Mume AIMume AI

Changelog

New updates and improvements to Mume AI.

v1.10.0August 2026
  • Four new agents, each at its own address: CSV Analyst reads a spreadsheet and answers questions about it, Repo Explainer reads a codebase and writes what it finds, Test Writer writes tests and runs them, and Voice-over Translator turns a recording into another language and speaks it back
  • A new Agents directory at /agents, with a page for each agent covering how it works, how it is set up, and what it will not do
  • A terminal under the file pane, and a Run button on Python, JavaScript, TypeScript and shell files — it saves first, runs in the same sandbox the agent uses, and can be minimised or closed
  • Any file can now go into a workspace by drag and drop, paste, or the file picker — not just images, PDFs and audio
  • Chess plays like a chessboard: click to move, a dot on every square the piece can reach, a promotion picker instead of an automatic queen, and captured pieces with a material score
  • The Files panel is now in the same place in every agent — at the far right, closed until you ask for it
  • Agents keep your workspace tidy, putting scripts, charts and notes in folders rather than leaving everything in the root
  • Every agent now opens with its own introduction and starter prompts, instead of a generic greeting
  • New API docs for Speech and Transcription, Embeddings and Reranking, Moderation, and reading your credits, usage and gateway status from an API key
  • Fixed: an agent asked to generate speech, an image or a video failed every time — a configuration difference between the two services that had been there since those tools were written
  • Fixed: some replies came back without their token count and cost, including every reply you stopped part-way
  • Fixed: cancelling a run left its last tool call spinning forever instead of saying it had been stopped
v1.9.0August 2026
  • Model pages now show independent benchmarks — intelligence, coding and agentic scores, GPQA Diamond, τ-bench, and Design Arena rankings across UI, 3D, data viz, code and websites
  • Every score carries its error bar, how many tasks it ran over, and what it cost per task
  • Sources are credited and dated, so you can see how current a score is
  • New Specifications section on every model — supported parameters, knowledge cutoff, tokenizer, max output, cached input pricing and moderation status
  • Models scheduled for retirement now say so, with the date, so you can migrate before requests start failing
  • Sort the model catalog by intelligence, not just price and context
  • Model descriptions render their links properly instead of showing raw markdown
  • Per-key spending limits — cap what any API key can spend per day or per month, without touching your balance
  • Per-key usage on the dashboard: requests, tokens and spend for the last 24 hours and the current month
  • The dashboard now fits properly on a phone
  • API: /v1/models returns supported_parameters, knowledge_cutoff and an intelligence score
v1.8.0July 2026
  • Pay as you go — buy credits from ₹100, without a subscription
  • New blog at /blog — product updates and engineering write-ups, with RSS
  • Any purchased credit unlocks the Plus-exclusive models, and keeps them unlocked while you hold a balance
  • Purchased credit stays valid for 12 months; the free daily credit still resets each day
  • Settings, credits and billing activity are now one dialog instead of three separate screens
  • Your balance is the same number everywhere — settings, dashboard and usage previously disagreed
  • Running out of credit now offers a top-up instead of telling subscribers to wait for their renewal
  • Credit purchases appear in the activity log as grants, with the amount paid
  • Pages load in one step rather than revealing in pieces, and no longer flash 'Free' before your actual plan
  • Model and author pages are served from the CDN, so they open immediately
v1.7.0July 2026
  • Media Studio redesigned — chat-style composer with the model, aspect ratio and send controls in one bar
  • Your latest generation now fills the canvas, with the last three a click away above it
  • Reference images: edit an image, or animate one as the first frame of a video, on models that support it
  • Reuse any generation as the reference for the next one
  • Media Studio is now browsable without signing in
  • New API docs for image and video generation
  • Live per-model pricing shown before you generate
  • More reliable video generation — jobs now recover if a provider callback is missed
v1.6.0July 2026
  • Media Studio — generate images and videos on the web at /media
  • Generate images and videos directly inside a chat
  • Rebuilt chat engine shared with the mobile app, so sessions stay in sync everywhere
  • Videos are generated as background jobs with live status
  • Every generation is saved to a personal library, synced across web and mobile
  • Fixed tool calls failing on some MCP servers with a provider error
v1.5.5March 2026
  • Code editor with live preview
  • Video, audio and PDF support in the editor
  • Play chess against any model
  • Model search in the composer
  • Credit usage log
v1.5.0February 2026
  • Web app launch with full chat interface
  • Multi-model support — 300+ AI models
  • MCP server integration
  • File attachments and workspace support
  • Multi-chat sessions with branching
  • Bookmarkable and renameable chat sessions
v1.4.0January 2026
  • Extended thinking / reasoning mode for supported models
  • Image generation (mobile app)
  • Tool use with multiple categories
  • Improved session management and history
v1.3.0December 2025
  • Plus subscription with premium model access
  • Dashboard with usage analytics
  • Model explorer with search and filtering
  • Dark mode and theme support