Fasten your seatbelt: four major releases have landed back to back, and together they change what your agents can do. They can now hear you, answer out loud, take the wheel in your desktop apps and follow you from desk to phone — all inside an app that has been completely rebuilt to feel brand new.
Here is what your agents can do from today.
🎙️ Speak, and it is done
Voice mode turns TF Code into a conversation. Speak to your local agents and hear them reply — including spoken callouts when they hit a milestone or need you to unblock them. Flip it on with /voice and keep working hands-free in the terminal, the desktop app or the web.
Then take it further: computer use lets an agent take the wheel in your real applications — moving the mouse, typing, clicking and scrolling its way through the job while you watch. It is strictly opt-in, asks permission for every action and keeps a full audit log, so you stay firmly in charge.
And with Mobile Peering, your desktop agent lives in your pocket: pair your phone with a one-time code and chat with it — or set it to work — from anywhere in the world, with sleep protection keeping your machine awake and reachable.
- Voice mode — talk to your agents and be answered out loud, hands-free or push-to-talk
- Computer use — agents that move the mouse, type, click and scroll through your real apps, safely and with permission
- Mobile Peering — your desktop agent, in your pocket: chat and delegate from your phone, from anywhere
- TF Code in the app — a full TF Code panel floats inside ToothFairyAI, with session switching and Ask TF Engineer one click away
🎨 An app that feels brand new — because it is
Open ToothFairyAI today and you may not recognise it. Every screen has been rebuilt with a modern design language — cleaner menus, refreshed settings and admin panels, and polished agent views that make growing workspaces a pleasure to run.
The biggest workflows got the same treatment. The dedicated Agent Builder walks you from idea to a fully configured agent — personality, behaviour and tools — in clear, guided steps, all from one central hub.
🧠 Serverless fine-tuned models — served in your region
This is the headline for anyone who has ever wanted their own model: fine-tuning has been rebuilt end to end, and your finished models are now served serverlessly, in the region you choose. Bring your data, shape the training with friendly sliders or expert JSON, and put your model to work — no infrastructure to run, no servers to babysit. Your custom intelligence, ready wherever your users are.
- Serverless serving — your fine-tuned model goes live without you managing a single server, across the app, the API and MCP
- Regional serving — keep your model and your data close to home, served in the region you choose
- Versioned and reversible — every fine-tune becomes an immutable snapshot; roll back to any pinned version in a click
- Serving on your terms — on-demand, scheduled or always-on, with instant power control
- Smarter data prep — import multiple files with per-file validation, mark hold-out validation splits and bulk-delete training documents
🤖 A million tokens of memory
This month's model drop reads like a wishlist. DeepSeek V4.1 Flash sees images, reads text and reasons across a million-token context — while activating just 16B of its 552B parameters, so it stays remarkably affordable. GLM 5.3 arrives across standard, priority and fast tiers with 800K context. And Qwen3.8 Max brings vision and a 1M context to the flagship Qwen line.
- DeepSeek V4.1 Flash — multimodal, million-token context, cost-efficient agentic reasoning in standard and priority tiers
- GLM 5.3 — the full family: standard, priority and fast tiers, 800K context, tool calling and planning out of the box
- Qwen3.8 Max — a 2.4T-parameter flagship with vision input and a 1M default context
- DeepSeek-V4-Pro and Kimi K3 — official releases with dramatically enhanced agentic capabilities
- New voices — TF Mystica, Qwen 3.6 27B, Kimi K2.6 Fast and GPT OSS 120B make your voice agents sound better than ever
🔮 Agents that work while you sleep
If there was a theme this month, it was autopilot: hand your agents the routine work, then watch the results roll in.
- Scheduled agents — put TF Code agents on a cron schedule with missed-run recovery and full run logs. Daily dependency checks, weekly reports, periodic cleanups — done before you wake up
- TF Orchestrate — one concierge for cloud and local agents: delegate work, fan it out in parallel and let a worker/critic loop double-check every result
- Shared sessions — send an entire TF Code session to the Knowledge Hub with /share and bring it back losslessly on any machine with /import
- Alerts everywhere — know the moment something happens. Create, pause and resume alerts from the REST API, both SDKs, both CLIs and MCP, with in-app, push, email or webhook delivery
- Run the show from your phone — the mobile app now covers workspaces, analytics, members and billing, so the whole platform travels with you
✨ And there is more
- Stealthier browser automation — sharper anti-detection and proxy support for scraping tough sites, with a navigation model of your choice for every agent
- Email from your agents — send branded email notifications straight from a TF Code session, and get tailored plugin recommendations for your workspace
- Sharper plans, richer previews — review and refine plans before execution, and view PDFs and videos inline in TF Code
- Your fine-tuned models, front and centre — fine-tuned models now appear in the chat picker with the right pricing, and the MCP server gained triggers, predictions and the full alert toolkit
- Show your work — embeddable chat widgets can now display tool-call details inline, so your users see exactly what the agent is doing
- Smoother everything — improved chat streaming, more reliable Deep Search and tool calling, a per-provider model health dashboard, clearer consumption tracking and a faster sign-up
Every one of these features is live today. Open the app to feel the new design, run /voice to meet your talking agents, or put one on a schedule tonight and let it surprise you in the morning. And if you build something brilliant, tell us about it at support@toothfairyai.com — we genuinely love to hear from you.
