Devlog 14

New

• Added Anthropic as an AI provider
• Added the Claude Haiku 5.5 model
• Adapted the web search tool for Anthropic models
• Added input validation to setup_env.sh
• Added a “How to use” section to the /help command
• Added ANTHROPIC_API to env.example
• Added a section about Anthropic models to README.md
• The console now shows input and output token counts for Anthropic models

Important

Anthropic is an optional provider (NevAI requires at least one of the AI providers listed on GitHub). The bot keeps working with free APIs alone, and Anthropic is used only if you add an ANTHROPIC_API key.

Devlog 13

New

• New method for installing a Docker image (install.sh)
• Auto update (Watchtower install if you use install.sh)
• The bot can see the message a user is replying to
• The bot can respond on multiple channels through changes to the data.json logic

install.sh automatically

• Installs Docker (if not installed)
• Installs docker-compose.yml from my GitHub
• Creates data.json and .env files
• Asks for API keys for .env
• Runs the Docker image

Devlog 12

New pull request

• Fixed Exa image response issues
• Changed model qwen/qwen3-32b -> google/gemini-3-flash-preview (for faster response times)
• The model now also returns sources (a safeguard against model hallucinations)
• Updated link and added a note in README.md

Exa

Doubled response speed (when using Exa) by changing the model and limiting the data returned by Exa.

Planned

I plan to re-ship this project.

Devlog 11

New pull request

• Exa.ai now returns images
• Limited Exa.ai API calls per model
• Added an error embed to the /logs command
• Split model messages (when message length > 1990)
• Added a “How to use?” section to README.md
• Dynamic bot username in the default prompt
• Prohibited Markdown tables and HTML tags
• Handled the tool_use_failed error with the Groq provider
• Shortened the returned Exa results

Exa returns images

Exa searches for images based on the queries the AI model makes using the tool (e.g. The model wants to search:, current temperature Poland 2026-09-15).

Planned

I plan to improve the /logs system and the processing of data returned by Exa.ai so that the model doesn’t make things up.

Devlog 10

New pull request

• Added Exa.ai for AI models to search the internet
• Added internet search information to /logs
• Added support for TXT files
• Added AI response time tracking
• Added information about Node version and system

Exa.ai response system

The model decides for itself whether it wants to use Exa.ai. When the model uses Exa.ai, information from the internet is returned. The model uses this information to respond in the Discord chat. When the model decides it doesn’t need to use Exa.ai, it won’t. If you don’t add an EXA_API token, the bot will continue to function, but it will not search for information on the web.

Exa supported on the following providers:

• Groq
• HackClub
• OpenRouter
• Gemini (coming soon)

Planned

I plan to add another Exa integration, based on the HackClub API for redundancy. I also plan to add information about Exa.ai to README.md.

Devlog 9

New

• Started working on a debugging system with a .log file.
• Added colorful console output to make errors easier to understand.
• Added a new .env option, DEBUG_MODE, with true/false values (when true, logs are saved to debug.log).

Planned

I plan to continuously improve the logging and debugging system.

Devlog 8

New

• Added OpenRouter free models API support
• The AI can use the Discord server’s custom emojis in its responses (emojis from the Discord cache)

Models (reminder)

All AI models are free and use free API keys (Groq, Gemini, HackClub, OpenRouter).

const geminiModels = ['gemini-2.5-flash', 'gemini-2.5-flash-lite']
const groqModels = ["openai/gpt-oss-120b", "openai/gpt-oss-20b", "qwen/qwen3.6-27b"]
const HackClubModels = ['meta-llama/llama-3.3-70b-instruct']
const visionModel = ["qwen/qwen3.6-27b", "qwen/qwen3.8-27b"]
const openroute = ["openrouter/free"]

Planned

I plan to continuously develop and improve this project.

Devlog 7

New

• Added CONTRIBUTING.md (how to add your language)
• New /help command with a description of all commands
• Finalized README.md with features, how to run it, etc.
• Added /logs command with live bot logs (question and AI model)

Planned

I plan to ship this project.

Devlog 6

New

• Added image support (vision, Qwen models)
• Added AI response error handling
• Added validation for Add custom AI reaction emoji (supports Discord emoji format)
• Added .env example in README.md
• Added MIT LICENSE on GitHub
• Added base AI prompt with a character limit (1500 characters)

Planned

I plan to add a /logs command.

Devlog 5

New

• New settings image
• Changed /ai_settings panel design
• Created 2 new bot emojis (success and failure)
• Added Discord emoji format support in the Add custom AI reaction emoji panel
• Added bot_name typing... indicator

Planned

I plan to add an embed AI response with an on/off option in /ai_settings.

Devlog 4

New

• Created the bot’s profile image
• Added the AI response system (with 15-message context history)
• Created the /ai_settings panel with custom prompt and thinking emoji options
• Extended the language files with new translation keys

Planned

I’m planning to add support for Discord’s custom emoji format in /ai_settings, and start working on a logs system showing the question asked and which AI model responded.

Devlog 3

New

• Added multi-language support
• Created /ai_settings panel
• Added English and Polish language support
• Added mention detection - the bot now checks if it was mentioned on the configured AI channel

Planned

I’m planning to connect the AI system to the bot’s replies and finish the custom prompt settings (via the modal opened from /ai_settings).

Devlog 2

New

• Created the Dockerfile and .dockerignore
• Set up a pre-built Docker image (built automatically via GitHub Actions)
• Created the /ai slash command
• Created save_data.js and a save system for the AI channel

Planned

I’m planning to create the bot’s profile picture and start working on getting AI responses inside Discord.

Devlog 1

New

• Added 3 AI providers: Groq, Gemini, HackClub
• Added 6 LLM models across those providers
• Tested the connection with each provider

LLM models

Gemini API

gemini-2.5-flash, gemini-2.5-flash-lite

Groq API

openai/gpt-oss-120b, openai/gpt-oss-20b, qwen/qwen3.6-27b

HackClub API

meta-llama/llama-3.3-70b-instruct

Planned

I’m planning to integrate these models with the Discord bot

Devlog 7

New

• Added a “Your flashcards folder” panel on the main page
• Fully functional flashcard study system (question/answer card with flip animation, folder selection, Next/Previous navigation)
• Pages now require login to access

Planned

I’m planning to ship my project.