mirror of
https://github.com/fosrl/pangolin.git
synced 2026-08-21 03:32:33 +02:00
Rename to v1_models and use with openai as well
This commit is contained in:
Owen
@@ -7,7 +7,7 @@ inference resource has more than one AI provider.
|
||||
|
||||
- Route → capability binding: `server/routers/aiGateway/createAiGatewayRouter.ts`
|
||||
- Request pipeline: `server/routers/aiGateway/pipeline.ts` (`selectProvider`)
|
||||
- Model discovery: `server/routers/aiGateway/anthropicModels.ts` and
|
||||
- Model discovery: `server/routers/aiGateway/v1Models.ts` and
|
||||
`server/lib/aiModelDiscovery.ts`
|
||||
- Tie-break scoring: `server/lib/aiProviderSelection.ts`
|
||||
- Allow/block matching: `server/lib/aiModelKeyMatch.ts`
|
||||
@@ -41,7 +41,7 @@ The incoming path selects a capability before any provider logic runs.
|
||||
| `POST /v1/chat/completions` | `openai_chat` |
|
||||
| `POST /v1/responses` | `openai_responses` |
|
||||
| `POST /v1/messages` | `anthropic_messages` |
|
||||
| `GET /v1/models`, `GET /v1/models/{id}` | `anthropic_models` |
|
||||
| `GET /v1/models`, `GET /v1/models/{id}` | `v1_models` |
|
||||
| Gemini / Vertex / Bedrock routes | their respective capability ids |
|
||||
|
||||
Only attached providers that advertise that capability stay in the candidate
|
||||
@@ -50,10 +50,10 @@ set. Default capabilities do not overlap for native OpenAI vs Anthropic:
|
||||
| Provider type | Default capabilities |
|
||||
|---------------|----------------------|
|
||||
| `openai` | `openai_chat`, `openai_responses` |
|
||||
| `anthropic` | `anthropic_messages`, `anthropic_models` |
|
||||
| `anthropic` | `anthropic_messages`, `v1_models` |
|
||||
| `openRouter` | `openai_chat` |
|
||||
| `vercelAiGateway` | `openai_chat`, `openai_responses` |
|
||||
| `microsoftFoundry` | `openai_chat`, `openai_responses`, `anthropic_messages`, `anthropic_models` |
|
||||
| `microsoftFoundry` | `openai_chat`, `openai_responses`, `anthropic_messages`, `v1_models` |
|
||||
| `custom` | whatever was configured |
|
||||
|
||||
### 2. Allow / Block Lists
|
||||
@@ -133,10 +133,10 @@ customs advertising the same capability for an unknown model.
|
||||
|
||||
## Model Discovery Is Not Selection
|
||||
|
||||
`GET /v1/models` and `GET /v1/models/{id}` (`anthropic_models`) skip steps 3-6
|
||||
`GET /v1/models` and `GET /v1/models/{id}` (`v1_models`) skip steps 3-6
|
||||
entirely. There is no requested model to disambiguate on, so the gateway does
|
||||
not pick one provider - it returns the **union** of what every attached
|
||||
provider advertising `anthropic_models` would accept, deduplicated by model id
|
||||
provider advertising `v1_models` would accept, deduplicated by model id
|
||||
(lowest `providerId` wins a collision).
|
||||
|
||||
Discovery is answered from the gateway's own view of the allow/block lists,
|
||||
|
||||
+2
-2
@@ -1923,8 +1923,8 @@
|
||||
"aiCapabilityOpenaiResponsesDescription": "Supports /v1/responses",
|
||||
"aiCapabilityAnthropicMessages": "Anthropic Messages",
|
||||
"aiCapabilityAnthropicMessagesDescription": "Supports /v1/messages",
|
||||
"aiCapabilityAnthropicModels": "Anthropic Models",
|
||||
"aiCapabilityAnthropicModelsDescription": "Supports /v1/models model discovery",
|
||||
"aiCapabilityV1Models": "Models List",
|
||||
"aiCapabilityV1ModelsDescription": "Supports /v1/models model discovery",
|
||||
"aiCapabilityGeminiGenerateContent": "Gemini Generate Content",
|
||||
"aiCapabilityGeminiGenerateContentDescription": "Supports the direct Gemini API",
|
||||
"aiCapabilityBedrockModelInvoke": "Bedrock Model Invoke",
|
||||
|
||||
@@ -135,8 +135,8 @@ export const AI_CAPABILITY_DEFS: Record<AiCapability, AiCapabilityDefinition> =
|
||||
joinUpstreamUrl(base, pathFromRequest(req)),
|
||||
isStreaming: isBodyOrSseStreaming
|
||||
},
|
||||
anthropic_models: {
|
||||
id: "anthropic_models",
|
||||
v1_models: {
|
||||
id: "v1_models",
|
||||
protocolFamily: "anthropic",
|
||||
routes: [
|
||||
{ method: "GET", path: "/v1/models" },
|
||||
|
||||
@@ -472,7 +472,7 @@ const REQUEST_NORMALIZERS: Record<
|
||||
openai_responses: normalizeOpenAiResponsesRequest,
|
||||
anthropic_messages: normalizeAnthropicRequest,
|
||||
// Model discovery carries no transcript to normalize.
|
||||
anthropic_models: () => null,
|
||||
v1_models: () => null,
|
||||
gemini_generate_content: normalizeGeminiRequest,
|
||||
google_generate_content: normalizeGeminiRequest,
|
||||
google_raw_predict: normalizeBestEffortRequest,
|
||||
@@ -487,7 +487,7 @@ const RESPONSE_NORMALIZERS: Record<
|
||||
openai_chat: normalizeOpenAiChatResponse,
|
||||
openai_responses: normalizeOpenAiResponsesResponse,
|
||||
anthropic_messages: normalizeAnthropicResponse,
|
||||
anthropic_models: () => null,
|
||||
v1_models: () => null,
|
||||
gemini_generate_content: normalizeGeminiResponse,
|
||||
google_generate_content: normalizeGeminiResponse,
|
||||
google_raw_predict: normalizeGoogleRawPredictResponse,
|
||||
|
||||
@@ -336,7 +336,7 @@ const EXTRACTORS: Record<
|
||||
openai_responses: extractOpenAiResponses,
|
||||
anthropic_messages: extractAnthropicMessages,
|
||||
// Model discovery never runs a model, so there are no tokens to bill.
|
||||
anthropic_models: () => null,
|
||||
v1_models: () => null,
|
||||
gemini_generate_content: extractGoogleGenerateContent,
|
||||
google_generate_content: extractGoogleGenerateContent,
|
||||
// rawPredict is a passthrough to whatever the underlying publisher
|
||||
|
||||
@@ -4,7 +4,7 @@ import {
|
||||
type AiCapability
|
||||
} from "@server/lib/aiCapabilities";
|
||||
import { handleAiGatewayProxy } from "@server/routers/aiGateway/pipeline";
|
||||
import { handleAnthropicModels } from "@server/routers/aiGateway/anthropicModels";
|
||||
import { handleV1Models } from "@server/routers/aiGateway";
|
||||
|
||||
type CapabilityHandler = (
|
||||
req: Request,
|
||||
@@ -15,7 +15,7 @@ type CapabilityHandler = (
|
||||
// Capabilities the gateway answers itself instead of proxying upstream.
|
||||
// Everything else goes through the inference pipeline.
|
||||
const LOCAL_HANDLERS: Partial<Record<AiCapability, CapabilityHandler>> = {
|
||||
anthropic_models: handleAnthropicModels
|
||||
v1_models: handleV1Models
|
||||
};
|
||||
|
||||
export function createAiGatewayRouter() {
|
||||
|
||||
@@ -1,3 +1,3 @@
|
||||
export { handleAiGatewayProxy } from "./pipeline";
|
||||
export { handleAnthropicModels } from "./anthropicModels";
|
||||
export { handleV1Models } from "./v1Models";
|
||||
export { createAiGatewayRouter } from "./createAiGatewayRouter";
|
||||
|
||||
@@ -35,7 +35,7 @@ import {
|
||||
import logger from "@server/logger";
|
||||
import HttpCode from "@server/types/HttpCode";
|
||||
|
||||
const CAPABILITY: AiCapability = "anthropic_models";
|
||||
const CAPABILITY: AiCapability = "v1_models";
|
||||
|
||||
const querySchema = z.object({
|
||||
limit: z.coerce.number().int().min(1).max(MODEL_PAGE_MAX_LIMIT).optional(),
|
||||
@@ -163,7 +163,7 @@ function buildDiscoveryProviders(
|
||||
* list at all or would expose models the resource's allow/block lists forbid,
|
||||
* so the response is built from the same effective lists that gate inference.
|
||||
*/
|
||||
export async function handleAnthropicModels(
|
||||
export async function handleV1Models(
|
||||
req: Request,
|
||||
res: Response
|
||||
): Promise<any> {
|
||||
@@ -33,7 +33,7 @@ const capabilityLabels: Record<string, string> = {
|
||||
openai_chat: "OpenAI Chat Completions",
|
||||
openai_responses: "OpenAI Responses",
|
||||
anthropic_messages: "Anthropic Messages",
|
||||
anthropic_models: "Anthropic Models",
|
||||
v1_models: "Models List",
|
||||
gemini_generate_content: "Gemini",
|
||||
google_generate_content: "Vertex AI (Generate Content)",
|
||||
google_raw_predict: "Vertex AI (Raw Predict)",
|
||||
|
||||
@@ -20,7 +20,7 @@ const CAPABILITY_LABEL_KEYS: Record<AiCapability, string> = {
|
||||
openai_chat: "aiCapabilityOpenaiChat",
|
||||
openai_responses: "aiCapabilityOpenaiResponses",
|
||||
anthropic_messages: "aiCapabilityAnthropicMessages",
|
||||
anthropic_models: "aiCapabilityAnthropicModels",
|
||||
v1_models: "aiCapabilityV1Models",
|
||||
gemini_generate_content: "aiCapabilityGeminiGenerateContent",
|
||||
bedrock_model_invoke: "aiCapabilityBedrockModelInvoke",
|
||||
google_generate_content: "aiCapabilityGoogleGenerateContent",
|
||||
|
||||
@@ -2,7 +2,7 @@ export const AI_CAPABILITIES = [
|
||||
"openai_chat",
|
||||
"openai_responses",
|
||||
"anthropic_messages",
|
||||
"anthropic_models",
|
||||
"v1_models",
|
||||
"gemini_generate_content",
|
||||
"bedrock_model_invoke",
|
||||
"google_generate_content",
|
||||
|
||||
@@ -38,12 +38,12 @@ export const AI_PROVIDER_DEFAULTS: Record<
|
||||
openai: {
|
||||
upstreamUrl: "https://api.openai.com/v1",
|
||||
authType: "bearer",
|
||||
capabilities: ["openai_chat", "openai_responses"]
|
||||
capabilities: ["openai_chat", "openai_responses", "v1_models"]
|
||||
},
|
||||
anthropic: {
|
||||
upstreamUrl: "https://api.anthropic.com",
|
||||
authType: "x-api-key",
|
||||
capabilities: ["anthropic_messages", "anthropic_models"]
|
||||
capabilities: ["anthropic_messages", "v1_models"]
|
||||
},
|
||||
googleGemini: {
|
||||
upstreamUrl: "https://generativelanguage.googleapis.com",
|
||||
@@ -67,7 +67,7 @@ export const AI_PROVIDER_DEFAULTS: Record<
|
||||
"openai_chat",
|
||||
"openai_responses",
|
||||
"anthropic_messages",
|
||||
"anthropic_models"
|
||||
"v1_models"
|
||||
]
|
||||
},
|
||||
openRouter: {
|
||||
|
||||
Reference in New Issue
Block a user