Portkey Signals

Counted for All practitioners.

Positive, 16

1. Positive

   Jaraxxxx, owner of an OpenMontage fork

   feat(video): Portkey gateway video generation (omni-flash + Veo 3.1)GitHub pull request

   Adds video generation through a single Portkey API key with no per-vendor credentials, covering Gemini and Veo 3.1 behind one gateway.

   Published Oct 24, 2026self-described: OpenMontage video generation through one Portkey API key

2. Positive

   DEENUU1, vstorm-co team member

   feat: add an OpenAI-compatible provider for any gateway or serverGitHub pull request

   Replaces gateway-specific providers with one OpenAI-compatible provider that reaches LiteLLM, Requesty, OrcaRouter and Portkey by base URL.

   Published Oct 6, 2026self-described: vstorm's full-stack AI agent template

3. Positive

   Tim, founder building RevenueFlow

   Every "AI SDR" on the market is a wrapper around the same 60 tools.X post

   Lists Portkey beside OpenRouter and LiteLLM as model routing in the 19 tools RevenueFlow runs in production.

   "…the 19 we run in production: THE BRAIN 1. Models: @AnthropicAI, @OpenAI, Google Gemini 2. Model routing: @OpenRouter, @LiteLLM, Portkey…"

   Published Sep 11, 2026self-described: RevenueFlow's production stack

4. Positive

   AllenGrahamHart, collaborator on terminal-bench-science

   ci: inject the model-gateway host into agent-phase allowlistsGitHub pull request

   Terminal-bench-science's CI can send agent traffic through Portkey as one of four model gateways, adding api.portkey.ai to network allowlists.

   Published Aug 22, 2026self-described: terminal-bench-science CI, which can run agents through Portkey among four model gateways

5. Positive

   lchoquel, Pipelex team member

   Per-request Gateway API keys: survey, Temporal design, and pipelex-server handoff (docs only)GitHub pull request

   Pipelex designs a separate Portkey gateway key for each hosted run, bound per request under its Temporal workflows, to replace the single key cached in its SDK client.

   Published Aug 18, 2026self-described: Pipelex, which routes hosted runs through a Portkey gateway

6. Mixed

   sharon-wang, Posit Positron team member

   Assistant: Portkey gateway provider auth supportGitHub issue

   Posit Assistant added an experimental Portkey provider covering hosted Portkey and the self-hosted gateway, while Positron's direct path still lacks a sign-in for it.

   Published Aug 12, 2026self-described: Positron Assistant's experimental Portkey gateway provider

7. Positive

   sharon-wang, Posit ai-lib team member

   Custom provider: support Anthropic Messages API for Claude modelsGitHub issue

   Posit's ai-lib already sends Claude requests to Portkey in Anthropic's Messages format and other models in OpenAI's, and plans the same split for generic gateways.

   Published Jul 30, 2026self-described: Posit's ai-lib provider bridge

8. Positive

   Akashkhurana Hirana, Senior Software Engineer 2 at Porch

   Centralized AI gateway has streamlined model routing, costs tracking, and data protectionReview, PeerSpot

   Porch's team calls Gemini, OpenAI, Claude and its own fine-tuned models through Portkey, with provider keys held in its vault and guardrails that block personal data, and calls the documentation thin and the dashboard hard to learn.

   Published Jun 27, 2026self-described: Porch, which calls Gemini, OpenAI, Claude and fine-tuned models through Portkey

9. Positive

   Shravan Revanna, Software Engineer at Clydo (klydo.in)

   Centralized AI control has standardized our workflows and delivers faster, more reliable featuresReview, PeerSpot

   At Clydo every user query from the internal assistant goes through Portkey, which handles authentication and routing, and the team swaps models by updating routing config and skips code changes across services; it asks for richer historical analytics and finer role-based access.

   Published Jun 26, 2026self-described: Clydo, which routes internal copilots, catalog intelligence and support assistants through Portkey

10. Positive

    xinguaws, GitHub user running a LibreChat deployment

    [Enhancement]: custom endpoint implementation with openai responses api not compatible with anthropic native web searchGitHub issue

    Runs LibreChat against Claude on Vertex AI through a Portkey gateway, with each user's email passed as Portkey metadata.

    "…it's using claude models via google's vertex ai through portkey gateway…"

    Published Jun 23, 2026self-described: A LibreChat deployment at ai.drwcloud.com calling Claude on Vertex AI through Portkey

11. Positive

    sseshachala, owner of the ConductAI repository

    feat: Portkey gateway integration for enterprise routingGitHub issue

    An enterprise prospect asked ConductAI to route through its existing Portkey setup, which it treats as the organization's central AI governance layer, so ConductAI plans a Portkey integration.

    Published Jun 9, 2026self-described: ConductAI, asked by an enterprise prospect to route through its existing Portkey setup

12. Positive

    sseshachala, owner of the ConductAI repository

    docs(sales): Portkey onboarding friction notes for demoGitHub issue

    Finds getting a Portkey key very low friction and the one setup step an admin adding provider keys as virtual keys in about 5 minutes, after which developers never see the raw keys.

    Published Jun 9, 2026self-described: Notes from a sales follow-up on the Portkey integration

13. Positive

    SinghAman, Generative AI Engineer at an outsourcing company with 10, 001+ employees

    Unified ai gateway has standardized observability and routing for multiple llm applicationsReview, PeerSpot

    After two years as the gateway and observability layer, Portkey has been reliable about 99% of the time, with endpoints down once or twice in six months; onboarding needs more guided templates.

    Published Jun 5, 2026self-described: Two years of Portkey as the gateway and observability layer for LLM applications with US and EU endpoints

14. Positive

    evalstate, owner of the fast-agent repository

    OpenAI Providers custom HTTP HeadersGitHub pull request

    Adds configurable HTTP headers to fast-agent's providers so agents can route through Portkey configs and other gateways that need custom headers.

    Published Dec 6, 2025self-described: fast-agent's OpenAI-compatible providers

15. Positive

    clayroach, developer of an open-source observability platform

    Removing 11,005 Lines: Why We Replaced Our Custom LLM Manager with PortkeyEngineering blog post, DEV Community

    Replacing the team's own LLM manager with Portkey's gateway removed more than 11,000 lines of code while keeping routing and adding functionality.

    "Today I want to share how replacing our custom LLM manager with Portkey's gateway removed over 11,000 lines of code from our observability platform while actually improving functionality."

    Published Sep 16, 2025self-described: An observability platform that replaced its own LLM manager with Portkey's gateway

16. Positive

    chandureddyvari, Hacker News commenter

    Comment on: HN Slop: AI startup ideas generated from Hacker NewsHacker News comment

    Recommends a gateway such as OpenRouter or Portkey to manage fallbacks after an app hit provider rate limits.

    "You should use something like openrouter or portkey or similar for managing fallbacks…"

    Published Jul 1, 2025self-described: An app hitting provider rate limits

Mixed or negative, 1

1. Negative

   sakshamyadav090, GitHub user

   Streaming: when the upstream closes the stream mid-answer, the gateway ends its own response cleanlyGitHub issue

   When a provider drops a stream mid-answer, the gateway returns HTTP 200 with no error event and never tries the configured fallback, so the client shows a short answer as complete.

   "The configured fallback target is never tried."

   Published Sep 9, 2026self-described: Streaming with a configured fallback

GILD's fact checks, which don't count as sources

Checked Oct 28, 2026.Portkey's pricing page lists a universal API with fallbacks, load balancing, and retries on every plan, including the free Developer plan. Pricing

Checked Oct 28, 2026.Portkey's Claude Code integration routes Claude Code through Anthropic, Bedrock, or Vertex AI with workspaces, budgets, and rate limits per team. Claude Code integration

Checked Oct 28, 2026.Portkey documents a LangChain (Python) integration that sends LangChain calls and agents through the gateway. LangChain (Python) integration

Checked Oct 28, 2026.Portkey's AWS Bedrock integration calls Bedrock models on your own AWS credentials through the gateway. AWS Bedrock integration

Counted for All practitioners.

1. Mixed

   kwisatzh, GitHub user comparing four gateways

   OpenAI-format usage for Anthropic cache reads: cached_tokens exceeds prompt_tokensGitHub issue

   Running the same requests through four gateways, Portkey 1.15.2 matched provider usage on ordinary requests and kept every Anthropic cache field on the Messages path, with one cache-token translation error on the OpenAI-format path.

   "Portkey v1.15.2 did well: ordinary requests matched the provider, and the native Messages path kept every Anthropic cache field."

   Published Oct 5, 2026self-described: Usage accounting across four open-source gateways

2. Positive

   Ezekiel Oluyale, AI/ML engineer

   You can't fix what you can't see.X post

   Tracks cost, latency, and time to first token per request with Portkey.

   "I track cost, latency, TTFT per request with Portkey. Now I can see: → Cost per 1k tokens → Time to first token (TTFT) → Total latency…"

   Published Sep 7, 2026self-described: Personal LLM app monitoring

3. Positive

   lchoquel, Pipelex team member

   Per-request Gateway API keys: survey, Temporal design, and pipelex-server handoff (docs only)GitHub pull request

   Pipelex designs a separate Portkey gateway key for each hosted run, bound per request under its Temporal workflows, to replace the single key cached in its SDK client.

   Published Aug 18, 2026self-described: Pipelex, which routes hosted runs through a Portkey gateway

4. Positive

   Akashkhurana Hirana, Senior Software Engineer 2 at Porch

   Centralized AI gateway has streamlined model routing, costs tracking, and data protectionReview, PeerSpot

   Porch's team calls Gemini, OpenAI, Claude and its own fine-tuned models through Portkey, with provider keys held in its vault and guardrails that block personal data, and calls the documentation thin and the dashboard hard to learn.

   Published Jun 27, 2026self-described: Porch, which calls Gemini, OpenAI, Claude and fine-tuned models through Portkey

5. Positive

   Shravan Revanna, Software Engineer at Clydo (klydo.in)

   Centralized AI control has standardized our workflows and delivers faster, more reliable featuresReview, PeerSpot

   At Clydo every user query from the internal assistant goes through Portkey, which handles authentication and routing, and the team swaps models by updating routing config and skips code changes across services; it asks for richer historical analytics and finer role-based access.

   Published Jun 26, 2026self-described: Clydo, which routes internal copilots, catalog intelligence and support assistants through Portkey

6. Positive

   xinguaws, GitHub user running a LibreChat deployment

   [Enhancement]: custom endpoint implementation with openai responses api not compatible with anthropic native web searchGitHub issue

   Runs LibreChat against Claude on Vertex AI through a Portkey gateway, with each user's email passed as Portkey metadata.

   "…it's using claude models via google's vertex ai through portkey gateway…"

   Published Jun 23, 2026self-described: A LibreChat deployment at ai.drwcloud.com calling Claude on Vertex AI through Portkey

7. Positive

   SinghAman, Generative AI Engineer at an outsourcing company with 10, 001+ employees

   Unified ai gateway has standardized observability and routing for multiple llm applicationsReview, PeerSpot

   After two years as the gateway and observability layer, Portkey has been reliable about 99% of the time, with endpoints down once or twice in six months; onboarding needs more guided templates.

   Published Jun 5, 2026self-described: Two years of Portkey as the gateway and observability layer for LLM applications with US and EU endpoints

8. Positive

   jychp, Cartography collaborator

   feat(portkey): add intel moduleGitHub pull request

   Adds a Cartography module that inventories Portkey organizations, workspaces, API keys, virtual keys, guardrails and MCP servers for security review.

   Published May 6, 2026self-described: Cartography's security inventory of Portkey organizations

GILD's fact checks, which don't count as sources

Checked Oct 28, 2026.Developer keeps logs 3 days and metrics 30 days; Production keeps logs 30 days and metrics 90 days; Enterprise sets custom retention periods. Pricing

Counted for All practitioners.

Raised as a concern, 14

1. Negative

   dr-goo, GitHub user

   Google / Vertex AI streaming drops functionCall after thought or text partGitHub issue

   Streaming responses from Gemini and Vertex AI drop tool calls that follow a thought or text part and report the wrong finish reason.

   Published Oct 26, 2026self-described: Streaming tool calls on Gemini and Vertex AI

2. Negative

   sheredega303, GitHub user calling Amazon Titan embeddings through Portkey

   [BUG] embed(array) returns one vector for many texts with Bedrock Titan behind an OpenAI-compatible gatewayGitHub issue

   With Portkey in front of Amazon Titan embeddings, a batch request returned one embedding for the first text and silently dropped the rest.

   "We hit this with Portkey in front of Amazon Titan, which takes one input per request. The gateway returns a single vector for the first text and drops the rest."

   Published Oct 22, 2026self-described: RubyLLM embeddings on Amazon Titan through Portkey

3. Negative

   xinliu0727-sudo, GitHub user

   POST /v1/messages with x-portkey-provider: openai or google returns HTTP 500GitHub issue

   Anthropic Messages requests to OpenAI or Google models fail with HTTP 500 on gateway 1.15.2 before reaching the provider.

   "…fails with HTTP 500 on Gateway `v1.15.2`…"

   Published Oct 15, 2026self-described: Anthropic-format requests to OpenAI and Google models

4. Negative

   xinliu0727-sudo, GitHub user

   Google provider: system messages that are not the first message are dropped from contentsGitHub issue

   System messages after the first one, and assistant text before tool calls, disappear without an error when requests are routed to Gemini.

   "…every `system` message is skipped while building `contents`…"

   Published Oct 7, 2026self-described: OpenAI-format requests routed to Gemini

5. Mixed

   kwisatzh, GitHub user comparing four gateways

   OpenAI-format usage for Anthropic cache reads: cached_tokens exceeds prompt_tokensGitHub issue

   Running the same requests through four gateways, Portkey 1.15.2 matched provider usage on ordinary requests and kept every Anthropic cache field on the Messages path, with one cache-token translation error on the OpenAI-format path.

   "Portkey v1.15.2 did well: ordinary requests matched the provider, and the native Messages path kept every Anthropic cache field."

   Published Oct 5, 2026self-described: Usage accounting across four open-source gateways

6. Negative

   chrikrah, GitHub user

   Bedrock /v1/messages streaming drops exception eventsGitHub issue

   Bedrock stream exceptions never reach the client on the Messages route, which sees an empty HTTP 200 or a stream that stops without an error.

   "Bedrock /v1/messages streaming drops exception events: HTTP 200 with an empty body, or a stream that stops with no error event ### What Happened? On `/v1/messages` with a Bedrock model, an exception event in the ConverseStream response never reaches the client. `BedrockConverseMessagesStreamChunkTransform` (`src/providers/bedrock/messages.ts:546`) has no branch for its `{"message": ...}` payload and returns `undefined`. On `669825c`, with synthesized event-stream frames: ``` $ npx tsx repro.ts normal: HTTP 200, 859 bytes, events [message_start, content_block_start, content_block_delta, content_block_stop, message_delta, message_stop] exception only: HTTP 200, 0 bytes, events [] exception after a delta: HTTP 200, 580 bytes, events [message_start, content_block_start, content_block_delta] ``` Both exception cases lose the message from the provider, and the second stream just stops. This is the `/v1/messages` sibl…"

   Published Oct 3, 2026self-described: Bedrock models on the Messages route

7. Mixed

   joelp22-maker, maintainer of a model retirement dataset

   Optional lifecycle field on src/data/models.json: 95 listed models are already retired at their providerGitHub issue

   Of 126 catalog entries matched against provider retirement dates, 95 are models their provider has retired, and the 1,928-model catalog carries no lifecycle field.

   Published Sep 20, 2026self-described: The gateway's bundled model catalog

8. Negative

   arohou, GitHub user

   Bedrock: tool-result cachePoint nested where the AWS schema forbids itGitHub issue

   Three defects in the Bedrock translation shorten prompt-cache lifetimes, and one makes a valid request with a one-hour cache fail outright.

   "Three related defects in the Bedrock provider's Converse translation, all around prompt-cache checkpoints."

   Published Aug 11, 2026self-described: Prompt caching on Bedrock through the gateway

9. Negative

   TCK0423, GitHub user

   Gemini function calling: role function not transformed to role MODELGitHub issue

   Routing Hermes Agent and Gemini through Portkey fails on function calls because Portkey passes the OpenAI function role to Gemini unchanged.

   "When routing Hermes Agent + Google Gemini through Portkey Gateway, function calling fails with:…"

   Published Aug 9, 2026self-described: Hermes Agent with Google Gemini through Portkey's hosted API

10. Negative

    sophiaashi, GitHub user

    moonshot provider model list is four generations staleGitHub issue

    The gateway's model list names only three pre-Kimi K Moonshot models and none of the four generations since, so the integration looks abandoned.

    "…lists exactly three models under the `moonshot` provider…"

    Published Jul 28, 2026self-described: The gateway's bundled model catalog

11. Negative

    davidwhite-spec, GitHub user

    azure-ai-inference: no Responses API support, requests silently fall back to chatComplete with data lossGitHub issue

    Responses API requests to the Azure AI Inference provider return no error and are silently translated to chat completions, dropping fields on the way.

    "The `azure-ai-inference` provider has no Responses API support."

    Published May 30, 2026self-described: Responses API calls to an Azure AI Inference key

12. Mixed

    thisisnic, ellmer collaborator

    Fix chat_portkey() error when PORTKEY_VIRTUAL_KEY is not setGitHub pull request

    After Portkey's virtual key deprecation, ellmer's Portkey provider required a virtual key for any model outside the @provider/model format, which broke custom gateways passing Bedrock ARNs.

    Published May 26, 2026self-described: ellmer's chat_portkey() provider for R

13. Mixed

    harsh-m-patil, Pi user

    Claude Opus 4.7 fails on Bedrock via custom provider with reasoning enabled (wrong thinking format)GitHub issue

    Claude Opus 4.7 on Bedrock fails with a 400 thinking-format error when Pi calls it through Portkey with reasoning enabled.

    Published Apr 17, 2026self-described: Pi coding agent calling Claude on Bedrock through Portkey

14. Negative

    Nilesh Darji, Founding Engineer at Heizen (his own site)

    Tool calls and file inputs fail with Responses API via Portkey for OpenAI, Azure OpenAI, and BedrockGitHub issue

    Tool calls failed on the Responses API through Portkey and worked against the provider directly. Bedrock returned errors for empty message content and for more tool results than tool calls, Azure OpenAI returned ModelNotFound for a model that exists, and file inputs failed on both.

    Published Mar 28, 2026self-described: LangChain's ChatOpenAI on the Responses API, through Portkey's hosted API to Azure OpenAI and Bedrock

Discussed without concern, 1

1. Mixed

   mldangelo, Promptfoo team member

   fix(providers): align Portkey credentials with request routingGitHub pull request

   Fixes how Promptfoo's Portkey provider picks credentials, since Portkey separates its gateway key from upstream keys, virtual keys and catalog slugs and each route needs the right one.

   Published Sep 9, 2026self-described: Promptfoo's Portkey provider

Counted for All practitioners.

Raised as a concern, 8

1. Mixed

   shafiktanbir, GitHub user

   Live logs middleware hardcodes the port and shows an undefined endpointGitHub issue

   The self-hosted gateway's live log view shows an undefined endpoint whenever it runs on a custom port or behind a reverse proxy.

   "…in the Endpoint column instead of the request path…"

   Published Oct 26, 2026self-described: Self-hosted gateway on a custom port

2. Mixed

   dan-petty, owner of the devops-cli repository

   fix(k8s): the portkey gateway reads its routing config in the documented way, or the unused config goesGitHub issue

   Self-hosts gateway 1.15.0 in Kubernetes and found the mounted routing config file is never read by the gateway image, so its routes had no effect on live traffic.

   Published Oct 24, 2026self-described: devops-cli, which runs the Portkey gateway in its own Kubernetes cluster

3. Negative

   chrikrah, GitHub user

   Bedrock /v1/messages streaming drops exception eventsGitHub issue

   Bedrock stream exceptions never reach the client on the Messages route, which sees an empty HTTP 200 or a stream that stops without an error.

   "Bedrock /v1/messages streaming drops exception events: HTTP 200 with an empty body, or a stream that stops with no error event ### What Happened? On `/v1/messages` with a Bedrock model, an exception event in the ConverseStream response never reaches the client. `BedrockConverseMessagesStreamChunkTransform` (`src/providers/bedrock/messages.ts:546`) has no branch for its `{"message": ...}` payload and returns `undefined`. On `669825c`, with synthesized event-stream frames: ``` $ npx tsx repro.ts normal: HTTP 200, 859 bytes, events [message_start, content_block_start, content_block_delta, content_block_stop, message_delta, message_stop] exception only: HTTP 200, 0 bytes, events [] exception after a delta: HTTP 200, 580 bytes, events [message_start, content_block_start, content_block_delta] ``` Both exception cases lose the message from the provider, and the second stream just stops. This is the `/v1/messages` sibl…"

   Published Oct 3, 2026self-described: Bedrock models on the Messages route

4. Negative

   sakshamyadav090, GitHub user

   Streaming: when the upstream closes the stream mid-answer, the gateway ends its own response cleanlyGitHub issue

   When a provider drops a stream mid-answer, the gateway returns HTTP 200 with no error event and never tries the configured fallback, so the client shows a short answer as complete.

   "The configured fallback target is never tried."

   Published Sep 9, 2026self-described: Streaming with a configured fallback

5. Negative

   meitalbensinai, GitHub user

   Streamed responses fail under concurrent load with "Response object has been garbage collected"GitHub issue

   Under concurrent streaming load the gateway ends streams with garbage-collection errors, and downstream clients see terminated responses.

   "Under concurrent streaming load, gateway logs show:…"

   Published May 19, 2026self-described: Self-hosted gateway under concurrent streaming load

6. Negative

   Yupei Chen, Wise Rock (GitHub profile)

   Intermittent auth failures with Claude Code OAuth passthrough (mid-session)GitHub issue

   Mid-session, Claude Code behind Portkey's hosted API returned auth errors while still showing the user signed in, and only unsetting the Portkey base URL fixed it. After reading the gateway source, the author concluded that OAuth tokens are never forwarded correctly.

   Published Apr 10, 2026self-described: Claude Code signed in through Anthropic OAuth on a Team plan, pointed at Portkey's hosted API as Portkey's OAuth docs describe

7. Negative

   ianbbqzy, GitHub user running the gateway in production

   Connection errors in productionGitHub issue

   Occasionally sees errors likely related to connection health in production traffic, with OpenAI, Anthropic, and Mistral, that do not reproduce locally even under a load test of more than 40 concurrent requests.

   "I'm occasionally running into several errors likely related to connection health in my production traffic."

   Published Sep 24, 2025self-described: Production traffic through the gateway, with errors seen on OpenAI, Anthropic, and Mistral

8. Negative

   Leonardo Marciano, CTO by his own GitHub profile

   Bug Report: Simple Cache Incorrectly Converted to Semantic Cache Causing Platform OutageGitHub issue

   A Portkey change turned the author's platform's simple cache into a semantic cache with a hit rate near 100%, which caused a platform-wide outage for all its customers on Portkey cloud; Portkey acknowledged the bug and reverted it the same day.

   "This caused a complete platform failure while all customers were actively using the system."

   Published Jul 23, 2025self-described: A platform on Portkey's hosted cloud

Discussed without concern, 1

1. Positive

   SinghAman, Generative AI Engineer at an outsourcing company with 10, 001+ employees

   Unified ai gateway has standardized observability and routing for multiple llm applicationsReview, PeerSpot

   After two years as the gateway and observability layer, Portkey has been reliable about 99% of the time, with endpoints down once or twice in six months; onboarding needs more guided templates.

   Published Jun 5, 2026self-described: Two years of Portkey as the gateway and observability layer for LLM applications with US and EU endpoints

Counted for All practitioners.

1. Negative

   mrSingh007, GitHub user

   Email address mentioned under contact us not workingGitHub issue

   The support address linked from the self-hosted gateway bounced as nonexistent.

   "Getting error `The email account that you tried to reach does not exist`."

   Published Sep 15, 2026self-described: Seeking support for the self-hosted gateway

2. Negative

   pritamsoni-hsr, GitHub user

   is this going to be maintained?GitHub issue

   Asks after the Palo Alto Networks acquisition whether the open-source gateway will be maintained, pointing to a fork.

   Published Aug 5, 2026self-described: A self-hosted gateway user after the acquisition

3. Negative

   sophiaashi, GitHub user

   moonshot provider model list is four generations staleGitHub issue

   The gateway's model list names only three pre-Kimi K Moonshot models and none of the four generations since, so the integration looks abandoned.

   "…lists exactly three models under the `moonshot` provider…"

   Published Jul 28, 2026self-described: The gateway's bundled model catalog

4. Negative

   geo-chen, security researcher

   SSRF allowlist bypass and credential theft via /v1/proxy/* route, @portkey-ai/gateway 1.15.2 and earlierGitHub issue

   Discloses that the deprecated /v1/proxy/* route skips the allowlist check behind the v1.14.0 fix for CVE-2025-66405, allowing SSRF and credential theft, after an email report on June 1, 2026 got no response.

   "Sending any POST request to /v1/proxy/* bypasses all allowlist checks…"

   Published Jul 3, 2026self-described: Security review of the open-source gateway

5. Negative

   damray, GitHub user

   Build fails on 2.0.0: akto plugin committed to plugins/akto/ instead of src/plugins/akto/GitHub issue

   A clean checkout of the gateway's 2.0.0 branch does not build because a guardrail plugin was committed to the wrong directory.

   Published May 30, 2026self-described: Building the gateway's 2.0.0 branch

GILD's fact checks, which don't count as sources

Checked Oct 28, 2026.Palo Alto Networks announced general availability of Prisma AIRS AI Gateway, built on Portkey, on July 16, 2026. Palo Alto Networks: General availability of Prisma AIRS AI Gateway