Vani now remembers, connects and moves faster: Memory, Skills, Connectors and Fast mode

Our biggest update yet. Here's everything that's new since our last post.

Memory

Vani now remembers useful details from your conversations, so you don't have to repeat yourself.

• Memory used: replies that drew on something Vani remembered show it underneath, so you always know.
• You're in control: see, edit, add or delete what Vani remembers in Settings → Personalization, or turn memory off completely.
• Custom instructions: tell Vani about yourself and how you'd like it to respond.
• Bring your memory with you: import what ChatGPT, Claude or Gemini know about you, and export your memories at any time.

Skills

Skills are reusable instructions for tasks you do often, like a writing style, a review checklist or a weekly report.

• Type / in the message box to use a skill, or browse the Skills library.
• Start with 10 skills from Vani, or create your own.
• Bring your GPTs and Gems: import your custom GPTs from ChatGPT and your Gems from Gemini as Vani skills. A good moment, since OpenAI is retiring GPTs on December 11 and Google is changing Gems on November 17.

Connectors and your own MCP servers

• Connectors: connect Notion, Linear and Sentry in Settings → Connectors, then switch them on in any chat to let Vani read from them.
• Add your own MCP server: connect any remote MCP server. Free includes one.
• On the API too: use MCP tools with the Responses and Messages APIs.

Use Vani inside other assistants

Add https://mcp.vani.ai/mcp as a custom connector in Claude, ChatGPT and other assistants that support MCP, and use your Vani account to generate images and videos and run research from there. Setup takes a minute: see the guide. You can see and remove connected apps in Settings → Connected apps.

Fast mode for GPT models

• In chat: switch on Fast in the model menu for quicker replies from GPT models. Fast replies use 2.5× more of your allowance.
• On the API: fast versions are available as separate models, such as openai/gpt-6-astra-fast, priced at 2.5× the standard rate.

A simpler message box

Attachments, image and video creation, web search and deep research now live under one + menu, with shortcuts like /image, /video, /search and /research. Less clutter, same features.

Better answers from reasoning models

• No more blank replies: reasoning models now get enough room to think and still answer. If a reply comes back empty, Vani automatically tries again with lighter reasoning.
• Clear notes: if a long reply reaches its length limit, Vani tells you, so you can ask it to continue.
• Your effort setting is respected on every model.
• Links work again in messages to Claude Opus 5.5 and Claude Fable 5.1.

A better start for new users

New accounts now get a quick setup: pick what you'll use Vani for, and Vani suggests the right chat, image and video models, with a short tour and a checklist to get going. New chats now start on Claude Sonnet 5.5 by default; your own choices are always kept.

Your data

In Settings → Your data you can now download a copy of all your data, or delete your account. Deletion has a 7-day grace period in case you change your mind.

For developers

• Pay by card on the page: adding funds on platform.vani.ai now uses a simple card form right in the console.
• Saved cards and automatic top-up: save a card and top up automatically when your balance runs low, with limits you set.
• No charge for failed requests: if a request fails before you receive anything, it isn't charged.
• Better compatibility: more OpenAI and Anthropic request fields are accepted, including thinking, top_k, min_p, repetition_penalty and a top-level system prompt.
• Clearer errors: requests for a model that isn't available now get a clear model_not_available error.
• Prices kept up to date: model prices now follow our provider's rates automatically.

Model updates

• Muse Spark 1.3 from Meta is now available in chat and on the API.
• Usage rates: some models now use more or less of your allowance, following our provider's prices. GPT-6 Astra, GPT-5.5, GPT-5.6 Sol, GPT-5.6 Luna, GPT-6 Luna and Claude Haiku 4.5 use a little more; DeepSeek V4.1 Flash, MiniMax M3, Qwen 3.8 Flash, Qwen 3.8 Max and GPT-5.6 Terra use less.
• MiMo V2.5 Pro has been retired by its provider and is no longer available.
• GLM-5.3 Derisked has moved to new hardware after today's outage. Thanks for your patience.

Brand kit

Writing about Vani? Our logos, colours and guidelines are now at vani.ai/brandkit.

Fixes and improvements

• Adding a card on the API platform works again.
• Video requests in chat are adjusted to what each video model supports, instead of failing.
• Image generation and model routing are more reliable.

Coming soon

• Vani Studio: ready-made looks and camera moves for your photos and videos.
• Vani CLI, desktop apps and mobile apps.

We're still fixing bugs. If something doesn't look right, please use Report a bug in the app; we read every report.