From 6d799b034789afebbf575eeb03ae7467a6dfae76 Mon Sep 17 00:00:00 2001 From: Mahesh Vagicherla Date: Tue, 29 Sep 2026 09:54:36 +0530 Subject: [PATCH 1/2] chore: update backend changelog --- changelog/backend.mdx | 84 ++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 83 insertions(+), 1 deletion(-) diff --git a/changelog/backend.mdx b/changelog/backend.mdx index b6ab11c0a..96f66ec9e 100644 --- a/changelog/backend.mdx +++ b/changelog/backend.mdx @@ -1,6 +1,6 @@ --- title: "Backend" -sidebarTitle: "Backend [1.26.0]" +sidebarTitle: "Backend [1.28.0]" rss: true --- @@ -8,6 +8,88 @@ rss: true Discuss how Portkey's AI Gateway can enhance your organization's AI infrastructure + + +## v1.28.0 +--- + +### Logs & Exports +- **Agent search on logs**: The logs list API accepts an `agentSearch` query parameter for case-insensitive partial matching on agent name or ID +- **Deployment filter for exports**: Log exports support filtering by `deployment_slug` +- **Leaner export listing**: The list exports API now returns a trimmed set of fields per export + - [Documentation](/product/observability/logs-export) + +### Prompts +- **Render override allow-list**: Override params on the prompt render endpoints are now limited to documented inference parameters and parameters already saved on the prompt. Other keys are ignored + - [Documentation](/product/prompt-engineering-studio/prompt-api#updating-prompt-params-while-retrieving-the-prompt) + +### Model Catalog +- **Model and pricing refresh**: Updated model definitions and pricing from the Portkey Models repository for Anthropic, Bedrock, Vertex AI, Google, Azure AI, OpenRouter, Lightning AI, and others + - [Documentation](/product/model-catalog/portkey-models) + +### Fixes and Improvements +- **User search**: Users can now be searched by first and last name +- **Deployments**: The deployment list API supports filtering by `workspace_id` +- **Gateway sync**: Gateway sync now self-heals after repeated cache misses +- **OAuth refresh tokens**: A refresh token is now spent only once when concurrent requests race to use it +- **MCP connections**: Fixed how upstream MCP server connection details are recorded and reported, and widened storage for OAuth scopes and expiry values +- **Agent identities**: Agents can no longer renew access into a workspace they no longer belong to; token exchange timeout increased +- Updated dependencies to patch security vulnerabilities. + + + + + +## v1.27.0 +--- + +### New Providers +- **SCX.ai**: Added to the Model Catalog (`scx-ai`) as an OpenAI-compatible provider + - [Documentation](/integrations/llms) + +### Pricing +- **Model-level pricing adjustments**: Pricing multipliers can now be set on individual models within an integration, overriding the integration-level adjustment for that model + - [Documentation](/product/model-catalog/pricing-adjustments#per-model-pricing-adjustments) + +Per-model adjustments require Gateway version 2.22.0 or higher. + +### Model Catalog +- **Auto-sync new models**: Newly added base models are now synced to existing integrations automatically, respecting each integration's "allow all models" setting + - [Documentation](/product/model-catalog) + +### Analytics +- **Guardrails analytics**: New endpoints for guardrail summary, trend, and grouped data +- **Top MCP and A2A**: Analytics groups now include breakdowns for top MCP and A2A usage +- **Multi-value metadata filters**: Metadata filters accept multiple values per key, as a comma-separated string (`{"env":"prod,staging"}`) or an array, matched with OR + - [Documentation](/product/observability/analytics) + +### Guardrails Updates +- **Azure PII**: Added a `piiCategories` parameter to restrict detection to selected PII categories + - [Documentation](/integrations/guardrails/azure-guardrails#azure-pii-detection) +- **Alibaba Cloud AI Guardrails**: Added new security check types + - [Documentation](/product/guardrails) + +### Agent Identities +- **Owner lookup**: New endpoint to search for users who can own an agent identity +- **Licensed agent limit**: Agent identity creation is capped at the organisation's licensed amount; exceeding it returns `402` +- **Federation status**: The agent identity list response reports identity federation setup status + +### Security +- **Agent integrations**: `auth_type` on update (`PUT`) is now restricted to `headers` and `none`, matching create-time validation + +### Fixes and Improvements +- **Not found responses**: GET lookups for missing resources (virtual keys, configs, and others) now return `404` instead of `403`, so tools like Terraform can detect out-of-band deletions +- **SCIM**: Correct status code returned for SCIM user `PATCH` requests + - [Documentation](/product/enterprise-offering/org-management/scim/scim) +- **API keys**: Added API key list filters, with validation for invalid IDs +- **Guardrails**: Workspace and organisation guardrails are sanitised to prevent crashes +- **Usage limits**: Fixed a crash when default usage limits were not set +- **JWT**: Fixed user attribution for JWT requests that skip API key auth +- **MCP**: MCP user access on SCM routes now uses the global user ID +- **Stability**: Fixed a process crash on unsupported auto-describe types + + + ## v1.26.0 From 9e161fd97afff1c2a8e9a65571fd8027f85341d3 Mon Sep 17 00:00:00 2001 From: Mahesh Vagicherla Date: Tue, 29 Sep 2026 10:28:13 +0530 Subject: [PATCH 2/2] chore: cleanup --- changelog/backend.mdx | 32 +++++--------------------------- 1 file changed, 5 insertions(+), 27 deletions(-) diff --git a/changelog/backend.mdx b/changelog/backend.mdx index 96f66ec9e..fb56437a9 100644 --- a/changelog/backend.mdx +++ b/changelog/backend.mdx @@ -13,16 +13,11 @@ Discuss how Portkey's AI Gateway can enhance your organization's AI infrastructu ## v1.28.0 --- -### Logs & Exports -- **Agent search on logs**: The logs list API accepts an `agentSearch` query parameter for case-insensitive partial matching on agent name or ID -- **Deployment filter for exports**: Log exports support filtering by `deployment_slug` +### Log Exports +- **Deployment filter**: Log exports support filtering by `deployment_slug` - **Leaner export listing**: The list exports API now returns a trimmed set of fields per export - [Documentation](/product/observability/logs-export) -### Prompts -- **Render override allow-list**: Override params on the prompt render endpoints are now limited to documented inference parameters and parameters already saved on the prompt. Other keys are ignored - - [Documentation](/product/prompt-engineering-studio/prompt-api#updating-prompt-params-while-retrieving-the-prompt) - ### Model Catalog - **Model and pricing refresh**: Updated model definitions and pricing from the Portkey Models repository for Anthropic, Bedrock, Vertex AI, Google, Azure AI, OpenRouter, Lightning AI, and others - [Documentation](/product/model-catalog/portkey-models) @@ -30,10 +25,6 @@ Discuss how Portkey's AI Gateway can enhance your organization's AI infrastructu ### Fixes and Improvements - **User search**: Users can now be searched by first and last name - **Deployments**: The deployment list API supports filtering by `workspace_id` -- **Gateway sync**: Gateway sync now self-heals after repeated cache misses -- **OAuth refresh tokens**: A refresh token is now spent only once when concurrent requests race to use it -- **MCP connections**: Fixed how upstream MCP server connection details are recorded and reported, and widened storage for OAuth scopes and expiry values -- **Agent identities**: Agents can no longer renew access into a workspace they no longer belong to; token exchange timeout increased - Updated dependencies to patch security vulnerabilities. @@ -66,27 +57,14 @@ Discuss how Portkey's AI Gateway can enhance your organization's AI infrastructu ### Guardrails Updates - **Azure PII**: Added a `piiCategories` parameter to restrict detection to selected PII categories - [Documentation](/integrations/guardrails/azure-guardrails#azure-pii-detection) -- **Alibaba Cloud AI Guardrails**: Added new security check types +- **Alibaba Cloud AI Guardrails**: Added new check types - [Documentation](/product/guardrails) -### Agent Identities -- **Owner lookup**: New endpoint to search for users who can own an agent identity -- **Licensed agent limit**: Agent identity creation is capped at the organisation's licensed amount; exceeding it returns `402` -- **Federation status**: The agent identity list response reports identity federation setup status - -### Security -- **Agent integrations**: `auth_type` on update (`PUT`) is now restricted to `headers` and `none`, matching create-time validation - ### Fixes and Improvements - **Not found responses**: GET lookups for missing resources (virtual keys, configs, and others) now return `404` instead of `403`, so tools like Terraform can detect out-of-band deletions -- **SCIM**: Correct status code returned for SCIM user `PATCH` requests +- **SCIM**: Correct status returned for SCIM user `PATCH` requests - [Documentation](/product/enterprise-offering/org-management/scim/scim) -- **API keys**: Added API key list filters, with validation for invalid IDs -- **Guardrails**: Workspace and organisation guardrails are sanitised to prevent crashes -- **Usage limits**: Fixed a crash when default usage limits were not set -- **JWT**: Fixed user attribution for JWT requests that skip API key auth -- **MCP**: MCP user access on SCM routes now uses the global user ID -- **Stability**: Fixed a process crash on unsupported auto-describe types +- **API keys**: Added filters to the API key list endpoint