Listen to any completed assistant reply without leaving the conversation.

Select Read aloud beside a message to hear it in a natural voice. Select Stop at any time, or start another message to switch playback.

Search is now built into every workspace.

Use Search to find what you need without retracing your steps, then jump straight back into the relevant work.

Video generation across the workspace is now more capable. Based on Arena benchmarks, it is 50% better than our previous generation.

Create stronger videos from a prompt, source material, or an existing project with our improved state-of-the-art video workflow.

Four new model families are joining the picker.

#Kimi K3, hosted your way

Choose a China-hosted or US-hosted Kimi K3 endpoint based on where you want it to run.

#Meta Muse Spark 1.2

Choose data sharing for a 75% lower price, or keep data sharing off at the regular price.

#DeepSeek V4 and Opus 5

DeepSeek V4 and Opus 5 are available alongside the rest of your models, with the same workspace, tools, and memory.

Pin important messages in chat so decisions, instructions, and useful answers stay easy to reach.

#A clearer Plan UI

Follow the agent's plan as work progresses, with a clearer view of what is active, complete, and still ahead.

#Stable model selection

Your selected model now stays consistent more reliably as you move through the workspace and continue conversations.

This release also brings many smaller UX improvements across the workspace, making everyday work feel smoother and more polished.

Turn a long conversation into a clear, chronological annotation thread without copying, reformatting, or losing your place.

#Keep every excerpt in order

Select text from a message and choose Quote. The excerpt is appended to your current draft without replacing anything you already wrote.

Keep quoting as you move through the conversation. Each excerpt stays in the order you collect it, so you can build a point-by-point response from multiple moments in the thread.

#Add your annotation

After each quote, the cursor is ready for your note. The Markdown-aware composer keeps quotes, headings, lists, emphasis, and code clear while you write.

Connect your existing Grok subscription from the Subscriptions tab.

#Your subscription, enhanced

Sign in securely with your X account, formerly Twitter. Your available Grok models load automatically and stay refreshed, ready to use in the model picker.

You keep using your existing subscription. SPIRITT adds a real workspace, tools, memory, and agent capabilities, so Grok can do more than in the native experience.

Disconnect any time.

Deep Research turns one goal into a coordinated effort that plans, hunts, verifies, and delivers a cited answer.

#The right model mix, automatically

Choose any model or any mixture. By default, SPIRITT automatically assembles the ideal model mix, using each model where it is strongest.

In internal evaluations, it took first place on every publicly known and verified dedicated deep-research benchmark we tested.

#Goal-driven when you choose it

Turn on goal-driven research when the outcome matters more than a fixed depth. It finds the gaps, launches new waves of research, and keeps going until the goal is reached.

#One investigation, every source

Combine the open web, connected integrations, personal accounts, and enterprise systems in a single investigation.

Enterprises uncover hidden insights across huge data estates. Startups find leads across the web and their CRM, then turn research directly into outreach and deals.

Today, we are making SPIRITT integrations much more robust and accessible, with many new integrations, features, and enterprise-level control and visibility.

#Everything visible, everything covered

See every connection, status, authorization, and scope in one place. Use thousands of ready-made and AI-built custom integrations. If something is missing, choose Create with SPIRITT and SPIRITT builds it end to end.

#Personal speed, enterprise control

Connect to one workspace in a click, or let owners and admins authorize access across the organization.

Set action-level rules for Read, Write, and Destructive operations: Always allow, Ask every time, or Never allow. Sensitive actions wait for a time-limited, single-use approval.

#Triggers work both ways

Agents can act inside connected apps, and apps can wake agents with triggers and nudges. A new lead, ticket, payment, calendar event, or deal update can start a durable workflow automatically, protected by the same permissions and approvals.

The workspace now handles complex answers more clearly and feels more polished across everyday interactions.

#A richer chat experience

Diagrams, formulas, highlighted code, complete diffs, and rich links now appear directly in your conversations.

#Polish throughout the workspace

We shipped dozens of small UI fixes and micro-interactions across chat, scrolling, messages, the sidebar, integrations, desktop glass, and onboarding. Together, they make the workspace feel calmer, faster, and more natural to use.

UI polish is a constant process for us. We improve it continuously, and when a wave of changes becomes meaningful enough, we collect it into a changelog update like this one.

Chat now follows the natural direction of every message more closely.

Right-to-left languages such as Arabic, Hebrew, and Persian align as expected, while conversations that mix right-to-left and left-to-right content keep phrases, links, code, numbers, lists, and tables easy to follow. The composer adapts as you type, too, so switching direction feels seamless.

A small UX change with a big benefit for people who work heavily with media.

Tap or click images to open them full screen with their original shape preserved. Videos keep familiar native playback, picture-in-picture, and full-screen controls.

Kay 1.4 is our latest generation of intelligence across every tier. You’ll see stronger reasoning, coding, and long-running agent work, all inside the workspace you already use.

#Existing conversations upgrade automatically

Open an earlier Kay conversation and it continues on the equivalent upgraded tier. Your history, files, tools, and memory remain in place, so there’s nothing to migrate yourself.

#The same range, now more capable

The Kay family still spans quick everyday work through frontier intelligence for the hardest problems. Pick the pace and capability you need; every tier benefits from the upgrade.

#How to choose the right Kay tier

For the best balance of capability, speed, and credit usage, we recommend:

• Kay — the recommended default for most tasks.
• Kay Nano — ideal for time-sensitive work where speed and efficient credit usage matter most.
• Kay Pro — suited to complex tasks that benefit from deeper reasoning.
• Kay Max — reserved for the most demanding work, when the highest level of capability justifies higher credit usage, or when another tier cannot make progress.

Subscription plans are live for workspaces, with new monthly and yearly options built for teams that use SPIRITT regularly.

#Recurring credits

Choose a monthly or yearly plan and your workspace gets a fresh credit allowance every month. Yearly plans include a big discount, paid upfront while credits still refresh monthly so usage stays predictable.

#Top-ups still work

Need extra room before the next refresh? One-time top-ups are still available and now sit clearly below subscription value, so subscriptions are the better default for ongoing work.

#Manage it yourself

Checkout, plan changes, cancellation, reactivation, and billing receipts all stay self-serve from the billing page.

Meta Muse 1.1 is live in the model picker.

We enhance Muse with a real workspace, tools, and memory, so it outperforms the native Muse experience.

Grok 4.5 is available in the model picker.

We enhance Grok with a real workspace, tools, and memory, so it outperforms the native Grok experience.

Long agent tasks now spend less.

The workspace reuses stable context across a task instead of re-paying for the same ground every turn. On real work, that can cut credit use by up to about 60% for the same task.

You get the same quality of work. Your credits just go further.

Connect the AI accounts you already pay for and keep using that same subscription budget here.

#Claude Max

Connect Claude Max from the model picker. Full agent turns, live streaming, history, and a dedicated Subscriptions tab so you can see when it's connected.

#ChatGPT

Connect ChatGPT the same way. Supported models show up across chat modes, with clearer reconnect status when the connection needs a refresh.

Your Claude or ChatGPT budget stays on that account. We enhance them with a real workspace, tools, memory, and end-to-end work, so they outperform their native apps.

The model picker keeps growing. Recent additions include GLM 5.2 Fast, DeepSeek V4 Pro, Kimi K2.7 Code, Gemma 4, Gemini 3.5 Flash, and Claude Opus 4.8.

Every model here is enhanced with a real workspace, tools, and memory, so it outperforms the native experience. Extra models also show short descriptions and brand icons, so choosing one is faster.

Claude Fable is live in the model picker.

We enhance Fable with a real workspace, tools, and memory, so it outperforms the native Claude experience.

SPIRITT Desktop puts your workspace on your computer.

#What you get

• Full workspace inside a desktop shell
• Approve specific folders so the agent can work with files on your machine
• First-run onboarding
• Signed installs and in-app update checks
• Public download links

Workspaces can do more end-to-end work for you.

#Shop real products

Search real Shopify stores and get a direct checkout link.

#Ship apps faster

Building full apps is simpler and more reliable.

#Scheduled jobs

Turn one-off work into recurring automations, and see their status in the jobs panel.

#Download what you built

Files your workspace creates are easy to download.

The workspace chat feels faster and cleaner.

#Performance

• Instant chat switching with no reload or stale flash
• Faster streaming answers
• Faster new workspaces and cold starts
• Faster workspace UI updates

#Design

• Cleaner chat shell, composer, and tool activity
• Clearer tool steps
• Clean chat copy that pastes as plain text
• Embed chat mode for hosting chat in other products

#Mobile

• Better mobile startup and recovery
• Composer works with on-screen keyboards
• Content clears the iPhone notch and home bar
• Full-screen mobile sidebar

Every workspace can use the browser.

Browser automation is built in, starts when needed, and can reach more public sites that usually block bots. Live control is more reliable, and browser-powered apps can get stable public links.

Chat can take more of the work you already do.

#Files

Drag and drop files into the workspace, upload them as chat attachments, and see upload progress. Image-capable models can see the images you attach.

#Voice

Talk into the composer with native voice input.

#Images

Ask your workspace to create images from chat.

#Clearer activity

Live activity while the agent works, expandable tool steps, and clearer “out of credits” states.

Less friction between intent and work.

#Landing prompt into chat

A prompt typed on the site becomes the first workspace message.

#Drafts that stick

Composer drafts survive refresh, remounts, and per-chat switching.

#Pick a project folder

Start a chat in a specific working directory.

#Return after top-up

After buying credits, you land back in the same conversation.