Every day, millions of people open an AI chat, ask a question, get an answer, and close the tab. The conversation is gone. The context is lost. The notes you meant to take never got taken. The file you generated is buried in a download folder. Tomorrow, you start over.
That is the problem QTChat was built to solve.
QTChat is the harness at the center of QuantumThreads, a private, local-first AI studio for text and image work. It is not another chatbot wrapper. It is a persistent workspace where conversations keep their notes, highlights, attachments, and folders. Where every answer shows its working. Where your threads stay yours, organized and searchable, long after the tab is closed.
What makes QTChat different
Most AI tools treat a conversation as a disposable exchange. You type, the model responds, and the thread fades into a sidebar list you never revisit. QTChat treats a conversation as a working document. It is a place where ideas accumulate, where context builds over time, and where the work you do with AI becomes something you can return to, search, and build on.
The difference is not just storage. It is structure. QTChat gives you the tools to organize, annotate, and navigate your conversations the way you would organize a project folder or a research notebook. Pin the conversations that matter. Group them into colour-coded categories. Search across every message, file, and generated output. Jump to the exact line where a keyword appears. Scroll back through the full timeline of any thread without losing your place.
Every conversation gets a deep link. Share it, bookmark it, or open it directly. The thread is always there, exactly as you left it.
The composer: two modes, one workspace
QTChat's composer adapts to what you are doing. In LLM mode, labelled "Ask anything," you have a full text conversation with the model. In Image mode, labelled "Describe the image," you generate images from descriptions. Switch between them inline without leaving the thread.
The model picker and agent picker sit right in the composer, so you choose the right tool for each turn without navigating away. Attach images, audio, PDFs, text files, Markdown, JSON, or CSV. Attach an entire folder as context. Drag items in from the drive. Select text anywhere in the conversation and reply with it quoted, keeping context precise.
A generation settings popover and stop-and-send controls give you command over every turn. You decide how the model responds, and you can interrupt mid-generation if the direction is wrong.
Answers that show their work
QTChat does not hide how it thinks. Every answer ships with an animated step trail showing the stages the turn moved through, with per-step status. You see the analysis, the research, the reasoning, and the final composition. Not as a debug log, but as a readable trail that builds trust and understanding.
A decision card accompanies each answer. It surfaces the route chosen, the confidence level, the depth of the answer (brief or deep), the specialist used, whether files were needed, and the stance taken. The card is presented under the name Jev. Jev shows the route chosen, its confidence, and the depth of the answer. The assistant shows its working instead of hiding it.
Answers render as Markdown with collapsible reasoning displays for models that expose their thinking. Generated files appear as download cards. Images open in a lightbox. Per-message actions include copy, retry, edit, and answer version history, where you can re-run a prompt and switch between earlier rewrites. Token statistics and a context window meter appear on every turn, so you always know how much room remains.
Sub-agents: delegate without leaving the thread
Sometimes a question needs a deeper dive than the main conversation can provide. QTChat lets you launch a nested run from any selected passage. Watch it live as it moves through queued, running, done, failed, or stopped states. Each sub-run gets its own child thread, kept out of the main sidebar so your workspace stays clean.
While a sub-agent works, a card shows it reading the parent conversation before it replies. If the digest is unavailable, the answer proceeds without one. A dedicated panel lists all runs and accepts new ones with prompt, agent, attachments, and model. You stay in control of the main thread while the sub-agent handles the deep work.
Notepad and highlights: turn conversations into documents
The notepad is a per-conversation scratch space. Clip a passage and it pre-fills a note, saved only once it has a title. Notes show their origin, whether from you or from the assistant, and link back to the source message. Highlights let you mark spans across the conversation in five colours, filterable, jumpable, copyable, and removable.
Together, these features turn a conversation into a working document. You are not just reading answers. You are building a record of your thinking, annotated and organized, ready to revisit or share.
Built-in messenger: talk to your team without leaving the studio
QTChat includes a one-to-one direct messenger inside the same application. The conversation list shows unread counts, mute, and search, with swipe-back navigation on mobile. Messages support emoji reactions, per-message reply, copy, edit, and delete. Live typing indicators and presence dots keep you connected. Read receipts, block and unblock, live updates over a stream, and a mobile full-screen sheet variant complete the feature set.
Small teams get shared workspaces plus direct messaging without adopting a separate SaaS chat tool. Everything lives in one place.
Notifications that respect your attention
A bell in the top bar shows an unread count. The list is keyboard-operable with severity styling across info, warn, and critical levels. Mark read, mark all read, clear all, or click through to the relevant chat. User preferences include a master switch plus sound and desktop notice channels. You control what interrupts you and what waits.
Settings: seven sections, full control
QTChat's settings system covers everything you need to make the studio yours.
Providers: connect any OpenAI-compatible provider. API keys are write-only and masked in the UI. Per-model context window, reasoning and vision flags, thinking control declarations, and model list fetch and preview. Built-in text and image models ship under the Agnes brand. A subscription provider option, OpenCode Go, exists as the alternative to a self-supplied key.
Behaviour: create and edit custom agents. Connect your own tool servers over stdio or URL transports with redacted secrets. Manage rules and skills. Toggle bundled plugins on or off.
Display: three interface themes (dark, gray, light), six message bubble colour themes, choice of how model thinking is shown, and chat font size, all with a live preview.
Workspaces: a registry of member areas with create, rename, colour, and delete. Each workspace owns its own chats, files, and members. A monogram switcher moves between them.
Security: a four-digit lock screen passcode, an idle auto-lock window, and lockout behaviour. A danger zone for bulk deletion.
General: product facts such as default text and image models, a studio shortcut, shared links management, and the danger zone.
Notifications: master switch plus sound and desktop channels.
Files and drive: your context, your control
A file explorer provides two roots: workspace storage and a personal per-user drive. The folder browser supports create, rename, copy, delete, upload, download, and image preview. A folder picker attaches a folder, including a real folder on your PC, as conversation context. Drag items from the drive into the composer to use them in a turn.
Your files are not an afterthought. They are first-class context for your conversations.
Sharing: you decide what goes public
Shared links turn a conversation into a read-only public link. Enable, disable, delete, and manage them from a shared links library in settings. You decide what becomes public and what stays private. Nothing is shared unless you choose to share it.
Account and presence
Avatar and banner upload (avatar up to 5 MB; jpeg, png, gif, webp, or svg), display name, username handle, and a presence status (online, away, and others). Theme choice persists per account. Full right-to-left layout with Persian and Arabic typography, plus Latin scripts in the same interface. The studio works in your language, not just in English.
Lock screen: protect your session
A four-digit passcode overlay can gate the session at any time. An idle auto-lock timer, a lock action in the sidebar, and a password bypass for a forgotten code. The idle lock is enforced server-side, not only hidden in the UI. Your session stays protected even if someone walks up to an unlocked screen.
The model: QT-Pro
QTChat runs on QT-Pro, the default model for every account. QT-Pro is powered by Xiaomi MiMo-V2.6-Flash, a sparse Mixture of Experts model with 309 billion total parameters and 15 billion active per token. It offers a 1 million token context window and full omnimodal support across text, image, video, and audio. The model delivers strong reasoning, coding, and agentic capabilities at a cost that keeps the studio accessible.
Raw capability is only the starting point. QT-Pro is aligned, fine-tuned, and evaluated through rigorous protocols. Multi-layered safety filters operate at every stage of the response pipeline. Curated dataset fine-tuning elevates reasoning depth and factual accuracy. Proprietary prompt architectures shape consistent, high-quality interactions. Domain specialization enhances performance across coding, analysis, creative writing, and technical documentation. Conversational intelligence maintains coherence across extended exchanges. Output consistency ensures predictable, professional-grade results.
A high-throughput evaluation framework processes over ten thousand inference rounds weekly. Adversarial probing exposes failure modes and hallucination tendencies. Regression suites ensure improvements do not degrade existing capabilities. Human-in-the-loop review captures nuanced quality judgments that automated metrics miss.
On top of the model sits Jev, the decision intelligence layer. Jev analyzes each query to determine the optimal response strategy, depth, and approach. It calibrates confidence scores and triggers supplementary reasoning when needed. It captures quality signals from every interaction and feeds them back into the fine-tuning pipeline, creating a virtuous cycle where real-world usage continuously informs model improvement. Jev dynamically adjusts response depth and analytical stance based on query context, and coordinates specialized reasoning pathways for complex multi-domain questions.
Privacy and access
QuantumThreads is invitation-only. Access runs through a waitlist and manual approval. Registration collects name, email, age (must be 18 or older), password, terms consent, and a visual human check. Registering does not create a session. The account lands on the waitlist hold until reviewed and the email is verified. On approval, the user signs in and reaches the studio.
Conversations may be used to train and refine the app and its models. Conversation data is never sold and never shared with third parties. End-to-end encryption is not implemented yet, and the terms say so plainly. Keys are write-only and never re-displayed after entry.
The interface is Persian-first and right-to-left aware with full Latin support. Fonts and core assets are self-hosted. The build has no dependency on an external font CDN, which keeps it reliable on restricted networks.
Who QTChat is for
Individual builders, writers, researchers, and designers who want one private studio for text and image model work. Small teams that want shared workspaces plus direct messaging without adopting a general SaaS chat tool. Privacy-minded operators who prefer approval-gated access, bring-your-own providers, and self-hosted assets. Persian and Arabic speaking users who need a serious RTL-first interface rather than a translated afterthought.
QTChat is not for everyone. It is for people who want their AI conversations to last, to be organized, and to be theirs.
Get started
QuantumThreads is a private, local-first AI studio for text and image work. Access is by invitation through a waitlist. If you want to see how the model shows its work, request access at chat.quantumthreads.it