Releases: Fartis/MagicAI
Release list
🧙 MagicAI v0.1.1-alpha — Force of Will
🧙 MagicAI v0.1.1-beta — “Force of Will”
Fuerza de voluntad, para continuar con este sueño.
This release marks the first beta milestone of MagicAI, a local AI assistant specialized in Magic: The Gathering.
MagicAI combines official game knowledge, deterministic logic and Retrieval-Augmented Generation (RAG) to answer rules questions and assist players using:
- 📖 Official Scryfall Oracle card data
- 📚 Magic: The Gathering Comprehensive Rules
- 🧾 Official card rulings
- 💬 Persistent conversation context
- 🧠 Dynamic evidence construction
- ⚖️ Deterministic rule reasoning
- ♟️ An initial strategic assistant profile
Everything runs entirely locally using Ollama.
✨ Highlights
⚖️ Judge
The Judge remains MagicAI's factual authority.
It can:
- Detect cards automatically
- Resolve implicit card references
- Retrieve current Oracle text
- Retrieve relevant Comprehensive Rules
- Consult official rulings
- Track source versions and provenance
- Produce deterministic answers for supported interactions
- Validate LLM-generated answers against recovered evidence
The Judge does not rely exclusively on the language model. Whenever possible, answers are built directly from official card and rules data.
♟️ Strategist
This release introduces the first version of the Strategist, internally known as the Tactician.
The Strategist is designed to help with:
- Card interactions
- Play sequencing
- Synergies
- Combo identification
- Strategic risks
- Alternative lines of play
- Future deck analysis
The Strategist does not access factual sources independently.
Instead, it requests verified information from the Judge, analyzes it strategically and may challenge conclusions that appear inconsistent with the available evidence.
The Judge remains the final factual authority.
🎴 Card Recognition
- Automatic card-name detection
- Support for implicit card references
- Conversation-aware card retrieval
- Multi-card interaction context
- Local Oracle indexing
- Support for follow-up questions without repeating every card name
📚 Rules Engine
- Comprehensive Rules integration
- Automatic keyword and mechanic retrieval
- Official rules evidence
- Oracle-derived mechanic detection
- Deterministic renderers for supported rule families
- Structured validation contracts
- Source-aware answer generation
- Safer fallback behavior when evidence is incomplete
💬 Conversation Memory
MagicAI keeps track of the active conversation, allowing natural follow-up questions such as:
“What if I sacrifice it?”
or:
“Does it form a combo with Ghave and Ashnod's Altar?”
without requiring the user to repeat every previously mentioned card.
Conversation state can preserve:
- Active cards
- Previous questions
- Strategic topics
- Relevant mechanics
- Prior Judge evidence
🔍 Evidence and Traceability
MagicAI records the evidence used to produce an answer, including:
- Recovered cards
- Oracle text
- Comprehensive Rules references
- Rulings
- Retrieval queries
- Source versions
- Answer origin
- Validation attempts
- Reviewing profiles
- Authority trace
This makes incorrect answers easier to reproduce, audit and convert into regression tests.
🧪 Quality and Evaluation
The project now includes an expanded quality infrastructure with:
- Regression tests
- Performance tests
- Dynamic rule Gauntlets
- Deterministic route tests
- Oracle candidate audits
- Conversation continuity tests
- Manual community-feedback cases
- Reproducible evaluation artifacts
- Resumable and parallel test campaigns
Evaluation reports are used to improve the program logic and test coverage.
They are not automatically used as training data.
⚡ Local AI
- Ollama integration
- Local model execution
- CPU support
- GPU acceleration
- No mandatory cloud inference
- No model fine-tuning required
- Replaceable language-model backend
MagicAI's factual knowledge remains grounded in its local sources rather than being permanently embedded into model weights.
🌐 REST API
- FastAPI
- Swagger UI
- ReDoc
- Structured Judge results
- Strategy handoff support
- Conversation sessions
- Persistent history
- Feedback export
- Source-health information
🛡️ Why “Force of Will”?
Reaching beta is a major milestone for the project.
MagicAI has grown from a simple local RAG experiment into a source-grounded system with deterministic reasoning, validation, persistent conversations, quality Gauntlets and the foundations of a multi-profile architecture.
The name Force of Will represents the persistence required to keep testing, correcting and rebuilding the system whenever an apparently successful result hides a deeper failure.
This release is not the end of development.
It is the point where MagicAI has demonstrated that the idea works.
🚧 Current Status
Version: v0.1.1-beta
Codename: Force of Will
MagicAI is now in beta, but it remains under active development.
The Judge already provides a functional source-grounded rules assistant. The Strategist is still in an early stage and will continue gaining:
- Iterative access to Judge tools
- Better conversational behavior
- Play-sequence analysis
- Formal combo verification
- Commander Spellbook integration
- Authorized strategic statistics
- Decklist analysis
- Commander bracket awareness
- Format-sensitive recommendations
- Collection-aware suggestions
- Stronger adversarial validation
Incorrect or incomplete answers may still occur, especially for complex interactions that are not yet covered by deterministic logic.
🗺️ The Road Ahead
The next development milestones will focus on:
- Full Judge-to-Strategist orchestration
- An iterative Judge Tool Gateway
- Strategic investigation planning
- Formal combo and loop verification
- Improved conversational personality for the Strategist
- Commander Spellbook integration
- Deck and collection analysis
- Additional deterministic rule coverage
- Stronger second-pass validation
- Preparation for the future MagicAI 1.0 — “NicolAI Bolas”
❤️ Personal Note
MagicAI started as a personal project combining two lifelong passions:
- Programming
- Magic: The Gathering
What began as a small experiment has become a much larger challenge: building a local assistant that does not merely sound confident, but can show where its answers come from, recognize its mistakes and improve through testing.
Reaching this beta is an important personal milestone.
Thank you for taking the time to try MagicAI, report problems and follow its development.
I hope you enjoy using it as much as I have enjoyed building it.
🧙 More Gathering. Less Guessing.
🧙 MagicAI v0.1.0-alpha — Que empiece el duelo.
🧙 MagicAI v0.1.0-alpha
Que empiece el duelo.
This is the first public release of MagicAI, a local AI assistant specialized in Magic: The Gathering.
MagicAI combines official game knowledge with Retrieval-Augmented Generation (RAG) to answer rules questions using:
- 📖 Official Scryfall Oracle Cards
- 📚 Magic Comprehensive Rules
- 💬 Conversation memory
- 🧠 Dynamic context building
Everything runs entirely locally using Ollama.
✨ Highlights
🎴 Card Recognition
- Automatic card detection
- Implicit card references
- Conversation-aware retrieval
📚 Rules Engine
- Comprehensive Rules integration
- Automatic keyword lookup
- Official rules retrieval
💬 Conversation Memory
MagicAI remembers the conversation, allowing natural follow-up questions such as:
"What if I sacrifice it?"
without mentioning the card again.
⚡ Local AI
- Ollama integration
- CPU support
- GPU acceleration (CUDA)
🌐 REST API
- FastAPI
- Swagger UI
- ReDoc
🧪 Testing
Current automated tests include:
- Regression Suite
- Performance Suite
Current status:
- ✅ 25 regression questions
- ✅ 0 failures
🚧 Current Status
Version: v0.1.0-alpha
Although still in active development, the core architecture is stable and future releases will focus on performance improvements, usability and new Magic features.
❤️ Personal Note
MagicAI started as a personal project combining two lifelong passions:
- Programming
- Magic: The Gathering
Thank you for taking the time to try it.
I hope you enjoy using it as much as I've enjoyed building it.