tf_code/packages/opencode/src/provider/provider.ts

import z from "zod"
import path from "path"
import { Config } from "../config/config"
import { mergeDeep, sortBy } from "remeda"
import { NoSuchModelError, type LanguageModel, type Provider as SDK } from "ai"
import { Log } from "../util/log"
import { BunProc } from "../bun"
import { Plugin } from "../plugin"
import { ModelsDev } from "./models"
import { NamedError } from "../util/error"
import { Auth } from "../auth"
import { Instance } from "../project/instance"
import { Global } from "../global"
import { Flag } from "../flag/flag"

export namespace Provider {
  const log = Log.create({ service: "provider" })

  type CustomLoader = (provider: ModelsDev.Provider) => Promise<{
    autoload: boolean
    getModel?: (sdk: any, modelID: string, options?: Record<string, any>) => Promise<any>
    options?: Record<string, any>
  }>

  type Source = "env" | "config" | "custom" | "api"

  const CUSTOM_LOADERS: Record<string, CustomLoader> = {
    async anthropic() {
      return {
        autoload: false,
        options: {
          headers: {
            "anthropic-beta":
              "claude-code-20250219,interleaved-thinking-2025-05-14,fine-grained-tool-streaming-2025-05-14",
          },
        },
      }
    },
    async opencode(input) {
      const hasKey = await (async () => {
        if (input.env.some((item) => process.env[item])) return true
        if (await Auth.get(input.id)) return true
        return false
      })()

      if (!hasKey) {
        for (const [key, value] of Object.entries(input.models)) {
          if (value.cost.input === 0) continue
          delete input.models[key]
        }
      }

      return {
        autoload: Object.keys(input.models).length > 0,
        options: {},
      }
    },
    openai: async () => {
      return {
        autoload: false,
        async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
          return sdk.responses(modelID)
        },
        options: {},
      }
    },
    azure: async () => {
      return {
        autoload: false,
        async getModel(sdk: any, modelID: string, options?: Record<string, any>) {
          if (options?.["useCompletionUrls"]) {
            return sdk.chat(modelID)
          } else {
            return sdk.responses(modelID)
          }
        },
        options: {},
      }
    },
    "amazon-bedrock": async () => {
      if (
        !process.env["AWS_PROFILE"] &&
        !process.env["AWS_ACCESS_KEY_ID"] &&
        !process.env["AWS_BEARER_TOKEN_BEDROCK"]
      )
        return { autoload: false }

      const region = process.env["AWS_REGION"] ?? "us-east-1"

      const { fromNodeProviderChain } = await import(
        await BunProc.install("@aws-sdk/credential-providers")
      )
      return {
        autoload: true,
        options: {
          region,
          credentialProvider: fromNodeProviderChain(),
        },
        async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
          let regionPrefix = region.split("-")[0]

          switch (regionPrefix) {
            case "us": {
              const modelRequiresPrefix = [
                "nova-micro",
                "nova-lite",
                "nova-pro",
                "nova-premier",
                "claude",
                "deepseek",
              ].some((m) => modelID.includes(m))
              const isGovCloud = region.startsWith("us-gov")
              if (modelRequiresPrefix && !isGovCloud) {
                modelID = `${regionPrefix}.${modelID}`
              }
              break
            }
            case "eu": {
              const regionRequiresPrefix = [
                "eu-west-1",
                "eu-west-3",
                "eu-north-1",
                "eu-central-1",
                "eu-south-1",
                "eu-south-2",
              ].some((r) => region.includes(r))
              const modelRequiresPrefix = [
                "claude",
                "nova-lite",
                "nova-micro",
                "llama3",
                "pixtral",
              ].some((m) => modelID.includes(m))
              if (regionRequiresPrefix && modelRequiresPrefix) {
                modelID = `${regionPrefix}.${modelID}`
              }
              break
            }
            case "ap": {
              const isAustraliaRegion = ["ap-southeast-2", "ap-southeast-4"].includes(region)
              if (
                isAustraliaRegion &&
                ["anthropic.claude-sonnet-4-5", "anthropic.claude-haiku"].some((m) =>
                  modelID.includes(m),
                )
              ) {
                regionPrefix = "au"
                modelID = `${regionPrefix}.${modelID}`
              } else {
                const modelRequiresPrefix = ["claude", "nova-lite", "nova-micro", "nova-pro"].some(
                  (m) => modelID.includes(m),
                )
                if (modelRequiresPrefix) {
                  regionPrefix = "apac"
                  modelID = `${regionPrefix}.${modelID}`
                }
              }
              break
            }
          }

          return sdk.languageModel(modelID)
        },
      }
    },
    openrouter: async () => {
      return {
        autoload: false,
        options: {
          headers: {
            "HTTP-Referer": "https://opencode.ai/",
            "X-Title": "opencode",
          },
        },
      }
    },
    vercel: async () => {
      return {
        autoload: false,
        options: {
          headers: {
            "http-referer": "https://opencode.ai/",
            "x-title": "opencode",
          },
        },
      }
    },
    "google-vertex": async () => {
      const project =
        process.env["GOOGLE_CLOUD_PROJECT"] ??
        process.env["GCP_PROJECT"] ??
        process.env["GCLOUD_PROJECT"]
      const location =
        process.env["GOOGLE_CLOUD_LOCATION"] ?? process.env["VERTEX_LOCATION"] ?? "us-east5"
      const autoload = Boolean(project)
      if (!autoload) return { autoload: false }
      return {
        autoload: true,
        options: {
          project,
          location,
        },
        async getModel(sdk: any, modelID: string) {
          const id = String(modelID).trim()
          return sdk.languageModel(id)
        },
      }
    },
    "google-vertex-anthropic": async () => {
      const project =
        process.env["GOOGLE_CLOUD_PROJECT"] ??
        process.env["GCP_PROJECT"] ??
        process.env["GCLOUD_PROJECT"]
      const location =
        process.env["GOOGLE_CLOUD_LOCATION"] ?? process.env["VERTEX_LOCATION"] ?? "us-east5"
      const autoload = Boolean(project)
      if (!autoload) return { autoload: false }
      return {
        autoload: true,
        options: {
          project,
          location,
        },
        async getModel(sdk: any, modelID: string) {
          const id = String(modelID).trim()
          return sdk.languageModel(id)
        },
      }
    },
  }

  const state = Instance.state(async () => {
    const config = await Config.get()
    const database = await ModelsDev.get()

    const providers: {
      [providerID: string]: {
        source: Source
        info: ModelsDev.Provider
        getModel?: (sdk: any, modelID: string, options?: Record<string, any>) => Promise<any>
        options: Record<string, any>
      }
    } = {}
    const models = new Map<
      string,
      {
        providerID: string
        modelID: string
        info: ModelsDev.Model
        language: LanguageModel
        npm?: string
      }
    >()
    const sdk = new Map<number, SDK>()
    // Maps `${provider}/${key}` to the provider’s actual model ID for custom aliases.
    const realIdByKey = new Map<string, string>()

    log.info("init")

    function mergeProvider(
      id: string,
      options: Record<string, any>,
      source: Source,
      getModel?: (sdk: any, modelID: string, options?: Record<string, any>) => Promise<any>,
    ) {
      const provider = providers[id]
      if (!provider) {
        const info = database[id]
        if (!info) return
        if (info.api && !options["baseURL"]) options["baseURL"] = info.api
        providers[id] = {
          source,
          info,
          options,
          getModel,
        }
        return
      }
      provider.options = mergeDeep(provider.options, options)
      provider.source = source
      provider.getModel = getModel ?? provider.getModel
    }

    const configProviders = Object.entries(config.provider ?? {})

    for (const [providerID, provider] of configProviders) {
      const existing = database[providerID]
      const parsed: ModelsDev.Provider = {
        id: providerID,
        npm: provider.npm ?? existing?.npm,
        name: provider.name ?? existing?.name ?? providerID,
        env: provider.env ?? existing?.env ?? [],
        api: provider.api ?? existing?.api,
        models: existing?.models ?? {},
      }

      for (const [modelID, model] of Object.entries(provider.models ?? {})) {
        const existing = parsed.models[modelID]
        const parsedModel: ModelsDev.Model = {
          id: modelID,
          name: model.name ?? existing?.name ?? modelID,
          release_date: model.release_date ?? existing?.release_date,
          attachment: model.attachment ?? existing?.attachment ?? false,
          reasoning: model.reasoning ?? existing?.reasoning ?? false,
          temperature: model.temperature ?? existing?.temperature ?? false,
          tool_call: model.tool_call ?? existing?.tool_call ?? true,
          cost:
            !model.cost && !existing?.cost
              ? {
                  input: 0,
                  output: 0,
                  cache_read: 0,
                  cache_write: 0,
                }
              : {
                  cache_read: 0,
                  cache_write: 0,
                  ...existing?.cost,
                  ...model.cost,
                },
          options: {
            ...existing?.options,
            ...model.options,
          },
          limit: model.limit ??
            existing?.limit ?? {
              context: 0,
              output: 0,
            },
          modalities: model.modalities ??
            existing?.modalities ?? {
              input: ["text"],
              output: ["text"],
            },
          headers: model.headers,
          provider: model.provider ?? existing?.provider,
        }
        if (model.id && model.id !== modelID) {
          realIdByKey.set(`${providerID}/${modelID}`, model.id)
        }
        parsed.models[modelID] = parsedModel
      }
      database[providerID] = parsed
    }

    const disabled = await Config.get().then((cfg) => new Set(cfg.disabled_providers ?? []))
    // load env
    for (const [providerID, provider] of Object.entries(database)) {
      if (disabled.has(providerID)) continue
      const apiKey = provider.env.map((item) => process.env[item]).at(0)
      if (!apiKey) continue
      mergeProvider(
        providerID,
        // only include apiKey if there's only one potential option
        provider.env.length === 1 ? { apiKey } : {},
        "env",
      )
    }

    // load apikeys
    for (const [providerID, provider] of Object.entries(await Auth.all())) {
      if (disabled.has(providerID)) continue
      if (provider.type === "api") {
        mergeProvider(providerID, { apiKey: provider.key }, "api")
      }
    }

    // load custom
    for (const [providerID, fn] of Object.entries(CUSTOM_LOADERS)) {
      if (disabled.has(providerID)) continue
      const result = await fn(database[providerID])
      if (result && (result.autoload || providers[providerID])) {
        mergeProvider(providerID, result.options ?? {}, "custom", result.getModel)
      }
    }

    for (const plugin of await Plugin.list()) {
      if (!plugin.auth) continue
      const providerID = plugin.auth.provider
      if (disabled.has(providerID)) continue
      const auth = await Auth.get(providerID)
      if (!auth) continue
      if (!plugin.auth.loader) continue
      const options = await plugin.auth.loader(
        () => Auth.get(providerID) as any,
        database[plugin.auth.provider],
      )
      mergeProvider(plugin.auth.provider, options ?? {}, "custom")
    }

    // load config
    for (const [providerID, provider] of configProviders) {
      mergeProvider(providerID, provider.options ?? {}, "config")
    }

    for (const [providerID, provider] of Object.entries(providers)) {
      const filteredModels = Object.fromEntries(
        Object.entries(provider.info.models)
          // Filter out blacklisted models
          .filter(
            ([modelID]) =>
              modelID !== "gpt-5-chat-latest" &&
              !(providerID === "openrouter" && modelID === "openai/gpt-5-chat"),
          )
          // Filter out experimental models
          .filter(
            ([, model]) =>
              ((!model.experimental && model.status !== "alpha") ||
                Flag.OPENCODE_ENABLE_EXPERIMENTAL_MODELS) &&
              model.status !== "deprecated",
          ),
      )
      provider.info.models = filteredModels

      if (Object.keys(provider.info.models).length === 0) {
        delete providers[providerID]
        continue
      }
      log.info("found", { providerID })
    }

    return {
      models,
      providers,
      sdk,
      realIdByKey,
    }
  })

  export async function list() {
    return state().then((state) => state.providers)
  }

  async function getSDK(provider: ModelsDev.Provider, model: ModelsDev.Model) {
    return (async () => {
      using _ = log.time("getSDK", {
        providerID: provider.id,
      })
      const s = await state()
      const pkg = model.provider?.npm ?? provider.npm ?? provider.id
      const options = { ...s.providers[provider.id]?.options }
      if (pkg.includes("@ai-sdk/openai-compatible") && options["includeUsage"] === undefined) {
        options["includeUsage"] = true
      }
      const key = Bun.hash.xxHash32(JSON.stringify({ pkg, options }))
      const existing = s.sdk.get(key)
      if (existing) return existing
      const installedPath = await BunProc.install(pkg, "latest")
      // The `google-vertex-anthropic` provider points to the `@ai-sdk/google-vertex` package.
      // Ref: https://github.com/sst/models.dev/blob/0a87de42ab177bebad0620a889e2eb2b4a5dd4ab/providers/google-vertex-anthropic/provider.toml
      // However, the actual export is at the subpath `@ai-sdk/google-vertex/anthropic`.
      // Ref: https://ai-sdk.dev/providers/ai-sdk-providers/google-vertex#google-vertex-anthropic-provider-usage
      // In addition, Bun's dynamic import logic does not support subpath imports,
      // so we patch the import path to load directly from `dist`.
      const modPath =
        provider.id === "google-vertex-anthropic"
          ? `${installedPath}/dist/anthropic/index.mjs`
          : installedPath
      const mod = await import(modPath)
      if (options["timeout"] !== undefined && options["timeout"] !== null) {
        // Only override fetch if user explicitly sets timeout
        options["fetch"] = async (input: any, init?: BunFetchRequestInit) => {
          const { signal, ...rest } = init ?? {}

          const signals: AbortSignal[] = []
          if (signal) signals.push(signal)
          if (options["timeout"] !== false) signals.push(AbortSignal.timeout(options["timeout"]))

          const combined = signals.length > 1 ? AbortSignal.any(signals) : signals[0]

          return fetch(input, {
            ...rest,
            signal: combined,
            // @ts-ignore see here: https://github.com/oven-sh/bun/issues/16682
            timeout: false,
          })
        }
      }
      const fn = mod[Object.keys(mod).find((key) => key.startsWith("create"))!]
      const loaded = fn({
        name: provider.id,
        ...options,
      })
      s.sdk.set(key, loaded)
      return loaded as SDK
    })().catch((e) => {
      throw new InitError({ providerID: provider.id }, { cause: e })
    })
  }

  export async function getProvider(providerID: string) {
    return state().then((s) => s.providers[providerID])
  }

  export async function getModel(providerID: string, modelID: string) {
    const key = `${providerID}/${modelID}`
    const s = await state()
    if (s.models.has(key)) return s.models.get(key)!

    log.info("getModel", {
      providerID,
      modelID,
    })

    const provider = s.providers[providerID]
    if (!provider) throw new ModelNotFoundError({ providerID, modelID })
    const info = provider.info.models[modelID]
    if (!info) throw new ModelNotFoundError({ providerID, modelID })
    const sdk = await getSDK(provider.info, info)

    try {
      const keyReal = `${providerID}/${modelID}`
      const realID = s.realIdByKey.get(keyReal) ?? info.id
      const language = provider.getModel
        ? await provider.getModel(sdk, realID, provider.options)
        : sdk.languageModel(realID)
      log.info("found", { providerID, modelID })
      s.models.set(key, {
        providerID,
        modelID,
        info,
        language,
        npm: info.provider?.npm ?? provider.info.npm,
      })
      return {
        modelID,
        providerID,
        info,
        language,
        npm: info.provider?.npm ?? provider.info.npm,
      }
    } catch (e) {
      if (e instanceof NoSuchModelError)
        throw new ModelNotFoundError(
          {
            modelID: modelID,
            providerID,
          },
          { cause: e },
        )
      throw e
    }
  }

  export async function getSmallModel(providerID: string) {
    const cfg = await Config.get()

    if (cfg.small_model) {
      const parsed = parseModel(cfg.small_model)
      return getModel(parsed.providerID, parsed.modelID)
    }

    const provider = await state().then((state) => state.providers[providerID])
    if (!provider) return
    let priority = [
      "claude-haiku-4-5",
      "claude-haiku-4.5",
      "3-5-haiku",
      "3.5-haiku",
      "gemini-2.5-flash",
      "gpt-5-nano",
    ]
    // claude-haiku-4.5 is considered a premium model in github copilot, we shouldn't use premium requests for title gen
    if (providerID === "github-copilot") {
      priority = priority.filter((m) => m !== "claude-haiku-4.5")
    }
    for (const item of priority) {
      for (const model of Object.keys(provider.info.models)) {
        if (model.includes(item)) return getModel(providerID, model)
      }
    }
  }

  const priority = ["gemini-2.5-pro-preview", "gpt-5", "claude-sonnet-4"]
  export function sort(models: ModelsDev.Model[]) {
    return sortBy(
      models,
      [(model) => priority.findIndex((filter) => model.id.includes(filter)), "desc"],
      [(model) => (model.id.includes("latest") ? 0 : 1), "asc"],
      [(model) => model.id, "desc"],
    )
  }

  export async function defaultModel() {
    const cfg = await Config.get()
    if (cfg.model) return parseModel(cfg.model)

    // this will be adjusted when migration to opentui is complete,
    // for now we just read the tui state toml file directly
    //
    // NOTE: cannot just import file as toml without cleaning due to lack of
    // support for date/time references in Bun toml parser: https://github.com/oven-sh/bun/issues/22426
    const lastused = await Bun.file(path.join(Global.Path.state, "tui"))
      .text()
      .then((text) => {
        // remove the date/time references since Bun toml parser doesn't support yet
        const cleaned = text
          .split("\n")
          .filter((line) => !line.trim().startsWith("last_used ="))
          .join("\n")
        const state = Bun.TOML.parse(cleaned) as {
          recently_used_models?: {
            provider_id: string
            model_id: string
          }[]
        }
        const [model] = state?.recently_used_models ?? []
        if (model) {
          return {
            providerID: model.provider_id,
            modelID: model.model_id,
          }
        }
      })
      .catch((error) => {
        log.error("failed to find last used model", {
          error,
        })
        return undefined
      })

    if (lastused) return lastused

    const provider = await list()
      .then((val) => Object.values(val))
      .then((x) => x.find((p) => !cfg.provider || Object.keys(cfg.provider).includes(p.info.id)))
    if (!provider) throw new Error("no providers found")
    const [model] = sort(Object.values(provider.info.models))
    if (!model) throw new Error("no models found")
    return {
      providerID: provider.info.id,
      modelID: model.id,
    }
  }

  export function parseModel(model: string) {
    const [providerID, ...rest] = model.split("/")
    return {
      providerID: providerID,
      modelID: rest.join("/"),
    }
  }

  export const ModelNotFoundError = NamedError.create(
    "ProviderModelNotFoundError",
    z.object({
      providerID: z.string(),
      modelID: z.string(),
    }),
  )

  export const InitError = NamedError.create(
    "ProviderInitError",
    z.object({
      providerID: z.string(),
    }),
  )
}