Release Notes

​

v5.2.1

v5.2.1 fixes how fal.ai image and video models are charged.

​

Changes to be aware of in v5.2.1

​

fal.ai ratios use the units fal.ai reports

Aikeedo charges fal.ai models by multiplying the usage fal.ai reports by your credit ratio. For some models, the Credit ratios page showed a different unit from the one fal.ai reports, so a ratio set for the unit on the page charged the wrong amount. The page now shows the unit fal.ai reports: If you set a ratio for one of these models for the old unit, enter it again for the new one. For example, a per-second ratio for Seedance charged about 20 times too much for a 720p video. The Aikeedo Price Calculator gives the ratio for each of these units. The credits Aikeedo reserves before a generation starts use the new units too.

​

Images with fractional fal.ai usage are charged in full

fal.ai reports fractional usage for some images: Nano Banana 2.1 reports its cost in dollars, and Nano Banana 2 counts a 2K image as 1.5 images. Aikeedo rounded this down, so Nano Banana 2.1 images cost no credits and 2K Nano Banana 2 images were charged as one image. They are now charged in full. If fal.ai reports no usage, Aikeedo charges the credits it reserved for the image.

​

v5.2

v5.2 adds scheduled tasks and notifications, invoicing and business details at checkout, an Emails page, custom exchange rates, a background job queue and scheduler, and a choice of who searches the web and reads pages and files in chat. It also refreshes the model catalog, adds ten interface languages and runs on PHP 8.5.

​

What’s new in v5.2

• Scheduled tasks — users run chat instructions on a schedule, such as a morning briefing or a weekly summary. They set a task up with a form, from one of ten templates, or by asking in chat. Each plan sets how many tasks a workspace may have, and the workspace owner can see and delete every member’s tasks. Off by default. See Scheduled Tasks.
• Notifications — a bell in the sidebar lists finished, failed and waiting replies, generations and scheduled task runs, and opens with Ctrl + Shift + B. Browser alerts reach users even with the app closed, with no setup. Admins can turn browser alerts off and set how long notifications are kept. See Notifications.
• Invoicing — an invoicing engine plugin issues an invoice in your invoicing service for each paid order. Customers and admins download the PDF from the order page, failed invoices are retried, and admins can retry or create an invoice by hand. See Invoicing.
• Business details at checkout — the billing address step asks for a phone number, checks the state or region and postal code against the country, pre-selects the customer’s country, and lets businesses enter a company name and tax ID. See Billing Address.
• Purchase confirmation email — sent after every paid order, with a receipt, and the invoice PDF attached when the order is invoiced.
• Emails page — Settings → Email → Emails lists every email Aikeedo and your plugins send. Turn each one off, or preview it with sample data. See Transactional Emails.
• Custom exchange rates — set fixed rates for particular currencies under Settings → Finance → Exchange, with or without a rate provider. See Exchange Rates.
• Web search providers — choose Jina, Serper or Perplexity under Settings → Features → Chat. Results reach the model as a short cited list, capped by Results per search and Search result size, which cuts the input each search adds. See Web Search & Web Pages.
• Web page readers — pages can be read on your server, by Jina, or in a real browser by Cloudflare Browser Rendering, with Use only as fallback so a paid reader is used only for pages your server can’t read.
• File reading — chat reads .xlsx, .ods, .pptx and .rtf files, and scanned PDFs through Jina or Mistral OCR. Small files go to the model in full, so it can summarise and compare them; larger ones are searched and read page by page. See File Reading.
• Background jobs and scheduler — slow work runs as queued jobs and recurring work as scheduled tasks, each with its own schedule, isolated from the others. Settings → System → Queue shows waiting and failed jobs, workers and every task, with searchable Jobs and Workers pages. Cron is still all you need; a background worker is optional. See Background Jobs & Queue.
• Queue Adapters plugin — keeps background jobs in Redis, Amazon SQS, Azure Queue Storage or Cloudflare Queues instead of the database. Requires Aikeedo 5.2 or later. See Queue Adapters.
• Model catalog refresh — adds Haiku 5.5, GPT 6.1 Terra, GPT 6 Sol, Gemini 4 Argon, Kimi K3.1 Pro, MiniMax M3.1 Flash, Mistral Large 4, Qwen 3.8 27B, Eleven v4, Eleven v4 Turbo, Flash v2.5, Simba 3.5, Simba 3.2, FLUX 3 Max, Kling 4.0, Hailuo H3 Max Turbo, Nano Banana 2.1, Krea 3 and Stable Diffusion 3.5 Large, Large Turbo and Medium.
• OpenAI speech and transcription — Voiceover uses OpenAI’s Realtime API with GPT Realtime 2.1 mini, and Transcriber uses GPT Transcribe. See OpenAI.
• Ten interface languages — Azerbaijani, Bulgarian, Croatian, Hindi, Indonesian, Kazakh, Nepali, Slovak, Thai and Uzbek, for 40 in all. Enable them under Settings → Languages. See Localization.
• Canvas and connectors — Canvas keeps streaming while it writes long documents, and MCP App interfaces can be up to 8 MB.
• PHP 8.5 — Aikeedo runs on PHP 8.2 to 8.5. PHP deprecation notices are now logged instead of shown on the page.

​

Changes to be aware of in v5.2

​

Updating from 5.1 finishes in two steps

The update now unpacks the new files first, then finishes in a separate step: migrations, plugins, translations and cache. The Update page does this on its own, showing Finishing the update. Because 5.1 runs the old updater, a 5.1 site may show an error once at the end of the upload, such as Call to undefined method MariaDB1010Platform::getUnquotedIdentifierFolding() (the platform name depends on your database). The new files are already in place: the 5.1 updater runs the migrations in the same request, with some of the old database library still loaded, and the libraries updated for PHP 8.5 don’t fit with it. Reload the page, or open any admin page. Admin pages lead to the Update page until the update is finished, and it finishes there automatically. Use Try again if it stops. Cron does nothing until the update is finished. If you update from the terminal with php bin/console app:update, the 5.1 command can stop with the same error. Run php bin/console app:update:finalize to finish the update.

​

Web search was renamed

The chat capability is now Web search, and its settings were not carried over. After the update:

1. Choose a Search provider under Settings → Features → Chat. None is selected by default, and Serper is no longer used automatically.
2. Turn on Web search in the Capabilities section.
3. Enable Web search in each plan, and save with Update snapshots.

Plugins that registered a tool under google_search should use web_search.

​

OpenAI speech and transcription models were replaced

• Transcription: Whisper (whisper-1) is removed, and Transcriber uses GPT Transcribe (gpt-transcribe). Set its credit rate, per second, under Settings → Finance → Credit ratios. A rate you set for whisper-1 no longer applies. New transcriptions have no timestamps, so SRT and VTT downloads are available only for earlier ones.
• Speech: TTS 1 and TTS 1 HD are replaced by GPT Realtime 2.1 mini (gpt-realtime-2.1-mini), on OpenAI’s Realtime API. Set its credit rate, per character. Voices on the old models keep working, and Fable, Nova and Onyx are now spoken by Ballad, Coral and Cedar. Your server must be able to connect to api.openai.com on port 443.

​

Retired models removed

GPT 4o, GPT 4o mini, GPT 4.1, GPT 4.1 mini, Kling 1.6, Kling 1.6 Pro, Kling 2 Master, Kling 2.1 Master, Pika 2.2, Veo 3.1 (all versions), SD3 Large, SD3 Large Turbo, SD3 Medium, Perplexity Sonar Pro and Sonar Reasoning Pro, and Groq’s Llama 3.1 8B and Llama 3.3 70B are removed. Reassign any plans, assistants or defaults that used them. ElevenLabs Turbo v2.5, Monolingual v1 and Multilingual v1, and Speechify Simba English and Simba Multilingual are removed too. Voices that used them now use, and are charged as, the models that replace them.

​

Checkout asks for more billing details

When Billing address is enabled under Settings → Billing, checkout now requires a phone number, and a company name and tax ID from customers buying as a business. See Billing Address.

​

API responses show less about other users

Where an API response includes the user who created an item, such as a conversation, library item, assistant or voice, it now has only their id, first_name, last_name and avatar. Integrations that read other fields from it, such as the email address, need to change.

​

Chat attachments are checked before they are stored

Files attached in chat go through the same checks as other uploads, with a new Max file size (MB) setting under Settings → Features → Chat → File reading. Left empty, the server’s PHP limit applies.

​

Cron runs scheduled tasks and the queue

Keep cron.php running every minute. It now runs the scheduled tasks that are due and, when no background worker is running, the waiting jobs. Users’ scheduled tasks, browser alerts, invoices and purchase emails all depend on it. Plugins’ cron listeners run as the Run plugin cron listeners task, each on its own: one that fails no longer stops the others, and the failure names its plugin.

​

Workers restart after updates

If you run a background worker, app:update and saving settings make it restart on its own. There is nothing to do after an update.

​

Upgrading to v5.2

---

​

v5.1

v5.1 adds MCP connectors, an optional Connector Catalogue plugin, an app-wide private mode and keyboard shortcuts, and refreshes the model catalog.

​

What’s new in v5.1

• Connectors (MCP) — chat can use tools from remote MCP servers. Each user signs in with their own account, every tool has an always, ask or never permission (the default is to ask), and an approval card in the conversation shows exactly what a tool will send. Admins turn Connectors on under Settings → Features → Chat and choose per plan which connectors are included and how many custom ones a workspace may add. Tool calls are not charged credits. See Connectors.
• Connector Catalogue plugin — 487 vetted, ready-made MCP servers you import under Settings → Starter content → Connectors. Requires Aikeedo 5.1 or later and is available in the marketplace.
• Private mode — replaces private chats. One switch makes everything a user creates temporary across the whole app, and it stays on across pages and tabs in the same browser until turned off. See Private Mode.
• Keyboard shortcuts — shortcuts for common actions, shown where they apply and listed in a cheat sheet. Users can switch each one off. See Keyboard Shortcuts.
• Model catalog refresh — adds Claude Opus 5.5, Claude Sonnet 5.5, GPT 6.1 Sol, GPT 6 Luna, Grok 4.7 and Grok Imagine Video 1.5 Lite.

​

Changes to be aware of in v5.1

​

Encryption key

Connectors store credentials encrypted, with the key in APP_ENCRYPTION_KEY in your .env file. The update writes a key if there is none. If .env is not writable, the admin Connectors page shows a warning: run php bin/console app:generate-key. Back the key up with your database. See Connectors.

​

Private mode replaces private chats

The chat-only temporary chat setting is removed. Turn on private mode under Settings → Features → Privacy & safety, set the retention period, and check your plans.

​

User-supplied URLs cannot reach private networks

Connector addresses, web pages read during chat and dataset URLs must be public. Addresses that point to localhost, private networks or cloud metadata endpoints are refused, including through redirects. An MCP server on the same machine or network as Aikeedo cannot be used as a connector.

​

Retired models removed

Claude Sonnet 4.5, GPT 5.1, GPT 5.4 nano, Command R and Command R+ are removed. Reassign any plans, assistants or defaults that used them.

​

Upgrading to v5.1

---

​

v5.0

v5.0 is a major release. Chat gains Canvas, clarifying questions and media generation. The Writer becomes a guided wizard that continues in chat. Imagine and Video get templates, a community feed and a new composer. The interface is rebuilt around a Recents sidebar and a new Settings area. Every official plugin is now translated, and languages can be managed from the admin panel and completed with new translation tooling.

​

What’s new in v5.0

​

Chat

• Canvas — the assistant writes documents, code, HTML and SVG to a side panel instead of reprinting them in the thread. Revisions are patches, every version is kept and can be restored, and users can comment on a selected passage. All documents are collected in the Library under Artifacts.
• Inline documents — short standalone writing (emails, messages, posts, letters) renders as a card in the conversation. Emails can be opened in Gmail, Outlook or the default mail app.
• Clarifying questions — the model can pause to ask multiple-choice questions before answering an ambiguous request. The question turn and the answer turn are billed separately.
• Media generation in chat — the chat capability now generates images, video, voice-overs and transcripts, can edit an earlier image by reference, and shows results inline, including async generations. For voice-overs, users pick a voice right in the conversation and can listen to it first.
• Personalization — users can add their own chat instructions and tell the AI what to call them.

​

Writer

• Preset wizard — opening a Writer template asks its questions one screen at a time, then continues as a normal chat conversation with tools, Canvas and follow-ups. The filled prompt is never shown to the user.
• Recommended model — admins can set a recommended model on a template. See Prompt Templates.

​

Imagine & Video

• Templates — admin-authored image and video templates with preview artwork, pinned settings, required input images and optional locking that hides the prompt and model. See Prompt Templates.
• Community — users can share generations to a feed visible to everyone on the installation (Settings → Features → Community).
• New composer — quick settings chips, @image reference tags, Enhance to rewrite a prompt and Surprise me to write one (Settings → Features → Utilities).
• Inputs are kept — reference images and frames are saved with each generation and shown in the details panel, and Remix restores them.

​

Interface

• Recents — each tool’s history moves into the left sidebar, with rename, share, delete and read/unread status.
• Head bar and Details panel — every tool shows the open item’s title, actions and a details panel with credits spent, the model used and other insights.
• Settings area — account, security, capabilities, API keys (now revocable), affiliates and workspace pages live under one Settings section with its own navigation.
• Contextual “New” button, a merged workspace and account menu, and redesigned dialogs that work properly on phones.

​

Localization and administration

• Languages page — enable languages, choose the default and see how complete each translation is from Settings → Languages. See Localization.
• Translation tooling — app:locale:extract, app:locale:translate (AI translator) and app:locale:build. Updates now keep your own translation edits.
• Starter content — import Writer, image and video templates at any time from Settings → Starter content or during installation. See Starter Content.
• Model catalog refresh — adds GPT 6 Astra, GPT 5.6 Sol/Terra/Luna, Opus 5, Fable 5.1, Command A+, Grok 4.6 and 4.5, GPT Image 2.5, Grok Image 2.0 and Video 1.5, Seedance 2.5, Nano Banana 2, Seedream 5.0 Pro, Kling 3.0 Turbo, Wan 3.0, Hailuo H3, Veo 3.1 Lite, Gemini Omni Flash and FLUX 3.

​

Plugins

All plugins are now translated. Update each installed plugin to get its translations and v5.0 compatibility.

​

Breaking changes in v5.0

​

Coder removed

The Coder tool is removed. Existing code documents are converted into chat conversations during the update, together with Writer documents.

​

Composer removed

The Composer (music) tool and its AI/ML API integration are removed. Existing compositions and their files are deleted during the update. Download anything you want to keep before updating.

​

Writer documents become conversations

The standalone Writer editor is replaced by the preset wizard. Existing Writer documents are converted into chat conversations during the update. The writer.model plan setting is no longer used; users pick the model in the wizard.

​

POST /api/ai/completions removed

The legacy completions endpoint is removed from the User API. Use the conversations API instead.

​

TitleServiceInterface removed

Ai\Domain\Title\TitleServiceInterface and the provider TitleGeneratorService classes are removed. Plugins that resolved them must use TextCompletionServiceInterface or the Titler service instead. This is why every plugin needs updating.

​

Image generation capability renamed

The generate_image plan tool is replaced by generate_media, which covers images, video, voice-over and transcription. Review the chat capabilities on each plan after updating.

​

Title generator model is global

The title generator model is no longer set per plan. Set it once in Settings → Features → Utilities.

​

Reference image uploads narrowed

Reference images for Imagine and Video accept JPEG, PNG, WebP and GIF only.

​

Retired models removed

o1, o3, o3-mini, o3-pro, o4-mini, GPT-4, GPT-4 Turbo, GPT-4.1 nano, GPT-3.5 Turbo, Sora 2, Sora 2 Pro, GPT Image 1, 1 mini and 1.5, Claude Sonnet 4.0, Imagen 3 and 4, Veo 2 and 3, FLUX.1 Pro and Lucy are removed. Reassign any plans, assistants or defaults that used them.

​

Starter content re-import

Writer templates now ship with fixed IDs. If you imported the built-in templates before v5.0 and import them again from Settings → Starter content, you get one set of duplicates to delete. New installations are unaffected.

​

Upgrading to v5.0

---

​

v4.1

v4.1 adds operator-facing billing and chat controls, plus self-service account management.

​

What’s new in v4.1

• Session-based usage mode — optional alternative to credit metering that paces usage through configurable session windows while hiding credit balances from users. See Billing Overview.
• Configurable detailed pricing page URL — set where the Plans page “Detailed pricing table” button links, including external marketing sites.
• Self-service account deletion — allow users to permanently delete their own accounts (admin toggle). See Account Settings.
• Temporary chat — let users start chats that are not saved to history, with a configurable retention period.
• Chat thinking / reasoning controls — configure per-model reasoning profiles and expose thinking on/off and effort controls in chat.

---

​

v4.0

v4.0 is the largest Aikeedo release to date. It introduces the Chatbots plugin, a full Tax Layer, new payment gateways, the Sources module, and a range of platform improvements.

---

​

v3.0

Please review these notes carefully if you are upgrading from v2.x.

​

Breaking Changes in v3.0

​

Plugins Architecture Changed

• Plugins are now installed in /extra/extensions instead of /public.
• Old plugins remain compatible, but you must reinstall them manually.

​

”Cohere Web Search” Chat Capability Removed

• The Cohere integration now uses Cohere API v2.0.
• The built-in web search feature from Cohere has been removed and replaced with Google Search.
• Please update your plan configuration accordingly.

​

Removed Models

The following legacy/deprecated models have been removed. Please use other available models:

• openai/gpt-3.5-turbo-instruct
• openai/o1-preview
• openai/o1-mini
• openai/gpt-4.5-preview
• openai/text-moderation-latest
• openai/dall-e-2
• fal-ai/flux-realism (not available anymore)
• grok-beta
• grok-vision-beta

​

Search Box Removed

• The search box on the user app dashboard has been replaced with a new quick access bar.
• To search for library items, visit the Library pages.

​

Ai/Infrastruture Namespace Renamed

• The namespace is now correctly spelled as Ai/Infrastructure.
• If you use services from this layer in custom code, update the namespace accordingly.

​

Change in ImageServiceInterface::generateImage() Method Signature

• The signature of Ai\Domain\Image\ImageServiceInterface::generateImage() has changed.
• It now returns ImageEntity directly.
• Update any custom image generator services to match the new signature.

​

Keys for the Following Models Have Been Updated

If you use any of the following models, set the credit rates and enable them both globally and at the plan level again:

• fal-ai/flux/dev
• fal-ai/flux/schnell
• fal-ai/flux-pro

---

​

Migration Guide: Upgrading from v2.x to v3.0

If you have custom plugins or integrations, review the breaking changes above and update your code as needed.