Language Weaver
AIRun enterprise machine translation through the RWS Language Weaver Cloud API, with terminology dictionaries that hold your approved wording. Agents translate at volume without drifting off approved terms.
What This Integration Enables
Language Weaver is what you reach for when translation is already governed. The wording has been agreed by legal, brand, or a regulator, it is written down, and the machine's job is not to be creative. RWS built this connector's surface around that reality: account dictionaries that pin source-to-target term pairs, linguistic options such as formality and regional spelling that vary by language pair, quality estimation scores returned alongside the output, and usage reports broken down by language pair, user, label, group, or application for chargeback.
The translation path is asynchronous by design. An agent submits an array of independent strings, in plain text, HTML, XML, XLIFF, or TMX, gets back a request ID, polls until the status reads DONE, and then collects the content. That shape suits Orchestration as a Service well, because the flow is not blocked while a large job runs, and a job that turns out to be wrong can be stopped with Cancel Translation before it consumes capacity. Feedback runs the other direction: Submit Translation Feedback records a human correction and a quality rating against the account, so what a reviewer catches does not stay in the reviewer's head.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Inbound ticket triage across a multilingual support desk
Tickets land in Zendesk from customers who write in whatever language they think in. The agent calls Identify Language on the ticket body, which returns candidate languages ranked by confidence along with the writing scripts detected and their proportions. That last detail matters for mixed-script content that simple detection gets wrong. The agent routes the ticket to the queue that handles that language, or, where no native speaker is on shift, translates the body with the support dictionary attached and attaches both versions. The routing decision is made on a confidence score, not a guess, and low-confidence tickets go to a human triager instead of a language queue.
Release notes into an internal knowledge base
Every release, the product team publishes notes that name features, plan tiers, and settings by their exact product names. Those names have approved renderings in each market and they are not negotiable. The agent submits the notes with the product-terms dictionary applied, polls until the job completes, collects the translated content with its quality estimation scores, and writes each language version into Notion as a page in the localized space. Strings whose quality estimation falls below the bar are written as drafts and flagged rather than published.
Consumption reporting and entitlement checks
Language Weaver bills against subscriptions with expiry dates and specific language pairs. On the first of the month, the agent calls Get Usage Report for the prior period, broken down by language pair and by group, and posts the totals to a finance channel in Slack for chargeback. In the same run it calls List Subscriptions and flags any entitlement expiring inside the next sixty days, and List Language Pairs to confirm that every pair the live flows depend on is still subscribed. A translation flow that fails at 2am because a subscription lapsed is an avoidable failure.
Human-in-Loop Highlight
The consequential write on this connector is not a translation. It is Add Dictionary Term. A dictionary term changes how a word renders in every subsequent translation for that language direction, on every flow on the account, forever, and Delete Dictionary removes a dictionary and every term inside it permanently. So when a reviewer submits a correction and the agent notices the same term has now been corrected the same way several times, it does not promote that correction into the dictionary on its own. It records the correction with Submit Translation Feedback, then goes to the terminology owner with the evidence: "The German rendering of Business plan has been corrected to the same alternative five times this month across three flows. The current dictionary entry says otherwise. Change the dictionary entry, keep it and treat these as one-off edits, or escalate to brand?" The owner decides, because the person accountable for approved wording is a person, not a frequency count. The agent gathers the evidence and holds the pen out. That is the difference between an automation that drifts and one you can defend in an audit.
Agent Capabilities
18 actionsTranslation
5- Translate Text Submits text for machine translation and returns a request ID. Accepts an array of independent strings in plain text, HTML, XML, XLIFF, or TMX. Source language can be set to auto for detection. Supports account dictionaries for terminology control, linguistic options such as formality, a speed-versus-quality mode, and optional quality estimation scores.
- Get Translation Status Reports the progress of an asynchronous translation request as INIT, TRANSLATING, DONE, or FAILED. Polled while the status is INIT or TRANSLATING.
- Get Translated Content Retrieves the finished output of a translation request. For text submissions the response contains the translated strings in the same order they were supplied, together with any quality estimation scores. Called once the status reports DONE.
- Cancel Translation Cancels a translation request that has not yet completed, releasing the reserved capacity. Used when a flow detects the wrong input was submitted before the job finishes.
- Submit Translation Feedback Records human feedback on a machine translation, optionally including a corrected version and a quality rating. Feedback accumulates in the account, so routing reviewer corrections here is how human judgment reaches the engine.
Language Identification
4- Identify Language Identifies the languages present in a text and returns them ranked by confidence, each with its three-letter code, English name, and IETF language tag, plus a breakdown of the writing scripts detected and their proportions. Synchronous, which suits routing decisions on short content.
- Start Language Identification Queues an asynchronous language identification job and returns a request ID. Used for large inputs that would exceed the synchronous call's limits.
- Get Language Identification Status Reports the progress of an asynchronous language identification job. Polled until the job finishes.
- Get Language Identification Result Retrieves the outcome of a completed asynchronous language identification job: the languages detected with confidence scores and the distribution of writing scripts in the content.
Dictionaries
5- List Dictionaries Retrieves the terminology dictionaries defined on the account, each with its ID, name, description, and the source and target languages it applies to. Dictionary IDs from this list are what get passed to a translation request.
- Create Dictionary Creates a terminology dictionary for a single source-to-target language direction. A dictionary starts empty and is populated before it affects anything.
- Delete Dictionary Permanently deletes a terminology dictionary and every term it contains. Translations that referenced it stop applying those terms, so this belongs behind a human check.
- List Dictionary Terms Retrieves the term pairs stored in a dictionary, each with its source text, enforced translation, optional comment, and case-sensitivity flag. Used to audit terminology or export a dictionary before editing it.
- Add Dictionary Term Adds a source-to-target term pair to a dictionary so that every subsequent translation using that dictionary renders the term exactly as specified. The combination of source text and case sensitivity must be unique within the dictionary.
Subscriptions and Reporting
4- List Language Pairs Retrieves every language pair the account is currently subscribed to, with its source and target language codes, model identifier, display name, and whether adaptation is supported. Called before translating to confirm the combination is actually available.
- Get Linguistic Options Retrieves the linguistic options a specific language pair supports, such as formality or regional spelling, along with each option's allowed values and system default. Values are case sensitive and vary by pair, so this is what populates them correctly.
- List Subscriptions Retrieves the account's subscriptions, including expiry dates and the language pairs each one grants. Used to monitor entitlement expiry before it breaks a live flow.
- Get Usage Report Produces a translation usage report for a date range, broken down by language pair, user, label, group, or application. Used for chargeback, capacity planning, and monitoring consumption against the subscription.
Frequently Asked Questions
What can FlowRunner do with Language Weaver?
FlowRunner agents can run Translate Text, Get Translation Status, and Get Translated Content in Language Weaver, plus 15 more actions.
Does connecting Language Weaver to FlowRunner require OAuth?
No. Language Weaver connects to FlowRunner with an API key, no OAuth flow required.
Can Language Weaver trigger a FlowRunner workflow automatically?
Language Weaver doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Language Weaver
$100 in credits. No card required. Connect in minutes.