01 / Start
Choose a subject
Select math, science, reading, typing, or another topic. A.R.C. spins up a specialized tutor configuration with the right domain strategy and tools.
Teaching loop
Assess, teach, play, adapt.
A.R.C. is designed around a reasoning loop, not a static lesson pack. The tutor decides what to ask, when to shift format, and how quickly to increase challenge.
02 / Reason
Teach through tool calls
The agent runs a reason-act-observe loop with structured actions like ask_question, check_answer, show_hint, and adjust_difficulty instead of relying on one big prompt.
03 / Play
Switch into the right game
When practice benefits from play, A.R.C. moves into one of seven Phaser-powered games. Platformers, typing battles, and wave-defense loops reinforce the skill currently being taught.
04 / Adapt
Adjust in real time
Correct ratios, streaks, hint usage, and XP all feed the next step. Strong learners are stretched, struggling learners receive more support, and the experience stays individual.
Learning system
A structured tutor, not a static lesson pack.
A.R.C. keeps the playful parts visible while retaining the control, observability, and local-first posture that define the Helix family.
Reason-act-observe agent loop
The tutor uses a structured loop rather than one-shot answers. Tool calls drive both conversational lessons and game transitions, so the next move reflects real observations from the learner.
Seven-game learning suite
The current lineup includes Number Runner, Froggie Math, Typing Defense, Lexicon Realm, Type Dash, Type Quest, and Endless Runner.
- Math platformers and arithmetic roadblocks
- Typing battles, wave defense, and story play
- Mode variants and difficulty progression across levels
Difficulty that moves on a spectrum
A.R.C. adjusts in real time using correctness, streaks, and hint usage. Students who need support get smaller steps and more explanation while strong performers are challenged immediately.
Deep gamification and sensory feedback
XP, levels, combo streaks, and achievements keep momentum high while the games layer in Web Audio feedback, camera shake, particle bursts, lawnmowers, boss waves, and localStorage personal bests.
Multi-provider LLM support
Anthropic, OpenAI, and Gemini integrations run through raw HTTP with tool-calling support. The teaching logic stays provider-agnostic and can also move to local Ollama when needed.
Child-friendly interface design
The product UI uses a friendly MudBlazor surface, a purple and pink play palette, Nunito typography, and an animated mascot that celebrates wins, encourages after mistakes, and reacts to processing state.
Local-first deployment
Blazor Server plus SQLite keep learner data on your infrastructure. No tracking pixels, no third-party analytics, and no forced cloud dependency when a local model or air-gapped posture is required.
Typed tool system for new teaching modes
The IAgentTool interface lets teams add new capabilities without rewriting the tutor loop. Tools describe their own schema, execute against shared context, and return both model-facing and UI-facing output.
Architecture
Agent reasoning above games, tools, and local storage.
A.R.C. keeps the teaching brain separate from provider integrations, UI actions, and game execution so the system can grow without losing clarity.
3
LLM providers
10
Agent tools
42
Unit tests
7
Learning games
Agent loop
The tutor loop reasons about what to ask next, observes the learner response, and keeps iterating until the lesson objective is met. That makes the experience adaptive without losing structure.
Tool registry
The current typed tool set covers AskQuestion, CheckAnswer, ShowHint, ShowExplanation, ShowExercise, ShowProgress, AdjustDifficulty, GenerateGameRounds, GenerateStoryBeat, and GenerateTypingDefenseWaves.
Provider layer
Anthropic, OpenAI, and Gemini implementations run over raw HTTP with tool calling. The same reasoning loop can pivot to local Ollama when privacy or offline operation matters more than cloud throughput.
Delivery surfaces
Sealed UI actions, the seven-game Phaser engine suite, Blazor Server, and SQLite storage work together so conversation, play, progress, and persistence stay coherent across a session.
Learning settings
Designed for home, school, and adaptive practice.
A.R.C. is broad enough for several environments, but the constant is the same: individualized learning with a sovereign deployment path.
K-12 math tutoring
Students practice arithmetic, algebra, and problem-solving with an AI tutor that can immediately translate the lesson into play.
- Adaptive difficulty per student
- Game-based practice with Number Runner and Froggie Math
- Immediate feedback on every answer
- Streak rewards that reinforce progress
Homeschool supplement
Parents can run A.R.C. locally and hand off repetition, hints, and lesson pacing while keeping the learner data inside the home.
- Zero cloud dependency when using local models
- No student data collection
- Self-hosted on everyday hardware
- Multiple subjects from one tutor surface
Classroom integration
Teachers can use A.R.C. for differentiated practice while focusing their own attention on live instruction and intervention.
- Per-student difficulty adaptation
- Works alongside existing curriculum
- No internet required with local inference
- FERPA-friendly on-prem posture
Corporate training
With custom tools, the same loop can support onboarding, compliance practice, and role-specific learning that still benefits from adaptivity and audit trails.
- Custom tool development via IAgentTool
- Domain-specific question generation
- Progress tracking per learner
- Sovereign deployment for sensitive content
Pilot program
Pilot sovereign adaptive learning.
A.R.C. is in development alpha for teams, families, and schools that want individualized tutoring without handing learner data to an opaque platform. Start with a classroom, a home setup, or a training cohort.