Changelog
Follow up on the latest improvements and updates.
RSS
New
Hatz Voice Mode
Any member can talk with Hatz in chat and hear it answer out loud. Click the wave icon in the chat box to start.
- Real work, by voice:ask for an artifact, a document, research, or something in a connected app, and Hatz does it as a normal chat turn while the conversation keeps going.
- Natural back-and-forth:talk any time to interrupt. Hatz listens, thinks, and replies out loud, with live captions under the orb.
- Voice commands:say "cancel that" to stop a running request, or "end voice" to finish.
- Every exchange is saved to the chat.PressEscorEnd voiceto go back to typing.
Download Menu for Artifacts
Any member viewing an artifact gets one
Download
menu, with the original file at the top and its export formats below.- Named files:downloads are named after the artifact's title, on public pages too.
- PDF:HTML apps download as PDF, and PowerPoint files can now download as PDF as well.
- Deck navigation:slide decks show a small Prev / Next control that tucks away while idle.
Improved
Research With Real Sources
When Hatz searches the web with Perplexity, it can bring back the actual pages and links, so answers cite sources people can open and check. Works in chats, agents, and phone agents with Perplexity turned on.
Role Settings in One Place
MSP admins manage a custom role from one
Role Settings
dialog under Customize Roles
, including moving several people onto a role at once.
User Rate Limiting
Requests made with Hatz API keys have limits on requests, tokens, and simultaneous calls for each key and the user and organization that own it. This is to improve reliability with our services. Using Hatz in the app isn't affected.
New
Refreshed Chat
Any member gets a redesigned chat, with new navigation, a new composer and appearance controls.
- Resizable sidebar:drag the sidebar to the width you want, or collapse it.
- Search every conversation:pressCmd+KorCtrl+Kto search chats, including older ones and chats inside folders.
- Appearance:chooseSystem,Light, orDarkfrom the account menu.
Microsoft 365: Agents Can Take Action
Agents connected to Microsoft 365 can now act in Outlook, OneDrive, and SharePoint, not just read from them.
- Send email:chat asks a person to approve each email before it goes out. Emails keep their formatting (bold, lists, links) in Outlook.
- Calendar:cancel meetings, and accept, tentatively accept, or decline invites.
- Files:upload to OneDrive and SharePoint, and share files with people inside the organization.
- Sharing outside the organization:a person approves each external share or "Anyone" link.
- SharePoint lists:add and update list items.
- Deletes are confirmed in chatbefore anything is removed.
- The organization's own Microsoft 365 sharing policies still apply. When Microsoft blocks a share, chat shows Microsoft's reason.
LLMs
Claude Sonnet 5.5
Claude Sonnet 5.5 was added to the model selector for chats, apps, agents, and workflows.
- Where Auto uses it:Performance mode, the default, sends its hardest tasks to Sonnet 5.5, and Turbo uses it for complex work, with the most demanding requests going to Claude Opus 5.5. Lighter coding tasks now go to GPT-6 Luna.
- Credits:1.25x per message, the same as Claude Sonnet 5.
- How it compares:Anthropic's benchmarks put it well ahead of Sonnet 5 on agentic coding and knowledge work, and close to Claude Opus 5.5 (2x).
- Good for:everyday work that needs a strong model: fixing bugs, and polished documents, slides, and spreadsheets.
DeepSeek V4.1 Flash
DeepSeek V4.1 Flash was added to the model selector for chats, apps, agents, and workflows. It's a fast, low-cost model for members who choose their own model, built for long documents, images, and tool use.
- Credits:0.33x per message, the lowest multiplier in Hatz, alongside GLM 5.3 Flash.
- How it compares:DeepSeek's benchmarks put it ahead of Kimi K3 and GLM-5.3, and level with or ahead of Claude Opus 5 and GPT-5.6 Sol on three of four agentic benchmarks.
- Good for:high-volume work where credits matter: summarizing long files, reading screenshots and images, and routine steps in agents and workflows.
- With Auto:Auto doesn't route to it today and already sends lighter work to low-multiplier models like GPT-6 Luna and GLM 5.3 Flash. Choose DeepSeek V4.1 Flash fromAll Modelswhen you want it specifically.
Retired models:
- Kimi K2.6, Kimi K2.7-Code, Inkling, and DeepSeek V4 Pro have been retired by their provider. Any agent, workflow, or default that used one now uses a replacement automatically, and DeepSeek V4 Pro moves to DeepSeek V4.1 Flash.
Presenter Tools for Artifacts
When you present an artifact full screen, a small toolbar now offers
Cursor
, Laser
, and Pen
, so you can point at and draw on what you're showing.- Slide decks:each slide keeps its own drawing. Move to the next slide and it starts clean; go back and your marks are still there.
- Clear and undowork on the current slide only.
- Drawings are temporary:they're never saved into the artifact or its downloads, and they clear when you leave full screen.
Improved
Timestamps and Thinking in Chat
While Hatz works on a reply, a status line now sits right under it in the conversation, showing what it's doing (thinking, using tools, responding) and how long it's been going. When it finishes, the reply shows how long Hatz thought.
MSP Admins
Activate: Add Your Own Use Cases
In
Activate
, a customer's Use Cases
tab now has an Add use cases
button. Pick Workshop items
you've shared with your tenants, write your own, or have AI suggest one.- Enrich this item:Hatz researches the use case in about a minute, then fills in its hours, value, and talking points from what it finds.
- Into the pitch:enriched use cases appear in the pitch you generate. You can leave one out of the proposal, though including it is recommended.
- Who Has Access:MSP admins using Activate.
Custom MCP Servers With Public OAuth Apps
- Members adding a custom MCP server can connect one whose OAuth app doesn't use a client secret. In Add Custom MCP Server, chooseOAuth, thenI have credentials, thenPublic client (no secret).
New
GPT-6 Sol and GPT-6 Luna
"More ways to bring frontier intelligence into the work you do every day." — OpenAI
Any member can select OpenAI's two newest GPT-6 models in chat, agents, and workflows.
- GPT-6 Solis built for complex reasoning, coding, and agentic work.
- GPT-6 Lunais the fastest GPT-6 model, made for high-volume everyday work like summarizing, drafting, and routine automation.
- Both have a 1.05M-token context window and up to 128K tokens of output.
Grok 4.6 and Grok 4.7
"Grok 4.7 is our most capable model for coding and knowledge work." — SpaceXAI
Any member can select xAI's two newest Grok models in chat, agents, and workflows.
- Grok 4.7is xAI's most capable Grok model for complex reasoning, coding, and agentic work.
- Grok 4.6is built for coding, agentic tasks, and knowledge work.
- Both read images and offer configurable reasoning. A custom role that blocks xAI models blocks both.
Claude Opus 5.5
"Opus 5.5 is a major step up from Opus 5. It's the new leading model." — Anthropic
Any member can select Claude Opus 5.5 in chat, agents, and workflows. It is Anthropic's latest Opus model for long-running agentic coding and knowledge work, with a 1M-token context window and adaptive thinking always on.
MSP Admins
Activate: Ready-Made Use Cases for Every Prospect
A prospect's
Use Cases
tab in Activate opens on five Hatz Official use cases, each modeled for that company's size.- Each card shows the hours and dollars a year it gives back, and the totals update when the headcount under Modeled forchanges.
- Open a card for talking points (Hook, Proof, Provide, Objection), tips to get started, and How we got this, the math and sources behind the estimate.
- Generate pitchcreates a branded PDF with the same numbers to help guide conversations with the prospect.
- Creating a tenant from a pitched prospect copies the included use cases into the new tenant, and its AI Quickstart Checklist lists their tips as first steps.
New
Weekly Credit Limits
MSP admins can set a weekly credit limit on a custom role, separate from the monthly one, and choose what happens when someone reaches it.
For MSP admins
- Pick the role, set the weekly credit amount, and choose the behavior at the limit: Block or Auto only.
- Blockpauses new AI requests until the limit resets. Anything already running finishes.
- Autoonlykeeps people working on the lower-cost Auto tiers you choose, instead of stopping them.
- Weekly limits reset every Wednesday at 12:00 PM Eastern.
For users with a weekly limit
- You get an in-app warning at 80%, and again at 95%, before anything changes.
- At the limit you either pause until Wednesday, or switch to your admin's approved Auto models with a banner showing which models and when your limit resets.
- Your account menu shows credits used, your limit, and the exact reset time.
- Request more sends your admin a request directly.
LinkedIn MCP Integration
Any member can connect their LinkedIn account and post from chat — through the
Tool Selector
, or under Workshop → Integrations → LinkedIn
.- Post text, pictures, videos, links, or polls through Hatz.
- Hatz uploads your files. Hand it a JPG, PNG, GIF, or MP4 — up to 20 pictures in one post.
- Links post as a card, with a headline you choose.
- Polls take up to four options.
- Posts go to your connections unless you ask for public.
- Delete a post by prompting Hatz — it confirms which one before removing it.
Larger File Uploads
Any member can attach files up to
75 MB
— documents, images, audio, and archives. Full decks, recorded calls, and scanned document sets go in whole, in chat and as sources for agents and apps.Your Hatz AI-as-a-Service Status
- Enrolled MSP partners see their current Hatz business status and first sale window on the dashboard beside credit usage.
- The four tiers are Getting Started, Launched, Proven Model, and Established AI Practice. Tiers are measured by account growth in relation to the Hatz partner community.
- The window shows time left to first sale. It turns urgent inside the last two weeks, as well as shows status as they build their AI-as-a-Service practice.
Fixed
- Files in a conversation: Revising a generated file replaces it, so one current version is left rather than two with the same name.
- Microsoft 365 calendar: Calendar reads come back in the time zone the caller asked for.
- Autotask fields: Autotask picklist fields read as their labels, so a synced record shows the same words your PSA shows.
New
Shared Credentials now cover OAuth
MSP and tenant admins can connect an integration through their organization's own OAuth app, and it applies across the workspace.
- Shared custom MCP servers can be set up with OAuth as their sign-in method, with a redirect URI you can copy straight out of the dialog.
- Configure, rotate, remove, and check an app's status from Integrations & Tools.
More in Hatz Learn Resources
Everyone sees:
Hatz YouTube Channel
For MSP admins, under
Learn → Resources
in the Admin dashboard:- Roundtable: registration for the MSP roundtable
- Feature Friday: registration for the weekly feature session
- AI Adoption Leader Bootcamp: the full video playlist
- AI-as-a-Service Checklist
- MSP Go-To-Market Toolkit
- Trust Center
MSP Admins
: Activate ImprovementsRemove a Prospect From Your List
MSP users with a prospect-management permission can deactivate a PSA-sources prospect from the customer table or their company page, after confirming the exact prospect.
- Company details, contacts, and source presence keep syncing, so your next PSA sync leaves the prospect off the list.
- Recovery re-enables a prospect at any time, preserving source exclusions and restoring public links that were otherwise valid.
Activate: Shadow AI Scans
MSP admins can run device-discovery scans that finish in about three seconds, with detection counts reflecting distinct devices.
- A new scan can start for a prospect at any time.
- Scans past their 14-day link show an Expired label.
- Collector setup covers one platform at a time — Windows, macOS, or Linux.
- The collector script reports to the environment it came from, and rate limits scale for larger fleet rollouts.
Activate: PSA Reconnect
Reconnecting a PSA integration — ConnectWise, Autotask, or Halo — keeps your existing connection in place until the new credentials are confirmed.
New
Hatz Learn Resources
Hatz Learn has a Resources tab, with its own set of guides, walkthroughs, and videos for each audience. Everything opens in place — videos play on the page, and the reference material opens as a Hatz artifact.
For your team, under
Learn → Resources
:- Hatz AI Glossary— Hatz and AI terminology in one place, for anyone meeting the platform for the first time.
- Universal Use Cases— practical ways teams put AI to work, to hand someone who is not sure where to start.
- Hatz Tower— a dungeon crawler game built in Hatz that puts your AI knowledge to the test
- What Is Hatz?— a short video introduction.
- LAB (Learn And Build)— register for the Learn And Build webinar.
For MSP admins, under
Learn → Resources
in the Admin dashboard:- Admin Training— administering Hatz for the companies you manage.
- Hatz Activate Demo— a guided walkthrough of Activate.
- Hatz Activate Webinar— how Activate supports MSP growth.
The two sets are kept apart: your clients see their own Resources tab, and the MSP enablement material stays in the Admin dashboard.
Gemini 3.8 Flash
Google's newest Flash model is in the model selector for chat, apps, agents, and workflows, with a 1-million-token context window and a 65,536-token output ceiling.
- Google reports 61.4% on Vals Finance Agent v2 for financial analyst tasks, against 59.0% for Gemini 3.7 Flash, and 89.4% on Terminal-bench 2.1 for agentic terminal coding, against 85.8%.
Claude Fable 5.1
Anthropic's newest Fable model is in the model selector for chat, agents, and workflows, with a 1-million-token context window, a 128,000-token output ceiling, and adaptive thinking always on.
- Anthropic reports 52.6% on Terminal-Bench-Science 0.1 against 24.7% for Claude Fable 5, and 55.8% on Terminal-Bench 4.0 against 42.0%.
- Vision and tool calling are supported, so images and multi-step tool work stay in the same conversation.
- Auto routing sends the most demanding text work to it.
GLM 5.3
Z.ai's full-size GLM 5.3 joins GLM-5.3-Flash in the model selector for chat, agents, and workflows, with a 1-million-token context window and a 262,144-token output ceiling.
- Built for complex system engineering and long-running agent work.
- Z.ai reports a 50% gain over GLM 5.2 on its own Code Bench, and Terminal-Bench 3.0 moving from 4.6 to 28.3.
MSP Admins
Activate: Shadow AI Scans are Easier to Deploy Through Your RMM
- The collector download is named for the prospect it belongs to, and the prospect's name appears in the script itself, so a batch of scans is easier to follow.
- Setup splits into two labelled paths: deploy across a fleet through your RMM, and includes backup instructions to deploy by hand.
- macOS and Linux scans now work with NinjaOne System/root jobs — Open an existing scan, click Deploy to more devices, download the updated macOS/Linux collector, and replace the script in NinjaOne.
- The Windows collector runs under Syncro, and the macOS and Linux collectors run as System or root, so RMM-launched jobs report their devices back.
- Windows scans now work through Syncro — Open an existing scan, click Deploy to more devices, download the updated Windows collector, and replace the script in Syncro.
Activate: PSA Sync Updates
- Autotask sync is more reliable for MSPs with a large number of companies
- Pressing Sync while a run is already going tells you the run in flight and when it started.
- Faster PSA Sync for ConnectWise (no more timeouts)
Who Has Access:
Primary Admins, Admins, and Sales Users.New
GLM-5.3-Flash is now in the model selector for chat, apps, agents, and workflows.
- Approaches Claude Opus 4.8 on coding and agentic benchmarks — near-frontier intelligence in a much smaller, faster model.
- Built for speed, so it responds quickly even on demanding tasks.
- Extremely affordable and makes your credits go a lot further.
"
GLM-5.3-Flash shows that frontier intelligence does not have to come at frontier cost.
" — ZAI
Request a Phone Agent From Your MSP
- Tenant admins can now request a phone agent directly, using the same setup process MSP admins use — minus the number selection which occurs after approval.
- Primary admin and billing managers receive the request via email and in the phone agents page in their tenant. hey can review the exact configuration the tenant submitted, then approve or decline.
- The billing cycle begins once admins approve the request. Provisioning is official and the tenant is notified only once admins pick the phone number.
- Who Has Access:tenant admins can request; MSP admins with billing rights review and approve. Available for MSPs on per-minute phone billing.
Phone Agent: New Create Flow
Creating a phone agent starts with one screen: pick a curated template, copy one of your existing agents, or start from scratch.
- Templates ask a few setup questions specific to that template — business hours, emergency escalation, and the like — and write your answers straight into the agent's instructions.
Play Call Recordings in the App
Every recorded call now has a play control right in the call log, so you can listen to the whole conversation without needing a summary email.
Hatz Learn Moved In-House for All Tenants
- Hatz Learn now runs natively inside your tenant dashboard — no more logging into a separate TalentLMS site for training.
- Completed courses and certifications carried over automatically, so your record stays intact.
- If you were partway through a course when the move happened, you'll start it fresh in Hatz Learn — nothing is marked complete that wasn't.
- Find it directly from the Learn tab in your dashboard nav.
Improved
Compliance & Logs: Filter by Multiple Event Types
You can now select more than one event type at once when filtering Compliance & Logs, so a single search covers everything you're looking for.
MSP Admins
Various improvements to Activate, including:
Choose Which Companies Import From Your PSA:
Control exactly what gets imported instead of an all-or-nothing sync. Filter by each PSA's own live classifications, preview how many companies match, and only those sync going forward.- ConnectWise types and statuses
- Autotask company types
- Halo customer relationships
Edit Imported Company Details:
- Correct a company's name, domain, or primary contact right after import — your edits stick even through a later PSA resync, and updating the domain re-anchors matching and can kick off a fresh enrichment run.
Higher Margin Projections:
- The Revenue Calculator in Hatz Activate now supports margin inputs up to 500%, so you can model pricing that matches your services.
Fixes
- PSA Reconnects No Longer Duplicate Your Customer List:Disconnect and reconnect the same PSA instance and your existing imported customers carry over instead of duplicating — the filtering above still applies on every resync.
- Clearer Resource Access:The Resources button on a customer profile in Hatz Activate is labeled View Resources once resources exist, and opens straight to the Resources tab.
- HaloPSA Sync:Full Contact Details: Contact name, email, and domain now populate for every HaloPSA-synced customer, and a sync automatically refreshes its own access to keep running to completion.
New
Shared Integration Credentials
An admin connects a tool once for the whole workspace, and the key stays with them — nobody else ever sees it, and nobody else has to connect their own account.
- Works for any tool that uses a shared secret, like an API key or bearer token.
- Tenant admins, and MSP admins working inside a managed tenant, set them up under Settings → Integrations & Tools → Shared Credentials.
- Shared connections appear automatically for every workspace member, badged as managed by your organization.
- Anyone who already connected their own account keeps it — it sits dormant while the shared one is in use.
- Custom MCP servers work the same way. A personal server that a shared one replaces is marked Superseded.
Gemini 3.7 Flash
Google's newest Flash model is in the model selector for chat, apps, agents, and workflows, with a 64,000-token output ceiling.
"Our most intelligent workhorse model yet for coding and agents."
— Google- Google reports 43.6% on FrontierCode 1.1 Main for production code quality, against 34.4% for Gemini 3.6 Flash.
DeepSeek V4 Pro 0813
The dated 0813 release of DeepSeek V4 Pro is in the model selector for chat, apps, agents, and workflows, with reasoning, tool use, and native JSON Schema output.
"Major Agent upgrades with strong production gains."
— DeepSeek
Phone Agents: Live Look-Ups and Approved Callers
Two additions, both in the agent's settings.
- Look things up mid-call.An agent can search the web, read a page, check maps, or check the weather while it's on a call — so it can answer questions its instructions don't cover. Pick which of the seven tools each agent may use.
- Approved callers only.Turn it on and only numbers on the list reach the agent; everyone else hears a short notice and the call ends. Add numbers by hand, paste a batch, or import a CSV, and approve a caller straight from the call log.
Improved
Ask for Extra Usage Where You Hit the Limit
When your organization reaches its credit limit, "Request extra usage" is right there on the screen you were working in — chat, the workflow, agent, and app editors, and the workspace banner.
- One request covers every surface. File it from one place and the rest show it as already pending.
- Client admins see the button. Everyone else sees the explanation on its own, so nobody clicks a request that can't go through.
- Who Has Access:tenant client admins whose organization is on usage-based billing. MSP admins receive these requests rather than sending them, and enabling Extra Usage for a tenant removes the prompt.
MSP Admins
Your Client's Name on Their Sign-In Emails
Welcome, password reset, invite, and one-time-code emails arrive from Hatz AI (Client Organization), so a recipient recognizes which business the message is about before opening it.
New
Send Attachments On Their Own
Attach a file and send it with no message at all. The conversation answers from the file plus everything already in the thread.
- Attachments clear once they're sent, so your next message doesn't re-send the same source or spend credits on it a second time.
- Draft text stays with the chat you typed it in, rather than following you into whichever chat you open next.
Phone Agent Drafts
Step away mid-setup and your work is waiting when you come back. Leaving a phone agent's create flow asks first — save as a draft, discard, or keep editing — on every exit: the X, Escape, Cancel, or clicking the backdrop.
- Pick a saved draft back up and finish where you stopped.
Phone Agent Silence Controls
Set how long a phone agent sits through a quiet stretch before it re-engages, configured per agent.
- Choose any tolerance from 30 seconds up to 5 minutes, or switch silence detection off entirely for a given agent.
- Built for note-taker and walkthrough lines, where a caller is meant to go quiet or step away from the phone to do something.
- Existing agents keep the 30-second default, so nothing changes unless you change it.
Improved
Cleaner Document Output
PDF and PowerPoint generation delivers exactly the deliverables you asked for. Page renders, extracted text, thumbnails, form metadata, and draft decks resolve behind the scenes, so a request for one deck returns one deck.
Accurate Usage Status for Managed Tenants
On a managed tenant's overview, the Package credit usage card reads that tenant's own extra-usage budget. A tenant whose overflow is being absorbed reads as running normally, so the card matches what the tenant is actually experiencing.
Fixed
- Phone agent model selector— A phone agent powered by a Workshop agent shows that agent in its settings, and its dialogs name the action they perform.
MSP Admins
Tenant Renaming
Rename a tenant from Organization Settings — the Tenant name card sits directly above Branding.
- Who Has Access:MSP admins, from the Admin dashboard.
Invoices Over the API
Pull your invoices straight from the Hatz API so they land in your own AP system or archive, without anyone clicking through the dashboard.
- List them newest first, filter by issue date, and page through with a cursor.
- Fetch the hosted PDF link for any single invoice.
- Date filters cover the whole day in UTC, so an invoice issued late on your end date is still included.
- Authenticate with an API key or your existing session.
Who Has Access:
users with billing permission on the entity. If your account carries billing access on more than one entity, name the one you're asking about.New
DeepSeek V4 Flash 0731
is now in the model selector for chat, apps, agents, and workflows. - The model's 1-million-token context window makes it a fast, long-context model built for coding and multi-step tool use — a good pick when there's lots of material to work through.
Self-Reporting Credit Usage
Ask about your own credit usage right in chat and get an answer immediately — no admin access required.
Phone Agent: Pause, Duplicate, Reassign
Three new admin capabilities for phone agents:
- Pauseputs an agent on do-not-disturb. It stops answering calls but keeps its number and its call history, and you can bring it back anytime.
- Duplicatemeans you can copy a configured agent onto a new number. Build one front-desk attendant, then replicate it across as many numbers as a client needs instead of setting each one up from scratch.
- Reassignmoves an agent to a different tenant under the same MSP, keeping its number.
Improved
Present an Artifact Fullscreen
- Artifacts have a Present control in the viewer header. It puts the artifact into full browser fullscreen and hides the header and chat pane, so what's left on screen is just the artifact. Press Escape to come back.
Auto-Tool Tools Function Transparency
- When Auto-Tool is on, Hatz chooses the tools for you — switch Auto-Tool off in the Tools menu to choose them yourself. A notice at the top: "Turn off Auto-Tool to select tools" now appears in both the chat tools picker and the Workshop agent editor as a reminder.
Load More
→