Changelog

Clay connector

New

Teams enriching go-to-market data in Clay previously needed a separate integration or a manual export to bring that enrichment into an agent workflow. The Clay connector adds 11 tools across searches, routines, and table queries, so agents can enrich and query GTM data directly.

Unify connector

New

Go-to-market teams running signals and intent data through Unify had no direct way to let an agent act on records, objects, or attributes there. The Unify connector lets agents manage records, objects, and attributes, and run event queries against Unify data.

Google Analytics connector

New

Pulling reporting data out of Google Analytics for an agent to reason over meant hand-building calls against several separate APIs, since accounts, properties, audiences, and reporting each work differently. The Google Analytics connector covers accounts, properties, audiences, custom dimensions and metrics, data streams, key events, and reporting in one connector, so agents can query all of it through a single interface.

ServiceNow query building expansion

Improved

Encoded queries are one of the harder parts of working with ServiceNow's API. The syntax is compact and easy to get wrong. The ServiceNow connector now includes query building and validation tools, so agents can construct and explain encoded queries more reliably instead of guessing at syntax.

Connector reliability and scheduling improvements

Improved

A handful of connectors got targeted fixes this week. Calendly event booking now supports location, so invitees can be scheduled for event types that require a physical or virtual location. Google Drive adds resumable uploads for more reliable large file transfers and fixes downloads and exports that were returning 404 errors on files in shared drives. Podium contact and location lookups, campaign limits, and phone number validation are also more reliable.

Freshservice tickets and comments sync in real time

Ticketing [[hide]]

New

Ticketing integrations that only support scheduled syncs mean a new comment or ticket update can sit for the length of the sync interval before it reaches your unified data. GET /tickets and GET /comments for Freshservice now support manual webhooks, so ticket creations, updates, and deletions, along with new comments, sync in real time instead of waiting for the next scheduled run.

HRIS data quality improvements across the board

HRIS [[hide]]

Improved

Employee records pulled from HRIS integrations have been prone to small but consequential errors: wrong country codes on phone numbers, contractor emails landing in the wrong field, incremental syncs missing changes that never touched the core employee record. GET /employees for SAP SuccessFactors, Gusto, Microsoft Entra ID, Lucca, and UKG Ready now handles these cases correctly, from corrected country-code mapping on mobile numbers to contractor emails mapping to personal_email instead of work_email, to Entra ID group membership changes updating an employee's modified_at value so they're picked up on the next incremental sync. Rehired SAP SuccessFactors employees no longer carry outdated termination dates, and Dayforce, Paycor, and UKG Pro Workforce Management got sync performance and reliability improvements for large employee populations.

Sync performance and reliability across ATS, CRM, and File Storage

ATS, CRM, and File Storage [[hide]]

Improved

A handful of maintenance fixes landed across products this week. Candidate attachment downloads from Greenhouse handle edge cases more reliably, and candidate pulls from Lever no longer stall mid-sync. Zoho CRM's GET /users now paginates correctly so initial syncs complete for larger accounts, and Microsoft SharePoint got improved Live Search validation, region handling, and faster initial syncs for accounts with group membership data.

Confluence article sync sped up

Knowledge Base [[hide]]

Improved

Large Confluence knowledge bases could take a while to fully sync, especially on the first run. GET /articles for Confluence now syncs faster and uses fewer resources on both initial and incremental syncs, so bigger knowledge bases catch up sooner.

Account linking and dashboard reliability tightened

HRIS [[hide]]

Fixed

Broken account links tend to surface downstream as failed syncs instead of a clear setup error, which makes them hard to debug. Account linking for PeopleHR no longer throws validation errors that could block an initial connection, Lucca now accepts additional account URL formats, and Merge Link flags invalid Integration System IDs during Workday setup before they cause a sync failure. On the dashboard side, integration-wide field overrides on Workday SFTP accounts no longer produce null values, and Workday Selective Sync boolean filters now send lowercase true or false so filtered requests stop failing.

Unified docs onboarding revamped

Improved

Getting started with the Unified API meant piecing together core concepts from scattered pages before writing a first integration. The onboarding flow has been revamped to explain those concepts up front, so new users can get from signup to a working integration faster.

HubSpot opportunity webhooks go real time

CRM [[hide]]

Improved

Tracking a deal through HubSpot used to mean waiting on the next scheduled sync to notice a stage change or a new close date. Opportunity webhooks for HubSpot now fire in real time on GET /opportunities, so updates to the close date, stage, or associated account push through as soon as they happen.

SharePoint group and file sync go deeper

File Storage [[hide]]

New

Selective sync in SharePoint meant group membership changes made after the initial sync could go uncaptured, and custom, dynamic, and site groups did not always sync completely. GET /groups for SharePoint now captures membership updates after the initial sync and handles custom, dynamic, and site groups more completely. Initial sync speed for large drives and reliability for high-volume file changes both improved on GET /files, and sensitivity-label data now syncs more consistently across files and permissions.

HRIS sync reliability and data cleanup

HRIS [[hide]]

Improved

Employee and time-off syncs on UKG Pro Workforce Management, Dayforce, and Paycor could stall or lose progress on large employee populations or interrupted runs. Sync efficiency and interrupted-run recovery improved on GET /employees, GET /time-off, and GET /bank-info across those three systems. Elsewhere in HRIS, SAP SuccessFactors work emails no longer produce duplicate records, Humaans departments and teams with matching names stay distinct, and Darwinbox consultant and temporary employment types now map to standard values.

Attachments and comments get more flexible

Accounting, Ticketing [[hide]]

New

Writing content back into Xero and Freshservice used to run into format constraints. POST /attachments for Xero now accepts Base64-encoded file content directly, so an attachment no longer needs a hosted file URL first. POST /comments for Freshservice now supports both public replies and private notes, attributed to a specific user or contact author.

TikTok connector

New

Agents building social and creator tools have had no native way to read or write TikTok content without managing TikTok's OAuth flow and API pagination directly. The TikTok connector handles that infrastructure: agents can now manage user profiles, videos, and photo posts through TikTok's Display and Content Posting APIs.

Podium connector

New

Teams running customer messaging through Podium previously had to build direct API integrations to let agents read conversations or trigger campaigns. The Podium connector adds OAuth 2.0 support across contacts, messaging, conversations, users, webhooks, campaigns, and locations, all through a single normalized endpoint.

Oracle HCM expansion

Improved

Agents processing time-off requests need both absence balances and payslip data from the same system, and fetching them separately means two round-trips and two chances for sync lag. The Oracle HCM connector now adds payslip and absence-balance tools, with absence creation that reliably submits time-off requests.

Request logging hardened

Security

Connector request logs were capturing tokens, headers, and response bodies, meaning credentials and customer data could end up in log storage without any redaction. Logging across all connectors now strips that data at write time, so tokens, headers, and response bodies never reach the log.

Three new models added

New

Evaluating a new model used to mean rebuilding a request format from scratch and rechecking a provider's data retention policy each time. Gateway now supports GPT-5.6's Sol, Terra, and Luna models from OpenAI, Grok 4.5 from xAI, and Muse Spark 1.1 from Meta, all through the same request format already in use. Sol, Terra, Luna, and Grok 4.5 are available with zero data retention.

you.com web search tool

New

Agents that need current information have had to rely on a model's training data or wire up a separate search API outside Gateway. Gateway now supports you.com's web search tool directly, backed by their premium web search API, so a request can pull in live search results without leaving Gateway's request format. The integration runs with zero data retention.

Professional Plan pricing

New

Testing Gateway at low volume previously meant paying for every request from the first one, or staying on a free tier with limited headroom. The new Professional Plan includes $10 of credits every month at no charge, resetting automatically at the start of each billing cycle. That is enough to run real workloads before committing to volume pricing.

Custom complexity classifiers for Build Your Own Router

New

Build Your Own Router ships with Merge's default complexity classifier, but not every product routes prompts the same way. You can now train a custom complexity classifier tuned to your own use cases, so routing decisions reflect how your traffic actually breaks down instead of a general-purpose model. This is useful if your prompts cluster differently than the traffic the default classifier was tuned on.

Fable 5 by Anthropic

New

Fable 5 is back on Merge Gateway after being removed, and it is one of the premium frontier models available today. You can route to it through Build Your Own Router or call it directly like any other model in the catalog, with no separate setup required.

Slackbot connector

New

Connecting Slack as a human user meant agents ran under that person's account, inheriting their channels, their DMs, and their access history. The Slackbot connector gives agents their own identity: a dedicated bot with 48 tools across messaging, channels, and files. Activity in Slack shows up under the bot, not whoever set up the connection.

Generic record tools for Salesforce

New

Most Salesforce orgs have custom objects that do not fit any standard schema, and until now agents could only read and write the objects Merge had already modeled. Generic record tools give Salesforce agents full CRUD on any object, standard or custom, without waiting on Merge to add support for it. A query-injection guard ships alongside the new tools, so agents building dynamic queries cannot be tricked into reaching data outside the intended object.

Ramp purchase order support

New

Ramp agents could read purchase orders but not act on them, so any change still routed through a person in Ramp's UI. Ramp now supports full purchase order management: agents can create, update, archive, and manage line items on POs directly. That covers the full PO lifecycle end to end, not just the pieces that were easy to add.

Access Key scope updates

Security

Access Keys previously locked in their scope at creation, so narrowing or widening what a key could reach meant issuing a new one and rotating it through every service that used it. You can now update an Access Key's scopes directly in the dashboard, adjusting its access without generating a new key or breaking anything already using the old one. That matters after a security review, or when a key needs broader access as a new integration comes online.

Expense reports endpoint added to Accounting

Accounting [[hide]]

New

Pulling expense report data into a workflow previously meant reconciling it by hand outside the API, since Unified Accounting had no object for it. Accounting now supports GET /expense-reports, so you can fetch expense report records the same way you already fetch invoices or bills. The new endpoint accepts an employee_id filter, and GET /employees now supports filtering by email_address too, so you can look up the right employee or expense report without pulling the full list.

iTrent connector

HRIS [[hide]]

New

Companies running iTrent for HR previously had no way to sync that data through Merge. iTrent is now available in the HRIS category, covering GET /employees, GET /employments, GET /groups, and GET /locations. Data comes through the same normalized schema you already use for every other HRIS integration on Merge, so switching HR systems does not mean rewriting your integration.

OmniHR connector

HRIS [[hide]]

New

OmniHR joins the HRIS category this week, covering GET /companies, GET /employees, GET /employments, GET /locations, and GET /groups. Teams running OmniHR can sync that data through Merge's standard HRIS schema instead of building a custom integration against OmniHR's native API.

Smarter routing with use-case complexity classifiers

Improved

Intelligent Model Routing used to score every request against one general complexity signal, so a coding task and a creative-writing task could land on the same routing decision even though they need different things from a model. Routing now uses separate complexity classifiers for coding, creativity, reasoning, and knowledge tasks, so the routing decision accounts for what kind of task the request actually is.

Pi.dev and Factory Droid support

New

Coding agents built on Pi.dev or Factory's Droid couldn't route through Merge Gateway without custom setup. Both are now supported out of the box, so you can point either one at Gateway the same way you would any supported IDE integration.

Better provider cache control

Improved

Prompt caching only pays off if requests actually hit the cache, and that depends on how well Gateway forwards a provider's own caching parameters. Provider caching now has better cache control support for providers like Anthropic, so requests that rely on prompt caching are more likely to hit it instead of reprocessing the full prompt each time.

Plaud connector

New

Agents that need data from Plaud previously had no direct path into it without a custom integration. The Plaud connector is now live in Agent Handler, giving agents a normalized way to read and act on Plaud data through the same tool interface as every other connector.

Sprout Social connector

New

Teams running social media management through Sprout Social couldn't give agents direct access to that data without building a one-off integration. The Sprout Social connector is now live, exposing Sprout Social's data and actions through Merge's normalized tool interface.

Unily connector

New

Organizations running their intranet on Unily had no way to connect that content to agents without custom work. The Unily connector is now available, giving agents normalized access to Unily data through Agent Handler.

Redesigned Access Keys experience

New

A single production key used to control all API access, so scoping or rotating access for one agent or environment meant touching everything at once. The redesigned Access Keys experience replaces that single key with multiple scoped keys, each managed from its own card, so you can create a key scoped to a specific agent, environment, or use case and rotate or revoke it without affecting the rest.