe to resend? Yes. If unsure, recheck inputs and resend.

🚨Our systems are currently processing a large number of requests, so this feature is not available at the moment. Please try again in 5 minutes!

See this article https://helpcenter.ambersearch.de/en/articles/1780472-rotatinglightour-systems-are-currently-processing-a-large-number-of-requests

Best practices to avoid repeated errors

• Space out rapid requests (2–5 s) to avoid rate limits.

• Keep prompts focused; break complex workflows into smaller steps.

• When a feature is at capacity, either wait 1–5 minutes or switch models temporarily.

• If web search is throttled or blocked, start with internal sources and add web steps later.

• Refresh sessions periodically if you work for extended periods.

Contact support

If issues persist for more than 30 minutes, affect multiple users, or block a specific feature permanently:

• Email: support@amberoad.ai

• Include:time(s), affected features/models, exact prompt/step, screenshots, and whether the behavior is reproducible in other browsers or networks.