Lega Release Notes - Mar 19, 2026

This release introduces agentic orchestration to the Lab — the ability for apps to reason about requests, delegate to specialists, and begin producing real documents. It also brings new model provider options including Google Vertex AI, deeper knowledge search controls, and a faster way to navigate your apps.

⭐ Spotlight Features - Agentic Orchestration

Task Apps & Agentic Orchestration Preview   

Task Apps are the building blocks of agentic orchestration. Each one does one specific job — you give it defined inputs (a contract, a client name, a date) and it returns structured results (a risk summary, a formatted memo, an extract). They accept file uploads, produce downloadable documents, and keep a full history of every run. Every Task App your firm builds can be called on automatically by a coordinator — so the more you build, the more your apps can accomplish together. Learn more here.

"/" Command Autocomplete

In any Chat App that has Task Apps attached, type "/" in the chat input to open a searchable menu of the available Task Apps. Start typing to filter by name, then click or press Enter to run one. This works with Task Apps only — it does not surface other resources such as tools or knowledge sources. As your library of Task Apps grows, this becomes a faster way to get to the right one without leaving the conversation. You can learn more via the document linked above.

Document Generation Preview  

Apps can now produce formatted files as direct outputs — PDF, Word (.docx), Excel (.xlsx), and CSV. Run a document-generating task from the "/" menu in any Chat App and the file lands in the conversation, ready to download.

Task Apps can also pass documents to one another in a chain — a task that generates a file can hand it directly to the next task as an input, so the output of one step feeds into the next.

You can learn all about it here.

🔍Also in this release

Advanced RAG Configuration

For apps that draw on uploaded document libraries, this release adds more control over how those documents are searched. Admins and Creators can now configure how finely documents are broken up for searching, how many results are pulled back per query, and how closely a match needs to be before it's used. Getting these settings right for a particular document collection can meaningfully improve the accuracy of AI responses — especially for large or densely formatted document libraries.

These settings are available in the Knowledge Source   editor under Processing Settings   and Retrieval Settings  .

In-Chat Knowledge Source Creation

You can now create a knowledge source on the fly, directly from the chat — no advance setup required. Click the + button at the bottom of any Chat App, select Create Ad-hoc Knowledge Source  , and upload or select the documents you want the AI to draw on. The AI will reference them in that conversation straight away.

Once created, the knowledge source is saved and available to reuse from My Resources → Knowledge  . Learn more about both of these features here.

Google Vertex AI Support

Google Vertex AI (Gemini) models can now be deployed and managed from the Lega admin console alongside your existing Azure OpenAI and other providers. All Vertex AI traffic flows through Lega Gateway with the same audit logging, compliance scanning, and governance controls that apply to all other providers. Organizations with Google Cloud agreements can now use their Vertex AI credits and Gemini models within Lega without any additional governance overhead.

🛠️Platform Improvements

• Chat loading experience. New skeleton loading states for chat messages, streaming markdown, and system messages provide clear visual feedback while content loads.
• French language support. Extended translations covering app type settings, chat and task labels, and tool limit messages.
• Model catalog updates. GPT-5.2 (OpenAI provider), GPT-5.4 (OpenAI provider), Gemini 3.1 Pro (Vertex AI Provider), Claude Opus 4.6 (Anthropic), Claude Sonnet 4.6 (Anthropic, Federal Register OpenAPI tool, and updated UK Companies House MCP URL added.
• Background job worker. New standalone worker process for long-running tasks including document parsing, embeddings, and data retention.

🐞Bug Fixes

• Fixed duplicate file uploads when both a file parameter and a default value file were present simultaneously
• Fixed user invite flow and status change issues in the users table
• Fixed model settings not persisting correctly after changes
• Fixed controlled/uncontrolled input warning in prompt parameter fields
• Fixed enterprise document processor initialization and null handling

❗Known Limitations