Skip to main content

🧚 Some updates have dropped

Released on April 20, 2026

We're thrilled to announce that Agentic Tooling 2.0 is nearing the end of its incredibly successful Beta, alongside a brand-new Home experience, a redesigned chat organisation, a massive Canvas boost, and a ~33% reduction in UoI costs across all models! 🚀

🎯 Agentic Tooling 2.0 — Beta Nearing Completion​

The Most Successful Beta Feature Is Almost Ready​

Agentic Tooling 2.0 has been an overwhelmingly successful Beta — and we're now very close to ending the Beta period and making it the default execution mode for all reasoning-model agents. Based on your feedback and usage patterns, we've refined the system further and added several capabilities that make it even more powerful.

What's Been Polished Since Beta Launch:

  • Smarter Adaptive Planning - The agent now revises its plan more aggressively when new discoveries change the picture, reducing wasted tool calls on dead-end paths
  • Improved Parallel Execution - Better detection of truly independent tasks means more tools run simultaneously, reducing response times on complex requests
  • Enhanced Memory Persistence - Key findings from long-running sessions are now preserved more reliably, so the agent never loses critical context mid-task
  • Refined Safety Mechanisms - Stuck detection and failure circuit breakers are more sensitive, catching loops earlier and saving UoI
  • Better Delegation Handoffs - Sub-agents (Researcher, Analyst, Verifier) now receive richer context, producing higher-quality results on the first attempt

Agent Delegation — A Team of Specialists: Your agent can hand off tasks to specialist sub-agents that work in parallel:

  • Researcher - Searches your knowledge base, the internet, and deep-dives into sources
  • Analyst - Processes data, runs code, analyses files, generates reports and documents
  • Verifier - Cross-references sources, performs QA, and validates results

Built-In Skills — Active Expertise: Unlike Legacy Agentic Tooling where skills sit passively in the background, AT 2.0 agents actively choose to activate skills when the situation calls for it:

  • Verify Output - Double-checks results for accuracy and completeness before presenting them
  • Step Back and Rethink - Takes a fresh look at the problem when stuck or getting poor results
  • Structure Response - Organises findings from multiple sources into a clear, coherent answer

Execution Presets — Control How Long It Works:

PresetTime LimitMax Tool UsesBest For
Short (default)30 min150Most tasks — balanced performance
Medium1 hour300Complex multi-step tasks
Long2 hours500Extended research and deep analysis
Marathon4 hours1000Exhaustive, long-running investigations

📖 Documentation: Agentic Tooling 2.0

🏠 New Home Section​

Your Go-To Workspace for Everyday Tasks​

We've introduced a brand-new Home section that gives you immediate access to a fully configured Agentic Tooling 2.0 agent — ready to go out of the box with internet access, Canvas, and code execution enabled. This covers 90% of everyday use cases without any setup required.

What You Get:

  • Pre-configured Agent - A powerful AT 2.0 agent with reasoning model, internet search, Canvas editing, and Python code execution all enabled
  • Zero Setup - Just start typing — no need to create agents, configure tools, or adjust settings
  • Covers Most Tasks - Research, data analysis, document generation, code writing, image creation, and more — all from Home

Switching to Custom Agents: When you need a specialist agent or a custom configuration, you can easily switch back:

  • Browse to Agents - Navigate to the Agents section manually to interact with your custom agents as before
  • Use # to Delegate - Type # in any chat to delegate work to a different agent from your workspace without leaving the conversation

💬 Redesigned Chat Organisation​

One Unified Conversation List Per Agent​

We've completely rethought how chats are organised. Instead of fragmented views, each agent now displays a single, unified list of conversations — making it effortless to find and return to any past interaction.

What's Changed:

  • Single List - All conversations for an agent appear in one chronological list, no more switching between tabs or views
  • Source Indicators - Hover over any chat to see where it originated (App, Web, API, etc.)
  • Cleaner Navigation - Less clutter, more focus on your actual conversations
  • Faster Context Switching - Jump between different types of tasks without losing your place

🎨 Canvas — A New Way To Create Documents Collaborating RealTime With AI​

A Powerful Collaborative Document Editor, Supercharged​

Canvas has received a major overhaul, drawing directly from the capabilities of Agentic Tooling 2.0. It's now a far more capable collaborative document editor that your agents can build, edit, and refine incrementally.

What's New:

  • Iterative Editing - Agents can now write, append, and replace content in Canvas documents with a high consecutive-call limit (1000 operations), enabling complex document-building workflows
  • Parallel Operations - Canvas tools run in parallel with other agent tools — your agent can generate an image, run code, and update Canvas all at once
  • File-to-Canvas Streaming - Agents can read uploaded files and stream their content directly into Canvas documents in a single workflow
  • Smart Chunking - Content is automatically split into manageable chunks under 3,000 characters per operation, ensuring reliable edits even for large documents
  • Read Access - Agents can now read existing Canvas content before making changes, enabling true iterative refinement rather than blind overwrites

Canvas Tool Operations:

  • canvas_write - Write or overwrite content in a Canvas document
  • canvas_append - Append content to the end of a Canvas document
  • canvas_replace - Replace specific content using find-and-replace
  • canvas_read - Read the current content of a Canvas document

💡 UoI Cost Slashed by ~33%​

More Intelligence, Less Cost — Thanks to New Provider Agreements​

We've negotiated new agreements with local infrastructure providers, and the savings are being passed directly to you. Units of Intelligence (UoI) costs have been reduced by approximately 33% across all models — meaning you can do significantly more within your existing subscription.

What This Means for You:

  • More Queries Per Month - The same subscription budget now gets you roughly a third more agent interactions
  • Cheaper Long-Running Tasks - Agentic Tooling 2.0 tasks that involve many tool calls and extended reasoning are now much more affordable
  • All Models Benefit - The reduction applies across the entire model lineup, from lightweight models to the most capable reasoning models
  • No Action Required - The new pricing is already in effect — just keep using ToothFairyAI as normal

This is part of our ongoing commitment to make AI capability more accessible and cost-effective for every user.

📖 Documentation: Units of Intelligence

📋 Summary​

This update includes:

  • Agentic Tooling 2.0 Beta Nearing End - Polished adaptive planning, parallel execution, memory, safety, and delegation
  • Home Section - Pre-configured AT 2.0 agent with internet access, Canvas, and code execution for 90% of everyday tasks
  • Redesigned Chat Organisation - Single unified conversation list per agent with hover-to-reveal source indicators
  • Canvas Massive Boost - Iterative editing, parallel operations, file streaming, smart chunking, and read access
  • UoI Cost Reduced ~33% - New provider agreements mean more intelligence for the same subscription price
  • Delegation & Skills Improvements - More reliable sub-agent handoffs and active skill activation
  • Performance & Stability Fixes - Faster chat lists, accurate UoI billing, stable Canvas operations

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! 🧚‍♀️✨

v0.677.0