Changelog
Milly Lab changelog
The Milly Lab changelog lists user-visible changes by the date they were completed, newest first, from June 2026 onwards. Recent changes include built-in skills for Tajik and Russian business letters, agents in Spaces, schedules, MCP servers, an OpenAI-compatible API on the Max and Advanced plans, two-factor authentication and a Tajik interface.
Last updated:
What changed in October 2026?
October 2026 brought agents to Spaces, 15 built-in skills, schedules, remote MCP servers, plugins, notebooks and the Console, and a chat that shows each step and tool call while the model works.
| Date | Change |
|---|---|
| Built-in skills, including Tajik and Russian business lettersFifteen built-in skills are on for every account, among them official letters in Tajik and Russian, e-mail drafting in Russian and Tajik, translation with a glossary, contract review for CIS contracts and research briefs. You can also write or upload skills of your own. | |
| SchedulesA prompt or an agent can run on a schedule. Each run opens a new chat or adds to a conversation you choose, and its result lands in the Tasks inbox. | |
| MCP servers and pluginsAdd any remote MCP server by its https address, with OAuth sign-in, a bearer token or custom headers, and use its tools in chats and agents. Plugins built for Claude, Codex and Gemini can be installed, switched on and off, and exported. | |
| The ConsoleAgents, skills, MCP servers, connected accounts, schedules, API keys, usage and the audit log are managed in one place: inside the app, or in the full Console in its own tab. | |
| NotebooksA memory collection can become a notebook that answers questions with citations to its sources and can record a short audio overview. | |
| Stop, Regenerate and retryStop now ends a reply on the server as well, Regenerate replaces the previous answer, and a reply that failed can be retried. | |
| Agents in SpacesSave an assistant with its own instructions, model and tools, call it with @ in any chat, or let Milly Lab draft one from your chats. Nothing becomes an agent until you review the draft. | |
| Chat shows its workWhile a model works, the reply shows its plan, each tool call, the files it creates and a status line. |
What changed in September 2026?
September 2026 added an OpenAI-compatible API with API keys for the Max, Advanced and Enterprise plans, two-factor authentication, sealed Spaces and memory collections shared with a team.
| Date | Change |
|---|---|
| OpenAI-compatible APIOn the Max, Advanced and Enterprise plans you can create API keys and call Milly Lab models through an OpenAI-compatible endpoint at https://api.m-lab.app/v1. API usage is charged to the same account as the app. | |
| Two-factor authenticationTurn on two-factor authentication in Settings → Security with an authenticator app. Signing in with a password or with Google then asks for a code, and recovery codes are shown once. | |
| Sealed Spaces, shared collections and memory importA Space can seal its memory, so its chats use only their own memory and nothing learned there appears in other chats. Memory collections can be shared with your team and imported from files. |
What changed in August 2026?
August 2026 added Tajik as the third interface language, memory across chats, file previews, models that read attached Office files, textured 3D models, public documentation and web search that runs only when a question needs it.
| Date | Change |
|---|---|
| Memory across chatsWith memory switched on, Milly Lab keeps short facts about you and summaries of past chats, marks the answers that used them, and lets you read, edit or forget each fact. Memory is part of the paid plans. | |
| File previewsClick a generated file's name to preview it, with presentations shown as slides, or use the download icon to save it. | |
| Models read your attachmentsAttach Word, Excel, PowerPoint or text files to a chat and the model reads their content. Each chat keeps its files together in one place, and files can be dragged into the chat. | |
| Tajik interfaceTajik became the third interface language after English and Russian, across the website and the app. | |
| Textured 3D models and image-to-3D3D Studio builds textured models from a text description or from an uploaded image. 3D Studio is part of the Advanced plan. | |
| Public documentationGuides to plans, features, security and the API are published at /docs. | |
| Web search when a question needs itOn paid plans the model searches the web only when a question needs current information, and you can switch search off for faster answers. |
What changed in July 2026?
July 2026 introduced the Free plan, a document engine for Word, PowerPoint, Excel and PDF files, GitHub repositories and Google Drive folders connected to Swarm Mind with staged changes, and password reset by e-mail.
| Date | Change |
|---|---|
| New GitHub repositories on ApplyA Project or Council session can put its changes into a new private GitHub repository, which is created only when you press Apply. | |
| Connected GitHub repositories and Google Drive foldersProject and Council sessions can read a connected GitHub repository or Google Drive folder. Every change they propose is staged, and nothing is written until you press Apply and choose the default branch or a new branch. | |
| Password reset by e-mailIf you forget your password, request a one-time link by e-mail and set a new password. | |
| Free planNew accounts start on the Free plan without a card: 10 messages per model every 4 hours with GPT-5.6 Luna and DeepSeek V4 Flash. | |
| Document engine with Excel filesA new document engine builds designed Word, PowerPoint, Excel and PDF files from a chat, with themes and charts. Document files are part of the Pro plan and higher. |
What changed in June 2026?
June 2026 introduced Council Mode and Project Mode, sign-in with Google, voice messages and the first downloadable documents from a chat.
| Date | Change |
|---|---|
| Council ModeSeveral models answer the same brief on separate cards. You can compare the cards, start a debate between them and merge the result into a final answer. | |
| Project ModeAn orchestrator model splits a larger task into subtasks, several models work on them, and the results are combined into one deliverable. | |
| Sign in with GoogleCreate an account or sign in with a Google account. | |
| Voice messagesRecord a request instead of typing it: when you stop recording, the speech is transcribed and sent. | |
| Documents from a chatAsk for a Word, PDF, PowerPoint or Markdown file and download it from the conversation. Document files are part of the Pro plan and higher. |