Skip to main content

July 27, 2026

Highlights

A lot landed this week. Here is the short version:

  • Org admins: two redesigned hubs make integrations and models far easier to manage. The new Connectors Hub lays out your integrations as filterable cards with connection status and tool counts. The Models Hub groups models by vendor with clear provider routes, retirement warnings, and a count of the agents still using each model, so you can move them before access goes away.
  • Builders: turn a spreadsheet into a working app. Drop an Excel or CSV file into App Builder and it reads your columns, seeds a database with your rows, and proposes an app built around them. New apps can also pick up your company branding automatically.
  • Everyone: six new models to pick from. Claude Opus 5 joins the catalog, along with Kimi K2.6 and K3, GLM-5.2, Qwen 3.6 27B, and MiniMax M3 through our managed gateway. There are no keys to set up, and every one of them carries the same commitments as the rest of the platform: your data is not used for training, and the provider does not keep a copy.

There is more in the full notes below, including connector categories, smarter tool discovery in long conversations, and a long list of reliability fixes.

New features

A redesigned Connectors Hub (gradual rollout)

Managing integrations should not feel like hunting through a flat list. The new Connectors Hub presents connectors as visual cards with logos, descriptions, and how many tools each one exposes.

What you can do:

  • Filter by category to jump straight to CRM, communication, file storage, developer tools, finance, or other groupings
  • Filter by status to see what is already connected versus what is available to enable
  • Open a connector from its card to configure it or browse its tools

The hub is rolling out progressively behind organization capabilities. If you do not see it yet, your existing connectors page is unchanged until the rollout reaches you.

Connector categories for templates and custom connectors

Whether you are enabling a curated template or wiring up a custom MCP connector, you can now assign a category so it appears in the right place in the hub. Categories follow a consistent taxonomy (CRM & Sales, Communication, Files & Docs, Dev & Product, Finance & HR, and Other), making large connector libraries easier to scan.

A redesigned Models Hub

Choosing and retiring models should not feel like reading a spreadsheet.

The Models Hub now groups models by vendor with provider routes laid out in expandable rows, so you can see whether a model is served directly, through a gateway, or on a dedicated route. Creator filter chips, in-hub search, and clearer labels help you narrow the list quickly.

When models are retiring soon, a banner highlights the nearest retirement date. Retiring rows show how many agents depend on each model, with honest states when the list is long. You can jump into the agent editor on the model section to update defaults before access goes away.

Route tooltips explain provenance, naming the provider actually serving a route rather than the gateway vendor, and the image models hub follows the same retirement patterns for consistency.

Start from your data in App Builder (gradual rollout)

The fastest way to get a useful app is to start from the data you already keep in a spreadsheet.

On the App Builder landing page you can now drop in an Excel or CSV file and get a working app built around it. App Builder reads your columns, infers a schema, seeds a database with your rows, and proposes an app that fits what it found. You can add a sentence about what the app should help you do, or pick from suggestions tailored to your actual columns, such as tracking statuses and deadlines or a dashboard of totals. No file handy? There is sample data to try it with.

Your original file stays untouched: App Builder works on a private copy.

This is rolling out progressively, so the entry point may not be on your landing page yet.

Apps that arrive on brand (gradual rollout)

When you create a new app without choosing a design style, and your organization has no brand styles configured, App Builder can now infer a brand brief from your work email domain, including name, colors, and logo, and use it as the app's design style. The result is an app that looks like it belongs to your company from the first build. Personal mailboxes are ignored, and anything you style explicitly always wins.

This is also rolling out progressively.

Six new models in the catalog

This release expands what you can select in chat and agents.

From Anthropic:

  • Claude Opus 5 for high-capability workloads where your organization has access

Through our managed gateway:

  • Kimi K2.6 and Kimi K3 from Moonshot AI
  • GLM-5.2 from Z.ai
  • Qwen 3.6 27B from Alibaba
  • MiniMax M3 from MiniMax AI

You can start using these right away: there is no account to connect and no keys to manage. They carry the same data commitments as every other model on the platform: your data is not used for training, and the provider does not keep a copy.

Availability still follows your organization's model allowlist and provider configuration.

Smarter tool discovery in long conversations

When you have a long back-and-forth with the AI, it needs to find and remember the right tools as your request evolves. This release improves how tool discovery works under the hood:

  • Multi-keyword matching helps the AI surface the most relevant tools when your question spans several topics
  • State persistence means the AI keeps track of tools it already loaded, even when a conversation continues across multiple internal steps

The result: fewer moments where the AI reaches for the wrong capability or seems to forget what it was working with.

Improvements

  • Chat holds onto context better. Project chats keep their scope on follow-up turns, and output limits now follow each model's real catalog limits.
  • Signing in takes fewer detours. OIDC and MFA journeys are more reliable, including when you arrive from a satellite app or need to pick an organization.
  • A tidier sidebar. Deleted chats and their routine runs disappear right away and stay gone after a refresh.
  • App Builder feels more responsive. Sub-chats stay live with smoother transitions, you can keep typing while a turn runs, and cost estimates reflect the context actually being sent.
  • Knowledge and file handling are more forgiving. Retrieval keeps working for agents whose knowledge was never tagged, empty CSV uploads warn instead of failing, and file search handles more edge cases.
  • Connectors are steadier. Sign-in recovery and tool execution are more dependable, and the hub layout is calmer to scan.
  • Clearer billing dates with managed-by labels, so admins can tell AppDirect-managed from tenant-managed timing at a glance.

Bug fixes

Reliability fixes across the Models Hub, the Connectors Hub, sign-in, chat, and App Builder, plus billing display and platform updates under the hood. Together they should make model management, integrations, and everyday chat feel more predictable.

⌘J