(feat): Add Baseten Provider (#9461)

Co-authored-by: roomote[bot] <219738659+roomote[bot]@users.noreply.github.com>
Co-authored-by: AlexKer <[email protected]>
Co-authored-by: Matt Rubens <[email protected]>

Author: 3 people

packages/types/src/global-settings.ts

@@ -246,6 +246,7 @@ export const SECRET_STATE_KEYS = [
 	"featherlessApiKey",
 	"ioIntelligenceApiKey",
 	"vercelAiGatewayApiKey",
+	"basetenApiKey",
 ] as const
 
 // Global secrets that are part of GlobalSettings (not ProviderSettings)

packages/types/src/provider-settings.ts

@@ -4,6 +4,7 @@ import { modelInfoSchema, reasoningEffortSettingSchema, verbosityLevelsSchema, s
 import { codebaseIndexProviderSchema } from "./codebase-index.js"
 import {
 	anthropicModels,
+	basetenModels,
 	bedrockModels,
 	cerebrasModels,
 	claudeCodeModels,
@@ -120,6 +121,7 @@ export const providerNames = [
 	...fauxProviders,
 	"anthropic",
 	"bedrock",
+	"baseten",
 	"cerebras",
 	"claude-code",
 	"doubao",
@@ -424,6 +426,10 @@ const vercelAiGatewaySchema = baseProviderSettingsSchema.extend({
 	vercelAiGatewayModelId: z.string().optional(),
 })
 
+const basetenSchema = apiModelIdProviderModelSchema.extend({
+	basetenApiKey: z.string().optional(),
+})
+
 const defaultSchema = z.object({
 	apiProvider: z.undefined(),
 })
@@ -454,6 +460,7 @@ export const providerSettingsSchemaDiscriminated = z.discriminatedUnion("apiProv
 	fakeAiSchema.merge(z.object({ apiProvider: z.literal("fake-ai") })),
 	xaiSchema.merge(z.object({ apiProvider: z.literal("xai") })),
 	groqSchema.merge(z.object({ apiProvider: z.literal("groq") })),
+	basetenSchema.merge(z.object({ apiProvider: z.literal("baseten") })),
 	huggingFaceSchema.merge(z.object({ apiProvider: z.literal("huggingface") })),
 	chutesSchema.merge(z.object({ apiProvider: z.literal("chutes") })),
 	litellmSchema.merge(z.object({ apiProvider: z.literal("litellm") })),
@@ -496,6 +503,7 @@ export const providerSettingsSchema = z.object({
 	...fakeAiSchema.shape,
 	...xaiSchema.shape,
 	...groqSchema.shape,
+	...basetenSchema.shape,
 	...huggingFaceSchema.shape,
 	...chutesSchema.shape,
 	...litellmSchema.shape,
@@ -583,6 +591,7 @@ export const modelIdKeysByProvider: Record<TypicalProvider, ModelIdKey> = {
 	requesty: "requestyModelId",
 	xai: "apiModelId",
 	groq: "apiModelId",
+	baseten: "apiModelId",
 	chutes: "apiModelId",
 	litellm: "litellmModelId",
 	huggingface: "huggingFaceModelId",
@@ -715,6 +724,7 @@ export const MODELS_BY_PROVIDER: Record<
 	},
 	xai: { id: "xai", label: "xAI (Grok)", models: Object.keys(xaiModels) },
 	zai: { id: "zai", label: "Zai", models: Object.keys(internationalZAiModels) },
+	baseten: { id: "baseten", label: "BaseTen", models: Object.keys(basetenModels) },
 
 	// Dynamic providers; models pulled from remote APIs.
 	glama: { id: "glama", label: "Glama", models: [] },

packages/types/src/providers/baseten.ts

@@ -0,0 +1,127 @@
+import type { ModelInfo } from "../model.js"
+
+// Baseten
+// https://baseten.co/products/model-apis/
+
+export const basetenModels = {
+	"moonshotai/Kimi-K2-Thinking": {
+		maxTokens: 163_800,
+		contextWindow: 262_000,
+		supportsImages: false,
+		supportsPromptCache: false,
+		supportsNativeTools: true,
+		defaultToolProtocol: "native",
+		inputPrice: 0.6,
+		outputPrice: 2.5,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0,
+		description: "Kimi K2 Thinking - A model with enhanced reasoning capabilities from Kimi K2",
+	},
+	"zai-org/GLM-4.6": {
+		maxTokens: 200_000,
+		contextWindow: 200_000,
+		supportsImages: false,
+		supportsPromptCache: false,
+		supportsNativeTools: true,
+		inputPrice: 0.6,
+		outputPrice: 2.2,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0,
+		description: "Frontier open model with advanced agentic, reasoning and coding capabilities",
+	},
+	"deepseek-ai/DeepSeek-R1": {
+		maxTokens: 131_072,
+		contextWindow: 163_840,
+		supportsImages: false,
+		supportsPromptCache: false,
+		inputPrice: 2.55,
+		outputPrice: 5.95,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0,
+		description: "DeepSeek's first-generation reasoning model",
+	},
+	"deepseek-ai/DeepSeek-R1-0528": {
+		maxTokens: 131_072,
+		contextWindow: 163_840,
+		supportsImages: false,
+		supportsPromptCache: false,
+		inputPrice: 2.55,
+		outputPrice: 5.95,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0,
+		description: "The latest revision of DeepSeek's first-generation reasoning model",
+	},
+	"deepseek-ai/DeepSeek-V3-0324": {
+		maxTokens: 131_072,
+		contextWindow: 163_840,
+		supportsImages: false,
+		supportsPromptCache: false,
+		inputPrice: 0.77,
+		outputPrice: 0.77,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0,
+		description: "Fast general-purpose LLM with enhanced reasoning capabilities",
+	},
+	"deepseek-ai/DeepSeek-V3.1": {
+		maxTokens: 131_072,
+		contextWindow: 163_840,
+		supportsImages: false,
+		supportsPromptCache: false,
+		inputPrice: 0.5,
+		outputPrice: 1.5,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0,
+		description:
+			"Extremely capable general-purpose LLM with hybrid reasoning capabilities and advanced tool calling",
+	},
+	"Qwen/Qwen3-235B-A22B-Instruct-2507": {
+		maxTokens: 262_144,
+		contextWindow: 262_144,
+		supportsImages: false,
+		supportsPromptCache: false,
+		inputPrice: 0.22,
+		outputPrice: 0.8,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0,
+		description: "Mixture-of-experts LLM with math and reasoning capabilities",
+	},
+	"Qwen/Qwen3-Coder-480B-A35B-Instruct": {
+		maxTokens: 262_144,
+		contextWindow: 262_144,
+		supportsImages: false,
+		supportsPromptCache: false,
+		inputPrice: 0.38,
+		outputPrice: 1.53,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0,
+		description: "Mixture-of-experts LLM with advanced coding and reasoning capabilities",
+	},
+	"openai/gpt-oss-120b": {
+		maxTokens: 128_072,
+		contextWindow: 128_072,
+		supportsImages: false,
+		supportsPromptCache: false,
+		supportsNativeTools: true,
+		inputPrice: 0.1,
+		outputPrice: 0.5,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0,
+		description: "Extremely capable general-purpose LLM with strong, controllable reasoning capabilities",
+	},
+	"moonshotai/Kimi-K2-Instruct-0905": {
+		maxTokens: 168_000,
+		contextWindow: 262_000,
+		supportsImages: false,
+		supportsPromptCache: false,
+		supportsNativeTools: true,
+		inputPrice: 0.6,
+		outputPrice: 2.5,
+		cacheWritesPrice: 0,
+		cacheReadsPrice: 0,
+		description: "State of the art language model for agentic and coding tasks. September Update.",
+	},
+} as const satisfies Record<string, ModelInfo>
+
+export type BasetenModelId = keyof typeof basetenModels
+
+export const basetenDefaultModelId = "moonshotai/Kimi-K2-Thinking" satisfies BasetenModelId

packages/types/src/providers/index.ts

@@ -1,4 +1,5 @@
 export * from "./anthropic.js"
+export * from "./baseten.js"
 export * from "./bedrock.js"
 export * from "./cerebras.js"
 export * from "./chutes.js"
@@ -33,6 +34,7 @@ export * from "./deepinfra.js"
 export * from "./minimax.js"
 
 import { anthropicDefaultModelId } from "./anthropic.js"
+import { basetenDefaultModelId } from "./baseten.js"
 import { bedrockDefaultModelId } from "./bedrock.js"
 import { cerebrasDefaultModelId } from "./cerebras.js"
 import { chutesDefaultModelId } from "./chutes.js"
@@ -93,6 +95,8 @@ export function getProviderDefaultModelId(
 			return "meta-llama/Llama-3.3-70B-Instruct"
 		case "chutes":
 			return chutesDefaultModelId
+		case "baseten":
+			return basetenDefaultModelId
 		case "bedrock":
 			return bedrockDefaultModelId
 		case "vertex":

src/api/index.ts

@@ -42,6 +42,7 @@ import {
 	VercelAiGatewayHandler,
 	DeepInfraHandler,
 	MiniMaxHandler,
+	BasetenHandler,
 } from "./providers"
 import { NativeOllamaHandler } from "./providers/native-ollama"
 
@@ -190,6 +191,8 @@ export function buildApiHandler(configuration: ProviderSettings): ApiHandler {
 			return new VercelAiGatewayHandler(options)
 		case "minimax":
 			return new MiniMaxHandler(options)
+		case "baseten":
+			return new BasetenHandler(options)
 		default:
 			apiProvider satisfies "gemini-cli" | undefined
 			return new AnthropicHandler(options)

src/api/providers/baseten.ts

@@ -0,0 +1,18 @@
+import { type BasetenModelId, basetenDefaultModelId, basetenModels } from "@roo-code/types"
+
+import type { ApiHandlerOptions } from "../../shared/api"
+import { BaseOpenAiCompatibleProvider } from "./base-openai-compatible-provider"
+
+export class BasetenHandler extends BaseOpenAiCompatibleProvider<BasetenModelId> {
+	constructor(options: ApiHandlerOptions) {
+		super({
+			...options,
+			providerName: "Baseten",
+			baseURL: "https://inference.baseten.co/v1",
+			apiKey: options.basetenApiKey,
+			defaultProviderModelId: basetenDefaultModelId,
+			providerModels: basetenModels,
+			defaultTemperature: 0.5,
+		})
+	}
+}

src/api/providers/index.ts

@@ -35,3 +35,4 @@ export { FeatherlessHandler } from "./featherless"
 export { VercelAiGatewayHandler } from "./vercel-ai-gateway"
 export { DeepInfraHandler } from "./deepinfra"
 export { MiniMaxHandler } from "./minimax"
+export { BasetenHandler } from "./baseten"

webview-ui/src/components/settings/ApiOptions.tsx

@@ -26,6 +26,7 @@ import {
 	groqDefaultModelId,
 	cerebrasDefaultModelId,
 	chutesDefaultModelId,
+	basetenDefaultModelId,
 	bedrockDefaultModelId,
 	vertexDefaultModelId,
 	sambaNovaDefaultModelId,
@@ -67,6 +68,7 @@ import {
 
 import {
 	Anthropic,
+	Baseten,
 	Bedrock,
 	Cerebras,
 	Chutes,
@@ -274,6 +276,7 @@ const ApiOptions = ({
 
 	const selectedProviderModels = useMemo(() => {
 		const models = MODELS_BY_PROVIDER[selectedProvider]
+
 		if (!models) return []
 
 		const filteredModels = filterModels(models, selectedProvider, organizationAllowList)
@@ -355,6 +358,7 @@ const ApiOptions = ({
 				xai: { field: "apiModelId", default: xaiDefaultModelId },
 				groq: { field: "apiModelId", default: groqDefaultModelId },
 				chutes: { field: "apiModelId", default: chutesDefaultModelId },
+				baseten: { field: "apiModelId", default: basetenDefaultModelId },
 				bedrock: { field: "apiModelId", default: bedrockDefaultModelId },
 				vertex: { field: "apiModelId", default: vertexDefaultModelId },
 				sambanova: { field: "apiModelId", default: sambaNovaDefaultModelId },
@@ -564,6 +568,10 @@ const ApiOptions = ({
 				<Mistral apiConfiguration={apiConfiguration} setApiConfigurationField={setApiConfigurationField} />
 			)}
 
+			{selectedProvider === "baseten" && (
+				<Baseten apiConfiguration={apiConfiguration} setApiConfigurationField={setApiConfigurationField} />
+			)}
+
 			{selectedProvider === "bedrock" && (
 				<Bedrock
 					apiConfiguration={apiConfiguration}

webview-ui/src/components/settings/constants.ts

@@ -20,6 +20,7 @@ import {
 	fireworksModels,
 	featherlessModels,
 	minimaxModels,
+	basetenModels,
 } from "@roo-code/types"
 
 export const MODELS_BY_PROVIDER: Partial<Record<ProviderName, Record<string, ModelInfo>>> = {
@@ -42,6 +43,7 @@ export const MODELS_BY_PROVIDER: Partial<Record<ProviderName, Record<string, Mod
 	fireworks: fireworksModels,
 	featherless: featherlessModels,
 	minimax: minimaxModels,
+	baseten: basetenModels,
 }
 
 export const PROVIDERS = [
@@ -80,4 +82,5 @@ export const PROVIDERS = [
 	{ value: "roo", label: "Roo Code Cloud" },
 	{ value: "vercel-ai-gateway", label: "Vercel AI Gateway" },
 	{ value: "minimax", label: "MiniMax" },
+	{ value: "baseten", label: "Baseten" },
 ].sort((a, b) => a.label.localeCompare(b.label))

webview-ui/src/components/settings/providers/Baseten.tsx

@@ -0,0 +1,50 @@
+import { useCallback } from "react"
+import { VSCodeTextField } from "@vscode/webview-ui-toolkit/react"
+
+import type { ProviderSettings } from "@roo-code/types"
+
+import { useAppTranslation } from "@src/i18n/TranslationContext"
+import { VSCodeButtonLink } from "@src/components/common/VSCodeButtonLink"
+
+import { inputEventTransform } from "../transforms"
+
+type BasetenProps = {
+	apiConfiguration: ProviderSettings
+	setApiConfigurationField: (field: keyof ProviderSettings, value: ProviderSettings[keyof ProviderSettings]) => void
+}
+
+export const Baseten = ({ apiConfiguration, setApiConfigurationField }: BasetenProps) => {
+	const { t } = useAppTranslation()
+
+	const handleInputChange = useCallback(
+		<K extends keyof ProviderSettings, E>(
+			field: K,
+			transform: (event: E) => ProviderSettings[K] = inputEventTransform,
+		) =>
+			(event: E | Event) => {
+				setApiConfigurationField(field, transform(event as E))
+			},
+		[setApiConfigurationField],
+	)
+
+	return (
+		<>
+			<VSCodeTextField
+				value={apiConfiguration?.basetenApiKey || ""}
+				type="password"
+				onInput={handleInputChange("basetenApiKey")}
+				placeholder={t("settings:placeholders.apiKey")}
+				className="w-full">
+				<label className="block font-medium mb-1">{t("settings:providers.basetenApiKey")}</label>
+			</VSCodeTextField>
+			<div className="text-sm text-vscode-descriptionForeground -mt-2">
+				{t("settings:providers.apiKeyStorageNotice")}
+			</div>
+			{!apiConfiguration?.basetenApiKey && (
+				<VSCodeButtonLink href="https://app.baseten.co/settings/api_keys" appearance="secondary">
+					{t("settings:providers.getBasetenApiKey")}
+				</VSCodeButtonLink>
+			)}
+		</>
+	)
+}