> Source: https://builder.ema.ai/v2/release-notes
> Title: Release Notes

# Release Notes

_July 2026 — the next generation of the Ema platform_

This release is the most significant evolution of the Ema platform to date. Over the past several months we rebuilt the builder experience from the ground up, shaped by the work your teams have been doing and the innovations we built in response. This is not an incremental update — it is a generational leap in how you build, test, deploy, and scale AI Employees.

Below are the headline changes, followed by the full list of what shipped.

> [TIP]
> **New to the platform?** Start with the [Introduction](/builder/v2/introduction) and [Getting Started](/builder/v2/getting-started). This page is written for builders and partner teams who already know Ema and want to know what changed.

## Highlights

Capability

What it means for you

**Ema Autopilot**

Autopilot is now a hands-on building partner, not just an assistant. It builds and configures AI Employees, sets up connectors, explains what your AI Employees do, debugs runs and self-corrects, and generates artifacts on demand — with vastly improved performance.

**Ema MCP**

Invoke Ema from your preferred agentic development harness (Claude Code, Codex, or your own) through Ema's MCP layer. Ema can also call your custom MCP servers, so your AI Employees can use your other agents and every pre-built MCP.

**Improved Agent Library**

Many previous agents are consolidated into a shorter list of more powerful, more autonomous agents — stronger at tool-calling, extraction, and coding, in single resilient blocks that previously needed more complex workflows.

**Redesigned AI Employee Builder**

A cleaner, faster canvas with better error handling, version control with data, auto-layout, and card views. Building is more intuitive and less error-prone.

**Partner management**

Create and manage tenants directly from the UI, set up SSO, customize welcome-email branding, whitelist domains, configure custom LLM API keys, and get usage visibility across your teams — with no dependency on Ema.

## Ema Autopilot — your hands-on building partner

Autopilot has moved from an assistant that answers questions to a partner that does the work alongside you.

-   **Builds and configures AI Employees.** Describe what you want in plain language and Autopilot creates and edits workflows, configures agents, sets up knowledge bases, and connects Tools.
-   **Sets up connectors and explains your AI Employees.** Ask Autopilot to wire up an integration, or to explain what an existing AI Employee does and why.
-   **Debugs and self-corrects.** Autopilot can inspect an AI Employee's run, find where it went wrong, and correct its own work based on the execution details — closing the loop between building, testing, and fixing.
-   **Generates artifacts on the go.** Evaluation reports, test data, documentation, and graphics that help you understand and manage what you've built.
-   **Vastly improved performance** across all of these tasks.

Working with Autopilot is also smoother:

-   **Narrative and Developer views** — switch between a readable summary of what Autopilot did and a step-by-step view with full reasoning and tool outputs.
-   **Thinking control** — dial how much Autopilot reasons before it acts.
-   **Stop and steer mid-run** — cancel a turn in progress at any time.
-   **Session continuity** — your conversation auto-resumes and survives a page refresh; unsent drafts are preserved.
-   **Export conversations** — save a conversation as a readable file or a structured record.
-   **Clear approvals** — Autopilot pauses and asks before anything destructive or hard to reverse.

> [INFO]
> See the [Autopilot](/builder/v2/autopilot) section for the full guide, including modes, permissions, and tips.

## Ema MCP — connect Ema to your agents, and your agents to Ema

Ema now speaks the Model Context Protocol (MCP) in both directions.

**Invoke Ema from your own agentic harness.** Connect Claude Code, Codex, Claude Desktop, or any MCP-compatible client to Ema over a secure OAuth connection and drive the platform through conversation — list, create, and edit AI Employees; inspect and edit workflows; manage knowledge bases; connect Tools; and test or debug conversations. The connection acts as the signed-in user and respects your role and tenant boundaries.

**Let Ema call your MCP servers.** Builders can register custom MCP servers from the Integrations page, so your AI Employees can use your other agents and every pre-built MCP alongside Ema's native Tools.

> [INFO]
> See [Ema MCP](/builder/v2/ema-mcp) to connect a client and review capabilities, and [the Integrations hub](/builder/v2/integrations-data/integrations-hub) to register your own MCP servers.

## Improved Agent Library — fewer agents, more power

We consolidated a long catalog of narrow, single-purpose agents into a shorter list of more capable, more autonomous ones. Each of the new agents does more on its own:

-   **Stronger tool-calling** — agents plan and call the right Tools to complete multi-step work in a single block.
-   **More capable extraction** — pull structured data from documents and messages, with validation.
-   **Built-in coding** — run custom logic as a first-class step, with typed inputs and outputs.
-   **Autonomous search-and-respond** — retrieve, reason over, and answer from your knowledge in one agent.

The result: work that used to require several chained agents now fits into one resilient, higher-performing block — simpler to build and easier to maintain.

> [INFO]
> See the [Agent Reference](/builder/v2/agent-reference) for every agent in the builder library.

## Redesigned AI Employee Builder

The builder canvas has been rebuilt to make building faster and less error-prone.

-   **Cleaner, faster canvas** with auto-layout and card views so large workflows stay readable.
-   **Better error handling** — problems are surfaced clearly, where they happen, before you publish.
-   **Version control with data** — published versions are captured as a complete, reproducible snapshot, so you can compare versions, revert safely, and always know exactly what's live.
-   **Reusable, deterministic steps** for reshaping data between agents without spending an extra model call.
-   **Human-in-the-loop, built in** — pause any agent to collect a review, an approval, or structured input through inline forms, external forms, or a conversation.
-   **Knowledge bases as a first-class step** — create a knowledge base, add documents by upload or from a connected source, and attach it to an agent, all from the builder.

> [INFO]
> See the [Builder Guides](/builder/v2/builder-guides) for task-oriented walkthroughs, and [Core Concepts](/builder/v2/core-concepts) for how the pieces fit together.

## Test with confidence before you launch

Higher quality reaches production because you can prove behavior before shipping.

-   **Evaluation workflows** — build datasets, define reusable rubrics, and score your AI Employees at scale with aggregate results and side-by-side run comparison.
-   **Test mode** — run an AI Employee end to end with external calls mocked, so you can test without touching production systems or metrics.
-   **Unattended evaluation of human-in-the-loop steps** — evaluations can run across hundreds of rows without a person sitting in the loop, then hand you the transcript for scoring.

> [INFO]
> See [Evaluation](/builder/v2/testing-operations/evaluation) and [Testing AI Employees](/builder/v2/testing-operations/testing-ai-employees).

## Knowledge and integrations, unified

-   **Integrations hub** — one place to manage everything Ema connects to: Tools, data sources, channels, MCP servers, and an audit trail.
-   **Visual Tool Editor** — author your own Tools (HTTP or TypeScript) with input/output schemas and live testing, no separate build step required.
-   **New data connectors** — connect Box, Confluence, Google Drive, and SharePoint with guided, per-connector setup.
-   **Reusable connections** — define a connection once and reuse it across AI Employees, at either the tenant or the individual-user level.

> [INFO]
> See [Integrations & Data](/builder/v2/integrations-data), including [the Tool Editor](/builder/v2/integrations-data/the-tool-editor) and [data connectors](/builder/v2/integrations-data/data-connectors).

## Smarter model routing

EmaFusion™ — the model layer behind every AI Employee — is now more capable:

-   **Combines the best models** — for each request, EmaFusion draws on multiple leading models across providers and combines their outputs to return a stronger, more reliable answer than any single model could give on its own.
-   **Confidence-based selection** — it weighs how confident each model is and returns the highest-confidence result, falling back automatically when the top choice is uncertain.
-   **Capability-aware routing** — requests that need tool-calling or structured output are only sent to models that support them.
-   **Per-call token visibility** — every model call's usage is tracked and attributed, so cost is transparent.

> [INFO]
> See [EmaFusion](/builder/v2/emafusion) for how it works.

## Partner management

Running Ema across multiple teams and tenants is now self-serve.

-   **Create and manage tenants** directly from the UI — no dependency on Ema.
-   **Set up SSO** so your teams sign in with your identity provider.
-   **Customize welcome-email branding** for a consistent experience.
-   **Whitelist domains** to control who can join.
-   **Configure custom LLM API keys** to use your own model provider accounts.
-   **Usage visibility** across your teams.

> [INFO]
> See [Administration](/builder/v2/administration) for tenant, SSO, and user-management guides.

## Your existing work is safe

All of your AI Employees, execution history, audit logs, and configurations remain fully available in your current environment. Nothing is deleted or moved without you. The next-generation platform runs in a dedicated, newly hardened environment; during the transition your teams keep access to both, so production work is never disrupted while you explore and adopt the new experience.

> [TIP]
> Questions about the rollout or your migration? Your Ema AI Coach can walk your team through what's new and help you plan adoption.