Phrase Atlas

Available for

Team, Professional, Business, Enterprise and Software UI/UX plans

Get in touch with Sales for licensing questions.

Note

Phrase Atlas is available as part of the Early Access Program (EAP).

Phrase Atlas is a conversational AI agent for the Phrase Language Intelligence Platform, which provides direct access to Phrase platform capabilities through natural language. Instead of navigating menus or workflows, describe what you need, and Atlas executes it.

How Phrase Atlas Works

Atlas operates through a chat interface. Each chat is a separate thread with its own history. Atlas reads each message, decides whether the request is a question or an action, and responds in kind.

Conceptual questions, such as "What is a translation memory?": Atlas provides an explanation.
How-to questions, such as "How do I create a job?": Atlas provides step-by-step guidance.
Action requests, such as "Create a project" or "Update this job's due date": Atlas confirms the details, then executes the action.

Atlas can combine these capabilities in ways not exposed in the user interface, enabling tasks not otherwise possible without direct API use (e.g., bulk updates across multiple projects in a single request).

Access

Phrase Atlas is available to

TMS Administrators and Project Managers
Strings Owners, Administrators and Project Managers
Platform Owners, Administrators and those who have access to Atlas in TMS or Strings

Access applies at the organization level.

Atlas respects existing role and project permissions: available actions depend on the user's permissions.

Requesting Access During EAP

A Platform Owner or Administrator must request access from within the product on behalf of their organization.

Once access is granted, Atlas is immediately available to all Project Managers, Administrators, and Owners.

Using Atlas

Atlas is accessible from the main navigation or via the Ask Atlas button in the header, in the Platform Dashboard, Phrase TMS, and Phrase Strings.

Atlas operates only within the scope of Phrase products.

How Chat Works

To ask a question, follow these steps:

Open the Phrase Atlas chat via the Ask Atlas button or from the main navigation.
Describe your goal in plain language.
Send the message.
Atlas proposes a plan before making any changes.
Confirm to proceed, or refine your request.

Start a new chat using the New chat button at any time.

Reading Data vs. Making Changes

Atlas executes read actions (retrieving projects, jobs, translation memories, and similar data) immediately. Atlas asks one clarifying question at a time when a request is unclear or could apply to more than one product or resource. For example, if a user asks to set up a "term base" while connected to both Phrase TMS and Phrase Strings, Atlas asks which product the request applies to before proceeding, because both products use this term for related but distinct features.

If a user repeats a request without answering a clarifying question, Atlas states the assumed default and proceeds.

Atlas gates every write action (creating, updating, or deleting a project, job, translation memory, term base, or other resource) behind explicit confirmation. Atlas presents a plan and waits for explicit confirmation before executing. A short reply such as "yes, do it" confirms the plan and triggers execution.

Prerequisites: Some write actions require a specific account role or organization-level permission. For example, creating or updating a translation memory, term base, project, or user in Phrase TMS requires the Admin or Project Manager role with the relevant user permissions. The TMS Linguist, Guest, and Submitter roles cannot perform these actions.

Handling Credentials

Atlas never asks for a third-party secret or credential in the chat, including an API key, token, client secret, password, or service-account key file. For an action that needs such a credential, Atlas configures the non-secret parts of the resource through the API, then directs the reader to enter the credential in the named Phrase settings page, and offers to verify the connection afterward.

Chat History

Chats are saved and can be revisited at any time. Continuing a saved chat picks up from where it left off, including prior context.

Rename a chat by clicking Rename in the More menu.

Click the name of the current chat to display previous chats in a dropdown.

Chats can be deleted from the history by clicking the recycle bin icon after clicking the chat name or from the More menu.

Interrupting a Chat

Chats can be stopped before they finish.

Working with Files

Atlas reads files attached to a chat message and uses their content in a request. Attach a file using the paperclip icon, then reference it in the message (for example, "Count the segments in this XLIFF file").

Atlas reads common structured formats, including .CSV, .TSV, .JSON, .XML, .XLIFF, .PDF, and .XLSX.

The file upload limit is 50 MB per file.

Atlas also produces downloadable files from data already in the chat. To export data, ask Atlas to turn a result or table into a file. Atlas generates a downloadable file in .CSV, .XLSX, or .PDF format.

Providing Feedback

Atlas is an AI system and can occasionally produce inaccurate results. Atlas always requests confirmation before making changes. Review the proposed plan before proceeding.

Feedback can be submitted directly through the chat interface:

Thumbs up: submit positive reinforcement when Atlas performs as expected.
Thumbs down: report unexpected behavior with a brief description. All feedback is reviewed by the Phrase team.

Limits

Model: Anthropic Sonnet 5.

Usage Limits During EAP

Atlas usage is metered per user.

Chat allowance: each user receives a set number of chats for the duration of the EAP. This varies by plan.

Plan	Chats per user
Enterprise, Business	40
Starter, Team, Professional, Software UI/UX	20

Chat size cap: each chat is capped at approximately $5 in LLM costs.

Running out of chats: users can top up using existing AIUs/credits from their organization's pool by contacting Phrase.

Best Practices for Token-Efficient Chats

Since each chat has a cost cap, the structure of the chat affects how much of that allowance is used.

Start a new chat for each unrelated task. Atlas carries the full history of a chat into every response. Combining unrelated tasks (e.g., term base work, then a Strings task, then a user management request) in a single chat means each new task pays the cost of the previous ones' history, even though it doesn't need it.
Keep chats focused on one goal. A chat that stays on a single task uses meaningfully less of your allowance than one that drifts across multiple topics.
Avoid pauses mid-task where possible. Resuming a chat after a five-minute gap can cost more than continuing one that is still active, since some of the earlier context needs to be reloaded.

Example Use Cases

Area	Description	Prompt
User management	Bulk deactivating a filtered group of users	Deactivate all linguists whose name starts with 'JHA.' Show me a table with the first name, last name, masked username, masked email, and activated flag.
Analytics and usage reporting	A data query against project/translation-management records for a specific time frame	Give me the top 10 projects from July with the highest word count. For each, list the source and target languages, status, and project owner.
How-to guidance	Regional localization workflow setup	I want to tailor our marketing content, which lives in WordPress, by region, and need different translation versions for Spain and Latin America.
Quality benchmarking	Engine comparison against a quality benchmark	Retranslate the Atlas Demo Project with the AI Translation Agent, instead of the Machine Translation engine, and use the Atlas Demo Quality Profile to compare the results.