Skip to main content

🧚 Some updates have dropped

Released on April 26, 2026

This release marks Agentic Tooling 2.0 as the official default, introduces TF Code integration in desktop apps, enables end-to-end fine-tuning via MCP/SDK, launches a multilingual UI, and brings a smarter Fairy concierge agent! 🚀

🤓💻 TF Code Meets ToothFairyAI App​

Your AI Engineer & ToothFairyAI Engineer, Integrated Into Desktop​

You've asked for it, and we've delivered — TF Code is now natively integrated into all ToothFairyAI desktop apps. But this isn't just about coding locally — the pre-configured TF MCP means you can engineer directly in ToothFairyAI too. Manage agents, create documents, trigger fine-tuning jobs, set benchmarks, and automate workspace operations — all from a single tab. A dedicated tab at the top of the app lets you switch effortlessly between working with remote agents and engineering in ToothFairyAI.

What's New:

  • Dedicated Tab - Switch between Agents (remote AI) and Code (engineer locally + automate ToothFairyAI) with a single click
  • Engineer in ToothFairyAI - Thanks to TF MCP, you can manage agents, create documents, trigger fine-tuning, set benchmarks, and automate any ToothFairyAI operation — directly from Code
  • User-Friendly View - For those who prefer GUIs, TF Code now offers an intuitive visual interface right in the app — no terminal required to get started
  • Terminal Still Available - Of course, for the nerds among us, the full TF Code terminal remains fully accessible with all the power and flexibility you expect
  • VSCode/Cursor Extension - We've shipped a dedicated tfcode extension for VSCode and Cursor that launches the TF Code terminal directly within your IDE, faster than opening a separate terminal window
  • Automate Everything - Use Code commands to interact with agents, create documents, manage workspace operations, trigger fine-tuning workflows, and more — all from your local terminal or app view

📖 Documentation: TF Code

🎯 Agentic Tooling 2.0 — GA & Default​

The Beta Era Is Over — AT 2.0 Is Now Official​

After an incredibly successful Beta period, Agentic Tooling 2.0 is now the default and recommended execution mode for all Operator agents powered by reasoning models. We've graduated from Beta with confidence — your feedback and ideas on how far we could strech it have shaped something we are genuinly all at ToothFairyAI in love with so thank you!

What's Changed:

  • Official Default - AT 2.0 is now the recommended modality for new Operator agents
  • Agentic Tooling 1.0 Deprecated - Legacy Agentic Tooling is now deprecated. While it remains functional for existing agents, we strongly encourage migration to AT 2.0 for better performance, adaptive planning, and parallel execution. We will dismiss the modality in the coming weeks.
  • RAG Agents Continue - We hear you! Standard RAG agents remain fully supported. They are a beloved modality for many use cases, and we are committed to maintaining this capability for the foreseeable future

Why Upgrade to AT 2.0:

FeatureLegacy ATAT 2.0
PlanningFixed upfront planAdaptive, revises on the fly
ExecutionSequential onlyParallel for independent tasks
MemorySaved at end of sessionMid-task saves and recalls
DelegationBasic handoffsRich-context sub-agents
SkillsPassive backgroundActive when situations call for them

Migrating Is Easy:

  1. Open Settings > Agents > [Your Agent]
  2. Under Agentic mode, select Agentic 2.0
  3. Ensure your agent uses a reasoning model
  4. That's it — the agent immediately benefits from adaptive planning, parallel execution, and all AT 2.0 capabilities

📖 Documentation: Agentic Tooling 2.0

🧚 Fairy Concierge — Smarter Home Agent​

Your Intelligent Entry Point, Now With Full Delegation​

We've supercharged the Home module's Fairy concierge agent. Building on the positive feedback about its general productivity capabilities, Fairy (yeah, what a surprising name...) can now delegate work across all agents in your workspace — you just describe the task, and Fairy picks the right specialist or handles it directly.

What Fairy Now Does:

  • General Productivity - Research, analysis, document creation, code writing, image generation — all handled seamlessly as before
  • Intelligent Delegation - When a task suits a specialist agent (or you mention one), Fairy hands off the work automatically to the best-fit agent
  • Your Preference Matters - Simply mention an agent in your request (e.g., #ResearchAgent) to indicate your preferred specialist, and Fairy will route the task accordingly
  • Seamless Handoffs - Delegation happens invisibly — you get the result, not a handoff notification

How to Use It:

  • Direct Tasks - Describe what you need and Fairy handles it or delegates appropriately
  • Preferred Agent - Include #AgentName in your message to suggest a specific agent
  • No Setup Required - Fairy automatically knows about all agents in your workspace

🤖 Enhanced Sorcerer & Mystica Models​

Smarter Routing, Larger Pool, Better Performance​

The Sorcerer and Mystica model families have received significant upgrades to their underlying model pool and routing infrastructure. These improvements enhance response quality, speed, and reliability across all conversational interactions.

What Improved:

  • Larger Model Pool - More models available in rotation, reducing bottlenecks during high-traffic periods
  • Smarter Routing - Improved routing algorithms better match requests to optimal models based on task type, complexity, and current load

🌐 Multilingual UI — Now in 9 Languages​

ToothFairyAI Speaks Your Language​

The ToothFairyAI interface is now fully localised in nine languages — English, German, Spanish, French, Italian, Portuguese, Dutch, Polish, and Finnish. Every label, message, and setting is translated, making the platform accessible to a much broader global audience.

Available Languages:

  • 🇬🇧 English
  • 🇩🇪 German (Deutsch)
  • 🇪🇸 Spanish (Español)
  • 🇫🇷 French (Français)
  • 🇮🇹 Italian (Italiano)
  • 🇵🇹 Portuguese (Português)
  • 🇳🇱 Dutch (Nederlands)
  • 🇵🇱 Polish (Polski)
  • 🇫🇮 Finnish (Suomi)

How to Switch:

  1. Navigate to Profile > Language
  2. Select your preferred Language
  3. The interface updates immediately

🔌 End-to-End Fine-Tuning via MCP & SDK​

Build, Train, Own — Your Models, Your Business​

The ToothFairyAI MCP server and Python SDK now support complete fine-tuning workflows — from dataset creation to model deployment. This is a game-changer for businesses that need custom models tailored to their domain.

The Full Pipeline:

  1. Create Training Data - Generate datasets from previous conversations (MagicWand) or from your own documents (files, PDFs, etc.)
  2. Run Experiments - Configure and launch fine-tuning experiments with your chosen model and training method (SFT, DPO, GRPO, KTO, CPT)
  3. Set Benchmarks - Define benchmark tests to evaluate model performance against expected answers
  4. Evaluate & Iterate - Run benchmarks, review scores, adjust hyperparameters, and repeat until performance meets your standards
  5. Download & Own - Download the trained model weights and all experiment metadata — the model is yours, not ToothFairyAI's

MCP Tools Available:

  • generate_training_dataset_from_chat - Convert conversations into training data
  • generate_training_dataset_from_file - Transform documents into training data
  • list_trainable_models - View available models and estimated costs
  • start_training - Launch a fine-tuning job
  • get_training_status - Monitor training progress and download results
  • run_benchmark - Evaluate model performance
  • download_model - Download trained model weights

Availability: End-to-end fine-tuning is available for Business and Enterprise subscriptions. Individual users can create datasets but require higher tiers for full training.

✨ Simplified Chat UI​

Collapsed Tools, Cleaner Conversations​

We've decluttered the chat interface by collapsing tool displays into organised sections rather than showing everything at once. This keeps conversations focused while keeping capabilities accessible.

What's Changed:

  • One Section to Rule Them All - All available tools are now grouped into a single, collapsible section
  • Less Visual Noise - Tool names and status indicators no longer clutter the conversation view
  • Expand When Needed - Click to expand the tools section when you want to see what's available or monitor active tools
  • Faster Load Times - Collapsed by default means fewer DOM elements to render, especially for agents with many tools enabled

Before vs After:

BeforeAfter
15+ tool chips visibleSingle collapsed section
Scattered indicatorsOrganised expandable group
Visual overload on complex agentsClean, focused conversation

📋 Summary​

This update includes:

  • Agentic Tooling 2.0 GA - Now the official default, Legacy AT deprecated but RAG agents continue
  • Fairy Concierge - Smart delegation to any workspace agent, with preference hints via #
  • Sorcerer & Mystica - Larger model pool, smarter routing, improved quality
  • Collapsed Chat UI - Tools grouped into single expandable section, cleaner conversations
  • 9 Languages - Full UI localisation in EN, DE, ES, FR, IT, PT, NL, PL, FI
  • TF Code Desktop - Native integration with dedicated tab, GUI view, terminal for power users, and VSCode/Cursor extension
  • End-to-End Fine-Tuning - MCP/SDK support for dataset creation, training, benchmarks, and model download (Business/Enterprise)

Thank you for using ToothFairy! We're constantly working to improve your experience. If you have any feedback or encounter any issues, please don't hesitate to reach out to our support team at support@toothfairyai.com.

Keep building magic with AI! 🧚‍♀️✨

v0.678.0