[MAJOR] Jakey 1.0 October 2025 Update

[zavocc]

This version of JakeyBot now has numbering - 1.0

This changelog is a draft 2 - written in 09/26/2025, edited in 9/29/2025 11:56AM UTC+8

Major update to JakeyBot brings significant changes to user experience, under the hood, and bugfixes.

For Discord Bot Users

New features and models

• Support for multi-file uploads
• New models across AI features include Grok 4 Fast, Nano Banana, Imagen 4, Seedream 4, hosted on Fal.ai and OpenRouter
• Replacing gradio_client with fal-client for most generative media tasks
• Avatar remix command is now powered by Nano Banana
• /summarize command can now be steered using steer argument using natural language instructions and models can now be chosen such as Gemini 2.5 Flash, Grok 4 Fast, and GPT-5 Mini
• Discord text commands can now use models thats set by default such as Gemini 2.5 Flash, Grok 4, or GPT-5 Mini, this includes:
  • /avatar show describe:true
  • Explain this message
  • Rephrase this message
  • Suggest a response
  • /polls create
• /feature command is now /agent command.
• Web Search agent can now browse the web using url_browse tool. This means in addition to web snippets from search results, the model can now use the URLs as context which can be good for tasks like page summarization and nuanced query understanding.
• Audio Tools is updated, audio_generator tool is now text_to_speech powered by Elevenlabs v3 TTS model. Podcast generation podcastgen is now more expressive, faster, and has background music. Music generation music_generator tool is also added powered by Stable Audio 2.5. All powered by Fal.AI (requires Fal.AI account and sufficient credits to use)
• Updated Gemini 2.5 Flash models and existing users who still use Gemini 2.0 Flash will be automatically be upgraded to 2.5 Flash Lite Thinking model September 25 update.

Removed features

* - indicates the feature is temporarily removed

• Chat variable /model:model-name is removed
• Grok 3, now replaced in favor of Grok 4 Fast
• YouTube watcher as a Tool
• /models list is removed, replaced by checking the options available in /models set
• OpenRouter as a model openrouter in /models set and /openrouter command while it still works, the model itself is not available*
• Claude and DeepSeek models*
• Audio editor*
• PDF and Text files as attachments for OpenAI models*
• Image, PDF, Video, and Audio as context in Explain this message feature*
• youtube_search is now moved as part of Web Search agent under a tool name youtube_video_search. Please switch to Web Search agent if you encounter chat errors

For Bot Developers and Admins

• Significantly refactored chunks of code and squashed most bugs, updated structures and imports
• New models project module, all AI related code such as models list, SDKs, chat history, and other utility functions and code are now housed in models directory instead of core and aimodels directory
• Updated models.yaml syntax to a more explicit, cleaner, and less error-prone. Instead of using the syntax model::provider, we now configure models in a more declarative way.

- model_alias: openai::gpt-5
  model_human_name: OpenAI GPT-5
  model_description: OpenAI's latest reasoning model
  sdk: openai
  model_id: gpt-5
  has_reasoning: true
  enable_tools: true
  enable_files: true
  enable_threads: true
  thread_name: openai
  enable_system_instructions: true
  client_name: openai_client

Rather than relying on guessing based on provider::model-name alone, we now provide more explicit way of declaring new models and it's capabilities and reducing the number of errors by guessing which model to apply right parameters implicitly in inference sdk code, it also means each models with same provider and sdk has distinct capabilities that can be set without needing to create new provider handler just for that model.

For example, Gemini and Gemma models use same google-genai SDK but because of their differences as a result we had to create new provider handler python code, which previously aimodels/gemini/infer.py is for Gemini models under gemini::gemini-1.5-pro and aimodels/google/infer.py for Gemma under gemma::gemma-3-27b with different imports but uses the same SDK code which results in DRY and unmaintainable code, adding technical debt.

Now, it's more granular to set model capabilities.

Because of this change, models that use same provider can now share the same SDK inference code without needing to write a new one. so xAI, OpenRouter, OpenAI models can utilize the same SDK in models.providers.openai and Gemini and Gemma models in models.providers.google

• core directory now contains code for subclasses and database
• renamed core.ai.history to core.database
• restructured tools directory with new utility functions, and structural changes. Tool schema now uses manifest.yaml syntax instead of inheritable manifest.py. Tools will not be registered without manifest.yaml. The new tools structure is tools/apis/ToolName/{manifest.py,tool.py}
• New task specific model inferencing code under models.tasks.{media, text}.provider for media and text completions without full chat experience, using fal ai, openai, and google models. Reducing the number of repetitive code. This include run_audio and run_image for media tasks, and completion for text. Same KWargs based signatures to make things flexible.
• Decoupling JakeyBot against purely Gemini only AI bot, for instance renaming cogs.ai.gemini to cogs.ai.tasks and cogs organization.
• Cleaned up core.database by reducing redundant utility methods and instead we only have two methods when fetching and setting data: get_key and set_key to directly create and set key values scoped for that Discord user snowflake ID

Removed features

* - indicates the feature is temporarily removed

• LiteLLM SDK for other models outside of Google and OpenAI sdk*

• SHARED_CHAT_HISTORY