feat: add Avian as LLM provider

Add Avian (api.avian.io) as a new OpenAI-compatible inference provider with
four models: DeepSeek V3.2, Kimi K2.5, GLM-5, and MiniMax M2.5.

Changes:
- New provider module (web/src/llm-api/avian.ts) with streaming and
  non-streaming support, per-model pricing, usage tracking, and billing
- Route avian/* models through the Avian provider in chat completions API
- Add AVIAN_API_KEY to server env schema
- Register avian models in model-config constants and agent type definitions

Author: avianion

.agents/types/agent-definition.ts

@@ -418,6 +418,12 @@ export type ModelName =
   | 'deepseek/deepseek-r1-0528'
   | 'deepseek/deepseek-r1-0528:nitro'
 
+  // Avian (OpenAI-compatible inference API at api.avian.io)
+  | 'avian/deepseek-v3.2'
+  | 'avian/kimi-k2.5'
+  | 'avian/glm-5'
+  | 'avian/minimax-m2.5'
+
   // Other open source models
   | 'moonshotai/kimi-k2'
   | 'moonshotai/kimi-k2:nitro'

agents/types/agent-definition.ts

@@ -418,6 +418,12 @@ export type ModelName =
   | 'deepseek/deepseek-r1-0528'
   | 'deepseek/deepseek-r1-0528:nitro'
 
+  // Avian (OpenAI-compatible inference API at api.avian.io)
+  | 'avian/deepseek-v3.2'
+  | 'avian/kimi-k2.5'
+  | 'avian/glm-5'
+  | 'avian/minimax-m2.5'
+
   // Other open source models
   | 'moonshotai/kimi-k2'
   | 'moonshotai/kimi-k2:nitro'

common/src/constants/model-config.ts

@@ -3,6 +3,7 @@ import { isExplicitlyDefinedModel } from '../util/model-utils'
 // Allowed model prefixes for validation
 export const ALLOWED_MODEL_PREFIXES = [
   'anthropic',
+  'avian',
   'openai',
   'google',
   'x-ai',
@@ -51,6 +52,14 @@ export const openrouterModels = {
 export type openrouterModel =
   (typeof openrouterModels)[keyof typeof openrouterModels]
 
+export const avianModels = {
+  avian_deepseek_v3_2: 'avian/deepseek-v3.2',
+  avian_kimi_k2_5: 'avian/kimi-k2.5',
+  avian_glm_5: 'avian/glm-5',
+  avian_minimax_m2_5: 'avian/minimax-m2.5',
+} as const
+export type AvianModel = (typeof avianModels)[keyof typeof avianModels]
+
 export const deepseekModels = {
   deepseekChat: 'deepseek-chat',
   deepseekReasoner: 'deepseek-reasoner',
@@ -86,6 +95,7 @@ export type FinetunedVertexModel =
 
 export const models = {
   ...openaiModels,
+  ...avianModels,
   ...deepseekModels,
   ...openrouterModels,
   ...finetunedVertexModels,
@@ -114,6 +124,12 @@ export const providerModelNames = {
       'openai' as const,
     ]),
   ),
+  ...Object.fromEntries(
+    Object.entries(avianModels).map(([name, model]) => [
+      model,
+      'avian' as const,
+    ]),
+  ),
   ...Object.fromEntries(
     Object.entries(openrouterModels).map(([name, model]) => [
       model,
@@ -166,6 +182,7 @@ export function getModelFromShortName(
 }
 
 export const providerDomains = {
+  avian: 'avian.io',
   google: 'google.com',
   anthropic: 'anthropic.com',
   openai: 'chatgpt.com',
@@ -178,6 +195,8 @@ export function getLogoForModel(modelName: string): string | undefined {
 
   if (Object.values(openaiModels).includes(modelName as OpenAIModel))
     domain = providerDomains.openai
+  else if (Object.values(avianModels).includes(modelName as AvianModel))
+    domain = providerDomains.avian
   else if (Object.values(deepseekModels).includes(modelName as DeepseekModel))
     domain = providerDomains.deepseek
   else if (modelName.includes('claude')) domain = providerDomains.anthropic

common/src/templates/initial-agents-dir/types/agent-definition.ts

@@ -418,6 +418,12 @@ export type ModelName =
   | 'deepseek/deepseek-r1-0528'
   | 'deepseek/deepseek-r1-0528:nitro'
 
+  // Avian (OpenAI-compatible inference API at api.avian.io)
+  | 'avian/deepseek-v3.2'
+  | 'avian/kimi-k2.5'
+  | 'avian/glm-5'
+  | 'avian/minimax-m2.5'
+
   // Other open source models
   | 'moonshotai/kimi-k2'
   | 'moonshotai/kimi-k2:nitro'

packages/internal/src/env-schema.ts

@@ -6,6 +6,7 @@ export const serverEnvSchema = clientEnvSchema.extend({
   OPEN_ROUTER_API_KEY: z.string().min(1),
   OPENAI_API_KEY: z.string().min(1),
   ANTHROPIC_API_KEY: z.string().min(1),
+  AVIAN_API_KEY: z.string().min(1).optional(),
   FIREWORKS_API_KEY: z.string().min(1),
   CANOPYWAVE_API_KEY: z.string().min(1).optional(),
   SILICONFLOW_API_KEY: z.string().min(1).optional(),
@@ -53,6 +54,7 @@ export const serverProcessEnv: ServerInput = {
   OPEN_ROUTER_API_KEY: process.env.OPEN_ROUTER_API_KEY,
   OPENAI_API_KEY: process.env.OPENAI_API_KEY,
   ANTHROPIC_API_KEY: process.env.ANTHROPIC_API_KEY,
+  AVIAN_API_KEY: process.env.AVIAN_API_KEY,
   FIREWORKS_API_KEY: process.env.FIREWORKS_API_KEY,
   CANOPYWAVE_API_KEY: process.env.CANOPYWAVE_API_KEY,
   SILICONFLOW_API_KEY: process.env.SILICONFLOW_API_KEY,

web/src/app/api/v1/chat/completions/_post.ts

@@ -35,6 +35,12 @@ import type { NextRequest } from 'next/server'
 
 import type { ChatCompletionRequestBody } from '@/llm-api/types'
 
+import {
+  AvianError,
+  handleAvianNonStream,
+  handleAvianStream,
+  isAvianModel,
+} from '@/llm-api/avian'
 import {
   CanopyWaveError,
   handleCanopyWaveNonStream,
@@ -469,11 +475,12 @@ export async function postChatCompletions(params: {
     // Handle streaming vs non-streaming
     try {
       if (bodyStream) {
-        // Streaming request — route to SiliconFlow/CanopyWave/Fireworks for supported models
+        // Streaming request — route to provider for supported models
         const useSiliconFlow = false // isSiliconFlowModel(typedBody.model)
         const useCanopyWave = false // isCanopyWaveModel(typedBody.model)
-        const useFireworks = isFireworksModel(typedBody.model)
-        const useOpenAIDirect = !useFireworks && isOpenAIDirectModel(typedBody.model)
+        const useAvian = isAvianModel(typedBody.model)
+        const useFireworks = !useAvian && isFireworksModel(typedBody.model)
+        const useOpenAIDirect = !useAvian && !useFireworks && isOpenAIDirectModel(typedBody.model)
         const stream = useSiliconFlow
           ? await handleSiliconFlowStream({
               body: typedBody,
@@ -494,6 +501,16 @@ export async function postChatCompletions(params: {
               logger,
               insertMessageBigquery,
             })
+          : useAvian
+          ? await handleAvianStream({
+              body: typedBody,
+              userId,
+              stripeCustomerId,
+              agentId,
+              fetch,
+              logger,
+              insertMessageBigquery,
+            })
           : useFireworks
           ? await handleFireworksStream({
               body: typedBody,
@@ -544,13 +561,14 @@ export async function postChatCompletions(params: {
           },
         })
       } else {
-        // Non-streaming request — route to SiliconFlow/CanopyWave/Fireworks for supported models
+        // Non-streaming request — route to provider for supported models
         // TEMPORARILY DISABLED: route through OpenRouter
         const model = typedBody.model
         const useSiliconFlow = false // isSiliconFlowModel(model)
         const useCanopyWave = false // isCanopyWaveModel(model)
-        const useFireworks = isFireworksModel(model)
-        const shouldUseOpenAIEndpoint = !useFireworks && isOpenAIDirectModel(model)
+        const useAvianNonStream = isAvianModel(model)
+        const useFireworks = !useAvianNonStream && isFireworksModel(model)
+        const shouldUseOpenAIEndpoint = !useAvianNonStream && !useFireworks && isOpenAIDirectModel(model)
 
         const nonStreamRequest = useSiliconFlow
           ? handleSiliconFlowNonStream({
@@ -572,6 +590,16 @@ export async function postChatCompletions(params: {
               logger,
               insertMessageBigquery,
             })
+          : useAvianNonStream
+          ? handleAvianNonStream({
+              body: typedBody,
+              userId,
+              stripeCustomerId,
+              agentId,
+              fetch,
+              logger,
+              insertMessageBigquery,
+            })
           : useFireworks
           ? handleFireworksNonStream({
               body: typedBody,
@@ -622,6 +650,10 @@ export async function postChatCompletions(params: {
       if (error instanceof OpenRouterError) {
         openrouterError = error
       }
+      let avianError: AvianError | undefined
+      if (error instanceof AvianError) {
+        avianError = error
+      }
       let fireworksError: FireworksError | undefined
       if (error instanceof FireworksError) {
         fireworksError = error
@@ -641,7 +673,7 @@ export async function postChatCompletions(params: {
 
       // Log detailed error information for debugging
       const errorDetails = openrouterError?.toJSON()
-      const providerLabel = siliconflowError ? 'SiliconFlow' : canopywaveError ? 'CanopyWave' : fireworksError ? 'Fireworks' : openaiError ? 'OpenAI' : 'OpenRouter'
+      const providerLabel = avianError ? 'Avian' : siliconflowError ? 'SiliconFlow' : canopywaveError ? 'CanopyWave' : fireworksError ? 'Fireworks' : openaiError ? 'OpenAI' : 'OpenRouter'
       logger.error(
         {
           error: getErrorObject(error),
@@ -655,8 +687,8 @@ export async function postChatCompletions(params: {
             ? typedBody.messages.length
             : 0,
           messages: typedBody.messages,
-          providerStatusCode: (openrouterError ?? fireworksError ?? canopywaveError ?? siliconflowError ?? openaiError)?.statusCode,
-          providerStatusText: (openrouterError ?? fireworksError ?? canopywaveError ?? siliconflowError ?? openaiError)?.statusText,
+          providerStatusCode: (openrouterError ?? avianError ?? fireworksError ?? canopywaveError ?? siliconflowError ?? openaiError)?.statusCode,
+          providerStatusText: (openrouterError ?? avianError ?? fireworksError ?? canopywaveError ?? siliconflowError ?? openaiError)?.statusText,
           openrouterErrorCode: errorDetails?.error?.code,
           openrouterErrorType: errorDetails?.error?.type,
           openrouterErrorMessage: errorDetails?.error?.message,
@@ -681,6 +713,9 @@ export async function postChatCompletions(params: {
       if (error instanceof OpenRouterError) {
         return NextResponse.json(error.toJSON(), { status: error.statusCode })
       }
+      if (error instanceof AvianError) {
+        return NextResponse.json(error.toJSON(), { status: error.statusCode })
+      }
       if (error instanceof FireworksError) {
         return NextResponse.json(error.toJSON(), { status: error.statusCode })
       }