Changelog

Ticketing

Ticket writes on Jira accept the priority and status names configured in each customer's own Jira site, matched regardless of capitalization, so tickets created through Merge carry the priorities their teams already use. Read more in the Tickets docs.

‍

Gateway routes to your self-hosted and fine-tuned models

Add any OpenAI-compatible endpoint, such as vLLM, TGI, SGLang, or Baseten, and route to its models alongside the catalog.

• Add an endpoint under Configure > Self-hosted models with its URL, API key, and per-model settings
• Self-hosted models join Priority routing policies from the Self-hosted group in the model picker
• Traffic to your endpoint carries no Merge charges, and Gateway encrypts the API key and never shows it again
• Organizations with zero data retention can use them, and failover to curated vendors stays ZDR-compliant

Read more in the Gateway self-hosted models docs.

Route to Claude Haiku 5.5, and to Claude Sonnet 5.5 through deepinfra

Claude Haiku 5.5 by Anthropic is in the catalog, and deepinfra is a new model vendor serving Claude Sonnet 5.5 with zero data retention. Read more in the Gateway model catalog docs.

Priority routing policies filter and rank models by measured throughput and latency

A Performance section in the policy editor and the API sets thresholds on each Priority policy.

• A minimum throughput and a maximum time to first token
• Model ordering by measured latency or throughput
• Per-request min_throughput and max_latency overrides
• The routing reason explains how each threshold applied

Read more in the Gateway product page (docs link TBD - link to product page for now).

X-Merge-Customer-Origin-Id names the customer on each request with your own ID

Send your own customer ID in the header, and Gateway creates the customer on first use. You can also create or update customers in batches, mint customer API keys by origin ID, and read per-request customer usage records. Read more in the Gateway customer API keys docs.

You.com joins the catalog, and new tools for Greenhouse, QuickBooks, Linear and more

You.com is live, and three existing connectors gain new tools

• Greenhouse: 13 tools to hire applications, convert prospects, manage attachments, add employment and education history, and anonymize candidates, with full Harvest v3 write bodies
• QuickBooks Online: 34 tools across classes, departments, tax codes, expenses, bill payments, payment methods, and employees, plus a report tool covering 34 financial reports
• Linear: 11 tools for document management and cleanup, with priorities accepted by name

Read more in the Agent Handler connectors docs.

Apollo and HiBob accept OAuth 2.0 alongside their existing credentials

Apollo adds OAuth 2.0 next to its API key option, and HiBob adds it next to Service User authentication, ahead of HiBob retiring basic auth for partner apps. Read more in the Agent Handler connectors docs.

Route to Claude Sonnet 5.5, gpt-6.1-sol, and four more models

Three of the six run with zero data retention.

• Claude Sonnet 5.5 by Anthropic
• Claude Opus 5.5 through Amazon Bedrock, with zero data retention
• gpt-6.1-sol and gpt-live-1 by OpenAI, both with zero data retention
• mimo-v2.6-flash and mimo-v2.6-pro by Xiaomi Mimo

Read more in the Gateway model catalog docs.

OpenAI's ultrafast service tier is available on every Gateway API surface

Ultrafast is OpenAI's fastest processing tier and bills at six times the standard rate. Request it from any surface, alongside the flex and priority tiers. Read more in the Gateway service tiers docs.

previous_response_id chains on /v1/responses resume for 30 days

A stored turn stays available for 30 days, matching OpenAI's expiry, so a conversation picks up after a long pause. DELETE /v1/responses/{id} ends a chain early. Read more in the Gateway multi-turn conversations docs.

Gateway adds OpenAI audio translations, input token counting, and video management

Four more OpenAI endpoints pass through Gateway:

• Audio translations
• Responses input items
• Responses input token counting
• Video listing and deletion

Read more in the Gateway API reference docs.

Adobe Workfront, Tavily, WhatsApp Business, and nine more connectors join the catalog

Adobe Experience Manager (Cloud Service), Adobe Workfront, Alpha Vantage, Checkly, ConfigCat, Jamf Pro, Kit, Salesforce Data 360, Salesforce Marketing Cloud, Tavily, TinyPNG, and WhatsApp Business Cloud API are live. Jamf Pro responses include each device's OS and management state. Read more in the Agent Handler connectors docs.

Agent Handler adds Google sign-in and keeps SCIM roles in sync with your identity provider

Dashboard roles for SCIM-provisioned users are locked, so they always match the identity provider.

• A warning appears before a SCIM role mapping or role change would leave the organization without an Admin
• Identity provider checklists include Merge Admins and optional SSO setup steps
• The Agent Handler sign-in page offers single sign-on with Google

Change roles for SCIM-provisioned users in your identity provider. Read more in the Agent Handler SCIM provisioning docs

GET /remote-field-classes returns the selectable picklist and enum values for NetSuite custom fields and custom segments.

• Custom field types and formats normalize to one consistent set of values across Accounting
• POST /invoices and POST /expenses write relationship-type custom field values
• POST /invoices accepts both the previous and current custom field value formats

Read more in the NetSuite Accounting docs.

Batch inference runs through /v1/batches, with a Batches page in the dashboard

Submit chat completions, embeddings, responses, or Anthropic Messages requests as one provider batch, and results stream back as NDJSON. The Batches page sits next to Logs, filters by status and date, and opens a detail drawer with each batch's progress, cost, errors, and metadata. Read more in the batch inference docs.

Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and grok-4.7 are available with zero data retention

Seven models are new in the catalog, including two text-to-speech models billed by the second.

• Claude Opus 5.5 by Anthropic, GPT-6 Sol and GPT-6 Luna by OpenAI, and grok-4.7 by xAI, each with zero data retention
• GLM-5.3 FlashX by Z.AI
• Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS by Google, billed per second of generated audio and reported in the X-Merge-Billed-Audio-Seconds header

Read more in the Gateway model catalog docs.

Guardrails and data controls take per-customer and per-project overrides

A customer's Configuration tab sets its own prompt injection and data loss prevention policy, which inherits the organization policy until overridden and is enforced at request time. A new Controls tab on projects and customers overrides zero data retention, vendor restrictions, and region restrictions, and each unit resets to the organization setting on its own. Read more in the per-customer restrictions docs.

Routing policies set reasoning effort and accept an inline priority_order

Reasoning effort takes a policy-wide default, minimum, and maximum, plus per-model overrides, all enforced at dispatch and factored into vendor choice. Pass priority_order on /v1/responses or any compatible surface to route through an ordered model list, and Gateway saves it as a visible routing policy. Read more in the routing policy docs.

/v1/responses breaks down latency per request

Transfer, gateway, and provider time come back in the response body and the streaming done chunk. Non-streaming calls also carry a Server-Timing header. Read more in the Responses API docs.

Real-time voice sessions run over WebSocket at /v1/live/sessions

Each session is backed by gpt-live-1 by OpenAI. Sessions run with zero data retention. Read more in the live voice docs.

Agent Handler adds Power BI, Dealops, and GCC High versions of Outlook and Excel

Four connectors cover analytics, quoting, and the US Government sovereign cloud.

• Microsoft Power BI: 42 tools across workspaces, reports, semantic models, dashboards, dataflows, gateways, and tenant administration
• Dealops: read the CPQ catalog, build and price quotes, and follow them through approval
• Outlook GCC High and Excel GCC High: for teams on the US Government sovereign cloud

Read more in the Agent Handler connectors docs.

Admins set connector tool access per user from the dashboard

Grant a user access to a connector's tools, edit that selection later, or revoke it, one user at a time. The Group access dialog adds search boxes and consolidates tool selection into a single shared list. Read more in the managing tool access docs.

Access keys take explicit scopes at creation

Pick scopes when you create an access key. Privileged keys carry a badge in the dashboard, with revoke actions that reflect their elevated access. Read more in the access keys docs.

Link's OAuth step and success screen become configurable per organization

An optional "Powered by Merge" mark can appear on the Link OAuth step. Organizations can also skip the success screen, so the flow closes as soon as a connector authorizes. Read more in the Link customization docs.

Accounting

The invoice creation schema describes each line item as an expanded object. If your integration reads line-item structure from this endpoint, confirm your parser handles the object form. Read more in the invoices meta docs.

An agent can create and hand off its own organization

mgw agent signup creates a sandboxed organization and returns a claim link. A human opens that link and claims the organization from the dashboard.

Guardrails verdicts surface in the API Tester and on the API

Prompt injection and DLP results attach to each request, in the dashboard and in the response body.

• The API Tester shows prompt injection and DLP results for each test request
• include_guardrails_metadata returns the same verdicts alongside a live response
• The flag covers /v1/responses and every SDK-compatible endpoint

Read more in the prompt injection protection docs.

Eight models arrive in the catalog with zero data retention

Sakana AI, OpenAI, and augure models land alongside GLM-5.3 Flash on three inference providers.

• Fugu Max and Fugu Ultra 2.0 from Sakana AI
• GPT Image 2.5 Flare and GPT Image 2.5 Sunburst from OpenAI for image generation
• ossington-5, rosedale-1, and tofino-3 from augure
• GLM-5.3 Flash through Fireworks AI, Modal, and Together AI

Read more in the model catalog docs.

POST /v1/images/edits edits images with reference images and masks

The endpoint takes a source image, optional reference images, and a mask. It runs on the token-billed image models.

The Performance tab reports prompt caching hit rate and savings

Cache hit rate, cached token counts, and estimated savings sit in a prompt caching section under Usage and spend.