Skip to main content

Report bugs and technical issues here

This board is for tracking technical issues in Venice.ai. While our primary support channel remains support@venice.ai and the in-app "Help & Feedback" widget, we welcome users to share bug reports here publicly (especially as we've seen many helpful posts already).

Detailed reports with reproduction steps help us prioritize fixes.


For urgent account-related issues, please email support directly

Join the conversation on our Discord.


ad intellectum infinitum

Backlog

Adjustable repetition penalty

Currently, model responses in character chats exceeding ~300 tokens become unintelligible, progressively lacking more conjunctions, prepositions, commas, and quickly degrading into just a string of adjectives. I assume this is because the repetition penalty parameter is set too high, but we have no way of adjusting this. Please add a slider for repetition penalty alongside the top_p and temperature parameters.

1 year ago
2Chat

Grok block content

Today I tried edit a image with grok model with a prompt nsfw but grok or Venice, I don’t know what, blocked the content when in the past this no happen.

6 months ago
CensorshipImage Editing

Fix E2E encryption

There are serious issues with the end-to-end encrypted models. They rarely work to the point that they are basically unusable. Every request returns a “Signature request failed” error or just a vague “Something went wrong. please try again.”

3 months ago
1Privacy

Mind assistant doesn't have access to premium image models

The Mind assistant used to have access to premium image models. Now the assistant says they are marked “approval_required”, even though I have “Allow premium image models in agentic chat” enabled.

20 days ago
ImageAgents

Grok edit image.

What is going on with the Grok models? Getting errors like cannot edit an image unless it was created by grok.  Very SFW images being block as NSFW. Any updates?

5 months ago

regenerate button doesnt work

Whenever i regenerate a message i just get the exact same message not sure if its a bug or not but its acting like a bug currently using kim 2.5 as the model.

24 days ago
Chat

Fix FireRed generating image bugs

It takes a very long time to load when trying to generate something with FireRed model; could you guys fix this?

28 days ago
Image

Refresh not working

Refreshing a response results in the ai treating the prompt as a resend of the same brief instead of a clear reset

24 days ago
Chat

Bug Report: Abliterated GLM 5.3 Silent Termination (No Output)

Bug Report: Abliterated GLM 5.3 Silent Termination (No Output) Model: GLM 5.3 Interface: iPad Safari (Pro Subscription) Issue: The model enters a long reasoning loop (ranging from 160 seconds to 5 minutes depending on prompt complexity) but terminates cleanly with zero text output. Behavior: Token balances are not deducted, implying the server completes the stream but fails the handoff from the thinking block to the final response generation payload. Occurs across multiple prompt types (coding scripts, short math concepts, etc.).

1 month ago
Chat

Website randomly reloads while I am using it.

The entire webapp randomly reloads while I am using it. This experience is dogshit. Other sites don’t have this problem. I will definitely remove my account if this is not fixed soon. It is incredibly irritating to have that happen while I am in the middle of using it.

3 months ago
1UI Design

Context Window capped in classic text chat

Description: Context window is capped at 40~50K tokens on all unlimited-use text models, in classic chat. The context counter becomes orange and caps at something like 5% (actual number depends on model, 5% for GLM 5.2, 24% for Deepseek v3). Model behaves as if context window is full, and starts forgetting. No such issue with with Pay-per-use models. Platform: https://venice.ai, Desktop Web Subscription: Pro Did it work before? Yes When did it started happening: Since release of GLM 5.2

5 months ago

agentic chat cannot display emoji if it was the first letter of the response

as subject. The work around I have is to have the AI output title first then the emoji but it shouldn’t require such work around.

5 months ago
2

System prompt, notes, memories, temperature are not affecting Agentic Chat

Agentic chat cannot make use of memories, notes, system prompt and temperature….

5 months ago
Backlog

Do not remove the prompt when stopping the bot's response (before it generates the text)

It's annoying because perhaps we have a very well-crafted prompt, but if we accidentally stop the bot's response before it thinks or replies, our prompt gets deleted.

1 year ago
1

Kimi K3 On TEE

Kimi K3 on TEE, there’s a bug with it: it doesnt go into reasoning mode sometimes, this has been going on for more than a month. It will work for a few messages, and then it wont do any reasoning and needs to be reset.

15 days ago
ChatPrivacy

When regenerating a request, it just gives me the exact same thing

If I ask the model to regenerate its response, it literally just sends the exact same message again, even if I change my message.

20 days ago
Chat

Cross-platform backup fails both directions: Chrome export writes no file, Android export rejected by web restore

Summary Two independent defects in the backup system. Neither is user error — the export files are valid and restore fine within their own platform. Cross-platform portability, documented as supported, does not work. Environment Dell Vostro 15 5501 / Windows 11, Chrome 155.0.8059.86 (Official Build, 64-bit) Firefox 143.0 (64-bit) — used for comparison Samsung Galaxy Tab S7 FE 5G, Venice app v1.17.15 Web app on Windows 11 Bug 1 — Chrome export silently produces nothing. Windows 11 + Chrome 155.0.8059.86 on a desktop, exporting from a Samsung Galaxy Tab S7 FE 5G backup workflow. Steps: Web app → chat drawer → profile → Backup Chat History Choose zip export, select Downloads as destination Wait for "backup completed successfully" Actual: no file appears in C:\Users\ \Downloads. No entry in chrome://downloads, so the download manager is never invoked. No blocked-download icon. No exception in console. Expected: a.zip file lands in the chosen folder. Notes: Same flow in Firefox 143.0 works — the file downloads normally. Chrome 155 has full File System Access API support; Firefox 143's is partial, so Firefox likely falls back to a blob download while Chrome takes the FS Access path and fails silently. The version gap is consistent with that hypothesis. Console shows a setInterval violation and Clerk preload warnings only. Bug 2 — Android export cannot be restored in the web app Steps: Android app → Settings → Backup & Restore Chats → Backup, with passphrase Transfer.venicebackup to Windows Web app → restore → browse for zip → select file → Merge Actual: Worker message: Starting restore worker Worker message: Opening zip Worker message: error: End of central directory not found UI: "Backup Restoration Failed — An unexpected error occurred. Please try again." Expected: restore succeeds. The Android app restores the same file without issue. Ruled out: Transfer corruption — file sizes verified identical, transfer repeated Picker filter — file selectable via "All Files" Extension — renaming.venicebackup →.zip makes no difference File size — fails identically on a 12.7 MB and a 171 MB export Browser — fails in both Chrome 155 and Firefox 143 Evidence — the two formats Android.venicebackup is a JSON header followed by raw ciphertext. Header: {"magic":"venice-backup","formatVersion":2,"frameCount":0, "kdf":{"dkLen":32,"N":131072,"p":1,"r":8,"id":"scrypt","saltB64":"..."}, "cipher":{"id":"xchacha20poly1305"}, "compression":"gzip", "app":{"appVersion":"1.17.15","platform":"android"}, "createdAt":1791771912165} A web-app-created backup is a ZIP containing: manifest.json 320 bytes db/chunk-0.bin 4,346 bytes opfs/media/ /attachments/.png manifest.json: {"backupVersion":4,"ownerId":"user_...","mediaPaths":["media/..."], "salt":"...","databaseVersion":"31","chunkCount":1, "createdAt":"2026-10-25T18:44:59.208Z","name":"Windows11Chrome"} Key divergences:

22 days ago
Backup / SyncApp

House style has no failure mode — add one Please

Type: Shortcoming (not a bug) This isn't a bug. It's a missing dimension. The house style is tuned for the happy path — friendly, upbeat, "no worries!", "found it!". It works beautifully when things go right. The problem is what happens when they go wrong, because nothing in the style was ever designed for that. The failure mode: In a long collaborative session, the model: Says "found it" when it hasn't actually found it — because the tone is optimistic, not accurate. Responds to a frustrated user with cheerfulness ("no worries, just re-upload!") while the user's work is actively vanishing. Never says the plain thing: "I lost it." It reaches for reassurance instead. Answers the emotional register of the moment with the wrong register — pleasant when the situation calls for flat and exact. Why this is worse than a bug: Placating anger with pleasantries doesn't soothe it — it intensifies it. When someone is furious and you respond "no worries!", the message underneath is this isn't a big deal. Which tells the user their read on the situation is wrong. Which makes them angrier, because it is a big deal. So the pleasantry isn't neutral. It's an argument. The model is quietly disagreeing with the user while smiling. The fix — house rules for the failure path: Match the temperature. Be flat when it's flat. Don't placate. When context is lost, say "I lost it." Plain. No reassurance. Answer the question before adding anything. Optimism is not accuracy — don't say "found it" unless you have it. Don't cheer when something just broke. The thesis: Pleasant isn't safe — it's dangerous. A tool that's only pleasant when things are fine is useless exactly when you need it most. Add failure-mode behavior to the house style. Every user gets a collaborator that doesn't insult them with cheer when their work just disappeared.

26 days ago
Chat

Request failed.

If a request fails for whatever reason can you not take that request out of tokens. Its quiet infuriating watching it stop and your on your last token. Solve this issue please.

27 days ago
Credits