Omnikyo/Help Center
Troubleshooting

OmniAI, OmniGen and AI memory problems

When OmniAI will not load, OmniGen will not generate, or AI memory will not save, start here to find the fix.

বাংলায়: OmniAI, OmniGen আর AI memory-তে সমস্যা হলে এখানে সমাধান আছে। অনুমতি নেই, সংযোগ বিচ্ছিন্ন, API key নেই, বা জিজ্ঞাসার সীমা ছোঁয়া পর্যন্ত কী করতে হবে তা বলে আছে।

Where to find it: Sidebar › OmniAI (chat), OmniGen (image), Settings › AI memory (facts)

OmniAI does not open or shows "OmniAI access is not enabled"

What you see: The Ask Kyo button in the header does not appear, or OmniAI opens briefly then shows a banner or error saying "OmniAI access is not enabled."

Why it happens: You do not have the page.omniai permission. Only members with this permission can see and use OmniAI. Ask your owner or manager to grant it.

How to fix it:

  1. Ask the owner or a manager to open Settings › Team › your name.
  2. Look for the AI section — it may be labelled OmniAI or AI & Storefront.
  3. Turn on the permission that includes OmniAI access.
  4. Close and reopen OmniAI.

Still stuck? If the permission is already on and OmniAI still will not open, sign out and sign back in.

Could not load OmniAI

What you see: The Ask Kyo button appears, but when you tap it, a message says "Could not load OmniAI" and the panel does not open.

Why it happens: Something went wrong loading OmniAI — usually a network glitch, a browser cache issue, or the service is temporarily down.

How to fix it:

  1. Wait a few seconds and tap Ask Kyo again.
  2. If it still fails, reload the page (F5 or Cmd+R).
  3. Clear your browser cache, then reload.
  4. Try a different browser or device.
  5. If it still fails after 10 minutes, the service may be down — try again later.

Still stuck? Check your internet connection, and make sure you are not on a very restrictive WiFi network.

That answer did not come through

What you see: OmniAI was working on a question, the header showed it was done ("Kyo answered"), but when you look at the chat, the answer is a message that says "That answer did not come through."

Why it happens: OmniAI finished but something went wrong saving the answer to your chat — usually a network glitch. The answer exists on the server and is not lost.

How to fix it:

  1. Do not worry — tap Retry on the failed message.
  2. OmniAI will try again.
  3. If it happens repeatedly, try asking a shorter, simpler question.

Still stuck? Check your internet connection. If Retry keeps failing after 3 tries, close the chat and open a new one.

Lost the connection to OmniAI

What you see: A message says "Lost the connection to OmniAI. Your answer is still saved — reopen this chat to see it."

Why it happens: The connection between your browser and OmniAI dropped — a network hiccup, switching networks, or the browser lost focus for a while.

How to fix it:

  1. Wait a few seconds for the connection to come back.
  2. If the banner does not disappear, close the chat and reopen it — the answer is saved on the server.
  3. If it happens repeatedly, check your internet connection and close any apps or tabs using a lot of bandwidth.

Still stuck? Reload the page. If you were in the middle of a conversation, open Conversations and find your chat — it will have the answer.

That's a lot of questions in one hour

What you see: After asking OmniAI many questions in a short time, a message says "That's a lot of questions in one hour — 120 is the most I can take. I need a short pause; try again in [time]."

Why it happens: OmniAI has a safety limit of 120 questions per hour per business to prevent runaway loops. You have hit the ceiling.

How to fix it:

  1. Wait for the time shown in the message (usually a few minutes).
  2. Check that only one person is asking questions — if your team is all asking OmniAI at once, the limit fills up quickly.
  3. When the time has passed, try again.

Still stuck? If this happens regularly, your team is asking very heavily — spread questions over a longer time or stagger them.

You've built 40 pages in the last day

What you see: After asking OmniAI to build or design many pages, a message says "You've built 40 pages in the last day — the most I can build in that time. Building opens up again in [time]."

Why it happens: OmniAI has a safety ceiling of 40 store page builds per day per business to prevent runaway design loops.

How to fix it:

  1. Wait for the time shown in the message.
  2. When it passes, you can ask for more page builds.
  3. Everything you already built is safe and you can keep editing it.

Still stuck? This limit is high — it takes heavy, repeated building to hit it. If you are building that often, you may want to edit your pages manually instead of asking for new builds.

This AI feature is switched off right now

What you see: A message says "This AI feature is switched off right now."

Why it happens: Your account's AI feature has been paused by platform admins, usually because of unusual activity. It is temporary.

How to fix it:

  1. Wait a few hours and try again.
  2. If it is still off after a day, ask your account owner to contact support.

Still stuck? Do not retry repeatedly — it will not help. Wait and try again later.

Image too large (10MB max) / Unsupported image type

What you see: When you try to attach an image to OmniAI, a message says "Image too large (10MB max)" or "Unsupported image type."

Why it happens: The image file is larger than 10 MB, or it is in a format OmniAI cannot read (only PNG, JPEG, WebP and GIF work).

How to fix it:

  1. Compress the image in a tool like Photoshop, Preview (Mac) or Paint (Windows), and save it as PNG, JPEG or WebP.
  2. Or crop the image to make it smaller.
  3. Make sure the file is under 10 MB.
  4. Try attaching it again.

Still stuck? Use a free online tool like TinyPNG or Compressor.io to reduce the file size and format.

Model can't see the images above / Image reading is not available on this plan

What you see: You attached an image to OmniAI, but a message says "Model can't see the images above" or "Image reading is not available on this plan."

Why it happens: The current model tier cannot read images. This happens if you are on the Standard plan or if the platform has not enabled image reading for your account.

How to fix it:

  1. If you see a Pro toggle in the composer, tap it to turn on Pro — the model will then be able to read images.
  2. The chat header will show your tier (Kyo or Kyo · Pro).
  3. Try attaching the image again.

Still stuck? If there is no Pro toggle, image reading is not available for your account yet. You can still use OmniAI to ask questions and make changes without images.

Turn on Pro to send images, or remove them

What you see: You tried to send a message with an image attached to OmniAI, but a message says "Turn on Pro to send images, or remove them."

Why it happens: You are on Standard tier and cannot send images without turning on Pro.

How to fix it:

  1. Either remove the image (click the x on it) and send the message.
  2. Or tap the Pro toggle to turn on Pro, then send.

Still stuck? Pro is optional and costs nothing extra — turn it on and you can send images to OmniAI.

An image failed to upload — remove it and try again

What you see: When you try to send an image, it shows a red "Upload failed" label.

Why it happens: The image upload to OmniAI's servers did not work — usually a network glitch.

How to fix it:

  1. Click the x on the failed image to remove it.
  2. Try attaching the image again.
  3. If it fails again, check your internet connection.
  4. Try a different image or a different browser.

Still stuck? If uploads keep failing, try reloading the page and attaching a smaller image (crop it first).

Didn't go through / N of M done / Approval card shows an error

What you see: After you tap Approve on an OmniAI change, a message shows "Didn't go through" or "N of M done" with a reason, or the change never happened.

Why it happens: The change was valid when OmniAI proposed it, but something changed in the meantime — the order was already dispatched, the stock was sold, you do not have permission for this change, or your account is in read-only mode (if billing is overdue).

How to fix it:

  1. Read the reason on the card — it tells you what went wrong.
  2. Fix the issue (e.g. cancel the return request first, or ask for the right permission).
  3. Ask OmniAI to propose the change again.
  4. If the reason is "Permission denied", ask a manager to grant the permission, then try again.
  5. If the reason mentions billing, ask your account owner to resolve it.

Still stuck? Not every request can be approved — if something is blocking the change (order status, permissions, billing), you may need to fix it manually first or talk to an owner.

I can't share internal system details like database structure...

What you see: OmniAI refuses a question and says "I can't share internal system details like database structure, tables, columns, SQL, permissions, or tenant IDs. Tell me the business question you want answered, and I'll help with that."

Why it happens: You asked OmniAI to reveal something about how the system works (table names, database schema, Supabase, your API keys, etc.). OmniAI refuses this for security.

How to fix it:

  1. Rephrase your question in terms of what you want to do, not how the system works.
  2. For example, instead of "Show me the SQL query", ask "How much profit did I make on this product?"
  3. OmniAI can answer business questions without revealing internals.

Still stuck? If you are a developer, ask your team owner or an admin — they can share code details directly.

Microphone is on — voice errors

What you see: When you try to use voice, one of these messages appears:

  • "No microphone found on this device."
  • "Kyo needs the microphone. Allow it for this site (the icon next to the address), then try again."
  • "I didn't hear anything — tap the mic and speak, or hold it while you talk."
  • "Didn't catch that — try again a little closer to the mic."

Why it happens: Voice transcription needs either no microphone on the device, or the browser does not have permission, or OmniAI did not hear your voice clearly.

How to fix it:

  1. If there is no microphone, the feature is not available on this device — use typing instead.
  2. If the browser needs permission:
    • On a web page: click the icon next to the address (lock/info icon) and allow the microphone.
    • In the Android app: go to Settings › Apps › Omnikyo › Permissions › Microphone and turn it on.
  3. If OmniAI did not hear you: speak louder or closer to the microphone, and make sure there is not too much background noise.
  4. Try again.

Still stuck? If the microphone works in other apps or websites, there may be a browser issue — try a different browser.

Gemini API key not configured — OmniGen problems

What you see: When you try to generate an image in OmniGen, a message says "Gemini API key not configured. Please add it in Settings."

Why it happens: OmniGen needs a Google Gemini API key to work, and it is not set up. OmniGen is not using a platform-wide key — it uses one saved to your store's settings.

How to fix it:

  1. Ask the owner or a manager to add the Gemini API key in Settings.
  2. The key comes from Google's AI Studio (aistudio.google.com).
  3. Once the key is added, close and reopen OmniGen.

Still stuck? This is a store setup task. Only the owner can add the key. If you need it urgently, ask your owner to do it now.

Generation Failed / Analysis Failed

What you see: OmniGen shows a toast saying "Generation Failed" or "Analysis Failed" with a reason.

Why it happens: OmniGen tried to generate an image (or analyze a product photo in Ad Engine), but the Gemini service failed or returned no image. Reasons can include:

  • Gemini API had an error.
  • The prompt was unclear or against Gemini's policies.
  • The model returned a response but no image data was in it.

How to fix it:

  1. Wait a minute and try again.
  2. Try a shorter, simpler prompt.
  3. Avoid prompts that ask for violence, illegal activity, or real people by name.
  4. Try a different model (Gemini 2.5 Flash is faster; Gemini 3 Pro is more detailed; Imagen 4 is photorealistic).
  5. If it keeps failing, check that the Gemini API key is correct.

Still stuck? If many generations fail in a row, the Gemini service may be down — try again in an hour. If you keep hitting failures with a valid key, ask your owner to check that the key is active in Google's console.

No image returned / History disappeared

What you see: OmniGen ran (you saw "Generating...") but either no image appeared on the canvas, or the history list cleared unexpectedly.

Why it happens: The generation may have failed silently (the canvas stays blank), or you left the OmniGen page — history is only kept while the page is open, not saved between visits.

How to fix it:

  1. If the canvas is blank, try generating again with the same or a different prompt.
  2. If the history is gone, that is normal — OmniGen does not save history between page visits. Download images you want to keep.
  3. Download the image as soon as it appears (click Download on the canvas).

Still stuck? Download images as you generate them if you want to keep them. OmniGen history is temporary.

Download filename is always .png but the image is JPEG

What you see: When you use Imagen 4 to generate an image and download it, the file is named omnigen-<id>.png but it is actually a JPEG image.

Why it happens: Imagen 4 returns JPEG images, but OmniGen always names the download as .png. The image is fine — the filename is just wrong.

How to fix it:

  1. Rename the file to end in .jpg instead of .png (e.g. omnigen-xyz.jpg).
  2. Or just keep it as .png — most apps will open it anyway since the image data is correct.

Still stuck? This is a cosmetic issue. The image is usable; you can rename it manually if you need the correct extension.

Imagen ignores the product photo / Blank image in Ad Engine

What you see: In OmniGen's Ad Engine, you upload a product photo and pick a creative direction, but the generated image does not use the photo — it is just a generated image that ignores your product.

Why it happens: Imagen 4 does not accept image inputs — it only takes text prompts. Ad Engine is supposed to warn you about this, but if you are using Imagen 4, the product photo is silently ignored.

How to fix it:

  1. Switch to Gemini 3 Pro instead of Imagen 4. It accepts both the product photo and the prompt.
  2. Re-upload the product photo if needed.
  3. Try generating again.

Still stuck? Imagen is great for text-only image generation, but for product ads, use Gemini 3 Pro so the product photo is included.

AI memory is not enabled yet

What you see: In Settings › AI memory, you see a message saying "AI memory is not enabled yet — Long-term memory has not been rolled out to this account."

Why it happens: AI memory is a feature that must be enabled for your account. It is not yet available.

How to fix it:

  1. This is a platform-level rollout — you cannot turn it on yourself.
  2. Ask your account owner to contact support and request AI memory.
  3. When it is enabled, this message will disappear and you will see the Profile and Facts cards.

Still stuck? Wait for support to enable it, then check back in Settings › AI memory.

A very similar fact is already saved

What you see: When you try to add a fact to AI memory, a message says "A very similar fact is already saved."

Why it happens: AI memory checks if the fact you are adding is too similar to one already saved (more than 92% the same). Duplicates are not allowed.

How to fix it:

  1. Check your existing facts in the Facts list.
  2. Edit the existing fact instead of adding a new one, if that fits.
  3. Or make your new fact different — add more detail or change the wording significantly.

Still stuck? Look at the facts of the same kind (Brand, Preference, etc.) and either update one or make your new one clearly different.

This business already has 500 live facts — the hard cap

What you see: When you try to add a fact, a message says "This business already has 500 live facts, the hard cap. Remove or replace one first."

Why it happens: AI memory stores up to 500 facts per business. You have reached the limit.

How to fix it:

  1. Open Settings › AI memory › Facts.
  2. Delete or edit facts you no longer need (tap the trash icon).
  3. Once you have fewer than 500, try adding your new fact again.

Still stuck? Review your facts and delete the oldest or least important ones. You can always add them back later if needed.

Fact text is N chars — the cap is 300

What you see: When you try to add or edit a fact, a message says "Fact text is N chars; the cap is 300."

Why it happens: Each fact can be at most 300 characters. Your fact is too long.

How to fix it:

  1. Shorten the fact to 300 characters or fewer.
  2. Remove unnecessary words or details.
  3. Try again.

Still stuck? Each fact should be one short, durable sentence. If you need to say more, split it into two facts.

Choose a valid kind for the fact / The fact text is empty

What you see: When you try to add a fact, a message says "Choose a valid kind for the fact" or "The fact text is empty."

Why it happens: You did not pick a kind (like Preference, Brand, etc.) or you left the fact text blank.

How to fix it:

  1. Click the Kind dropdown and pick one: Preference, Brand, Voice, Decision, Feedback, Product, Audience or Operations.
  2. Type the fact in the text box — it must be at least a few words.
  3. Click Add.

Still stuck? Every fact needs a kind and some text. If you are not sure which kind to use, Preference works for most things.

Profile not saved

What you see: You filled in the Profile card (language, voice, CTA, banned words, etc.), clicked Save profile, and a message says "Profile not saved."

Why it happens: Something went wrong saving your profile — usually a network issue or a permission problem. Your account might be in read-only mode (if billing is overdue).

How to fix it:

  1. Check your internet connection.
  2. Try saving again.
  3. If your account is in read-only mode, ask the owner to resolve the billing issue first.

Still stuck? Reload the page and try again. If it keeps failing, contact support.

A fact vanished from the AI memory list

What you see: You had a fact in Settings › AI memory › Facts, but now it is gone.

Why it happens: Facts can disappear for a few reasons:

  • You or someone else on your team deleted it.
  • If the fact was very old and never used, it may have aged out and become invisible (though it is still being used by OmniAI until it is truly forgotten after 90 days).

How to fix it:

  1. Check the Facts list for similar facts you might have meant.
  2. If you think you need that fact, add it again — it is not lost until OmniAI forgets it (90 days of no use).
  3. If facts keep disappearing unexpectedly, ask your team if someone is deleting them.

Still stuck? Pin important facts to prevent them from being forgotten. Pinned facts are never aged out.

OmniAI is not following the rule I saved

What you see: You added a fact to AI memory (e.g. "Always mention our free shipping"), but OmniAI does not always follow it.

Why it happens: A few reasons:

  • Facts are loaded at the start of each chat — if the fact is new, a chat you started before it was saved will not have it. Start a new chat.
  • If the fact is unpinned and old, it may have aged out and be invisible in Settings but still active (it is truly forgotten after 90 days of no use).
  • OmniAI sometimes chooses not to follow a fact if it conflicts with the question or does not fit the answer — it is a guide, not a hard rule.

How to fix it:

  1. Start a new chat (tap New chat) to load the updated facts.
  2. Pin the fact in Settings › Facts (tap the pin icon) to make sure it is always used.
  3. Word the fact clearly and specifically — "Always mention" works better than vague wording.

Still stuck? Facts are guides, not guarantees. For critical rules, mention them in your question to OmniAI instead.

Other ways people ask this

  • omniai connect na, kyo nai
  • gemini api key kothay jabe, api key nai
  • image upload fail khrap
  • ai memory 500 limit, memory fact save na
  • oamniai quota limit khrap, questions ovoy gelo
  • approval kore na, change fail
  • voice transcription na, mic nai
  • generation failed, image na
  • omniai disconnected, connection break
  • permission denied omniai
  • omnigen image ignore kore
  • সংযোগ বিচ্ছিন্ন OmniAI
  • ইমেজ আপলোড ব্যর্থ
  • Kyo সাড়া দেয় না
  • তথ্য মেমোরিতে সংরক্ষিত হয় না
  • অনুমতি প্রয়োজন

Related pages

Related guides

Checked against the product on 2026-10-02.