Merge pull request #6956 from Kilo-Org/session/agent_24c88d05-f8c4-400c-81ae-fb27077b2072

Remove outdated vercel-ai-gateway architecture doc
2026-09-24 16:02:55 +08:00 · 2026-03-12 10:29:56 +01:00
parent 96733df56e e4558d8073
commit e328c8485f
4 changed files with 6 additions and 82 deletions
@@ -66,10 +66,6 @@ export const ContributingNav: NavSection[] = [
            href: "/contributing/architecture/track-repo-url",
            children: "Track Repo URL",
          },
-          {
-            href: "/contributing/architecture/vercel-ai-gateway",
-            children: "Vercel AI Gateway",
-          },
          {
            href: "/contributing/architecture/voice-transcription",
            children: "Voice Transcription",
@@ -18,7 +18,6 @@ These pages document the architecture and design of current or planned features,
 | [Organization Modes Library](/docs/contributing/architecture/organization-modes-library) | Shared modes for teams and enterprise                |
 | [Agentic Security Reviews](/docs/deploy-secure/security-reviews)                         | AI-powered security vulnerability analysis           |
 | [Track Repo URL](/docs/contributing/architecture/track-repo-url)                         | Usage tracking by repository/project                 |
-| [Vercel AI Gateway](/docs/contributing/architecture/vercel-ai-gateway)                   | Vercel AI Gateway integration                        |
 | [Voice Transcription](/docs/contributing/architecture/voice-transcription)               | Live voice input for chat                            |

 To propose a new feature design, consider using the [Spec Template](/docs/contributing/architecture/feature-template).
@@ -1,77 +0,0 @@
---
-title: "Vercel AI Gateway"
-description: "Vercel AI Gateway architecture"
---
-
-# OpenRouter and Vercel AI Gateway API compatibility
-
-Both OpenRouter and Vercel AI Gateway provide an OpenAI-compatible API.
-They are largely compatible and Vercel even adopted some of OpenRouter's non-standard extensions.
-
-OpenRouter API docs: https://openrouter.ai/docs/api/reference/overview
-
-Vercel AI Gateway API docs: https://vercel.com/docs/ai-gateway/openai-compat
-
-At time of writing, OpenRouter provides beta support for the Responses API, but Vercel does not: https://openrouter.ai/docs/api/reference/responses/overview
-
-## Available models and identifiers
-
-OpenRouter has more models than Vercel. The naming scheme is similar, sometimes identical but not always. For example `x-ai/grok-code-fast-1` (OpenRouter) has a different prefix from `xai/grok-code-fast-1` (Vercel). We will have to come up with a mapping from OpenRouter to Vercel model identifiers.
-
- https://openrouter.ai/models
- https://vercel.com/ai-gateway/models
-
-## Reasoning
-
-Vercel seems to have adopted OpenRouter's (non-standard) reasoning configuration:
-
- https://openrouter.ai/docs/api/reference/responses/reasoning#reasoning-configuration
- https://vercel.com/docs/ai-gateway/openai-compat#reasoning-configuration
-
-Both Vercel and OpenRouter use the same `reasoning` property for reasoning content (the standard is `reasoning_content`, although our extension will accept both):
-
- https://openrouter.ai/docs/api/api-reference/chat/send-chat-completion-request#response.body.choices.message.reasoning
- https://vercel.com/docs/ai-gateway/openai-compat#response-format-with-reasoning
-
-Vercel supports OpenRouter's way of preserving reasoning details in the Chat Completions API:
-
- https://openrouter.ai/docs/guides/best-practices/reasoning-tokens#preserving-reasoning-blocks
- https://vercel.com/docs/ai-gateway/openai-compat#preserving-reasoning-details-across-providers
-
-Other providers would require use of the Responses API (e.g. xAI) or Anthropic API (e.g. MiniMax)
-
-## Provider configuration
-
-OpenRouter's and Vercel's provider configuration are different:
-
- https://openrouter.ai/docs/guides/routing/provider-selection
- https://vercel.com/docs/ai-gateway/openai-compat#provider-options
-
-We will likely support OpenRouter's and convert between the two.
-
-Inference provider identifiers appear similar at first glance, but they may not always be identical.
-
-## Cost calculation
-
-Whether Vercel includes cost in the response (like OpenRouter) is unclear. In the docs there's an embeddings example, which suggests that it does, but in a different format than OpenRouter:
-
- https://vercel.com/docs/ai-gateway/openai-compat#embeddings
-
-We may have to estimate the cost in the extension instead.
-
-There's a generation endpoint, which looks very similar to OpenRouter's at first glance:
-
- https://openrouter.ai/docs/api/api-reference/generations/get-generation
- https://vercel.com/docs/ai-gateway/usage#generation-lookup
-
-# Deciding where to route traffic
-
-We want to route some traffic to Vercel, ideally just enough so we can be first place on both leaderboards. However, we will initially start with a small portion of traffic < 5% and will evaluate whether it make sense to increase volume.
-
-Vercel will also act as fallback if we encounter issues with OpenRouter (such as the issue with Grok Code Fast and OpenRouter recently). We need the ability to route certain providers and / or models to Vercel vs OpenRouter.
-
-Requests from the same user for the same model should go to the same gateway for caching purposes.
-
-We can randomize users to a gateway in a way similar to what was done for autocomplete: https://github.com/Kilo-Org/kilocode/pull/3857
-
-Since not all models are available on Vercel, some traffic will have to go to OpenRouter regardless.
@@ -801,4 +801,10 @@ module.exports = [
    basePath: false,
    permanent: true,
  },
+  {
+    source: "/docs/contributing/architecture/vercel-ai-gateway",
+    destination: "/docs/contributing/architecture/features",
+    basePath: false,
+    permanent: true,
+  },
 ]