New
- Your AI agents can now have their own email addresses. Give any agent an address, share it with your team, and anyone in your organization with a Hatz account, and access to the agent can email it — just like they'd email a coworker. The agent reads the message (and any files you attach), uses its tools to actually look things up in your connected systems, and replies directly to the email thread within seconds. Reply back and it remembers the whole conversation. CC it on a group thread and everyone gets the answer. The power isn't "agents have email addresses." The power is: AI that meets people where they already are, does real work, and requires zero adoption effort.
- Admins can now create custom user roles with granular permissions. Set a role name, description, and choose exactly which permissions to include. Roles can be assigned during user invites or updated later, and existing custom permission groups can be migrated in one click.
- Starting a new chat is now easier with Chat Templates. Use pre-built templates like Web Search, Image Generation, Build Agent, and Industry. Templates pre-fill prompts, configure the right tools, and select the right LLM - automatically kicking off a great chat!
- A new Atlassian integration has replaced the previous Confluence integration opening access to Jira, Confluence, Compass, and JSM in a single connection expanding available tools.
- Ten new LLMs added to the model selector. These models are US-based, lightning-fast, and built for efficiency — delivering powerful AI performance with low credit usage.
- GPT-5.4 Mini(OpenAI) — OpenAI's compact powerhouse that punches well above its weight for everyday tasks.
- GPT-5.4 Nano(OpenAI) — The tiniest OpenAI model with a surprisingly big brain, perfect for rapid-fire responses.
- Qwen3 VL 235B A22B(Qwen) — A massive vision-language model that sees and understands the world with stunning precision.
- Qwen3 Coder Next(Qwen) — Your new favorite coding companion, ready to write, debug, and ship code at speed.
- DeepSeek V3.2(DeepSeek) — A razor-sharp reasoning model that dives deep and surfaces answers fast.
- MiniMax M2.5(MiniMax) — A sleek, versatile model that keeps things smooth and efficient without breaking a sweat.
- GLM 5(Z AI) — Z AI's polished general-purpose model built for sharp, reliable, and snappy conversations.
- Kimi K2 Thinking(Moonshot AI) — A deep-thinking model that takes a breath, reasons carefully, and nails complex problems.
- Kimi K2.5(Moonshot AI) — Moonshot's fast and capable all-rounder that handles nearly anything you throw at it.
- Nemotron 3 Super 120B A12B(NVIDIA) — NVIDIA's supercharged 120B model optimized to run lean, mean, and incredibly smart.
Improved
- File size limits have increased to 50 MB per file and 20 files max per upload/chat
- Workflow limits have increased to support up to 40 user inputs and 10 constants.
- Improved Auto Tool Selection: Search tools, web browsing tools, and code execution tools are now added automatically, streamlining your workflow without manual configuration.
- The scheduled workflows view now shows the next run time for each trigger, making it easier to see when workflows will run next.
- Improvements to Code execution sessions for a more efficient process for inspecting and transforming files.
- File uploads are now faster and more reliable, with clearer status tracking throughout processing.
Fixed
- Fixed an issue where pasting from code blocks could use the wrong copy format.