diff --git a/packages/opencode/src/cli/cmd/models.ts b/packages/opencode/src/cli/cmd/models.ts
index 156dae91c67..4d950b10157 100644
--- a/packages/opencode/src/cli/cmd/models.ts
+++ b/packages/opencode/src/cli/cmd/models.ts
@@ -24,6 +24,10 @@ export const ModelsCommand = cmd({
         describe: "refresh the models cache from models.dev",
         type: "boolean",
       })
+      .option("local", {
+        describe: "show only locally detected models (e.g. Ollama)",
+        type: "boolean",
+      })
   },
   handler: async (args) => {
     if (args.refresh) {
@@ -60,7 +64,7 @@ export const ModelsCommand = cmd({
           return
         }
 
-        const providerIDs = Object.keys(providers).sort((a, b) => {
+        let providerIDs = Object.keys(providers).sort((a, b) => {
           const aIsOpencode = a.startsWith("opencode")
           const bIsOpencode = b.startsWith("opencode")
           if (aIsOpencode && !bIsOpencode) return -1
@@ -68,6 +72,17 @@ export const ModelsCommand = cmd({
           return a.localeCompare(b)
         })
 
+        if (args.local) {
+          const localProviders = Object.entries(providers)
+            .filter(([, p]) => p.source === "api")
+            .map(([id]) => id)
+          providerIDs = providerIDs.filter((id) => localProviders.includes(id))
+          if (providerIDs.length === 0) {
+            UI.println(UI.Style.TEXT_WARNING + "No local models detected. Make sure Ollama is running." + UI.Style.TEXT_NORMAL)
+            return
+          }
+        }
+
         for (const providerID of providerIDs) {
           printModels(providerID, args.verbose)
         }
diff --git a/packages/opencode/src/provider/ollama.ts b/packages/opencode/src/provider/ollama.ts
new file mode 100644
index 00000000000..4c20f1dc388
--- /dev/null
+++ b/packages/opencode/src/provider/ollama.ts
@@ -0,0 +1,83 @@
+import { Log } from "../util/log"
+
+const log = Log.create({ service: "ollama" })
+
+export interface OllamaModel {
+  name: string
+  model: string
+  modified_at: string
+  size: number
+  digest: string
+  details?: {
+    parent_model: string
+    format: string
+    family: string
+    families: string[]
+    parameter_size: string
+    quantization_level: string
+  }
+}
+
+export interface OllamaTagsResponse {
+  models: OllamaModel[]
+}
+
+export interface OllamaStatus {
+  running: boolean
+  url: string
+  models: OllamaModel[]
+}
+
+const OLLAMA_DEFAULT_URL = "http://localhost:11434"
+const OLLAMA_API_TAGS = "/api/tags"
+
+export async function detect(url: string = OLLAMA_DEFAULT_URL): Promise<OllamaStatus> {
+  const start = Date.now()
+
+  try {
+    const controller = new AbortController()
+    const timeout = setTimeout(() => controller.abort(), 3000)
+
+    const response = await fetch(`${url}${OLLAMA_API_TAGS}`, {
+      signal: controller.signal,
+      headers: { Accept: "application/json" },
+    })
+
+    clearTimeout(timeout)
+
+    if (!response.ok) {
+      log.info("ollama not responding", { url, status: response.status })
+      return { running: false, url, models: [] }
+    }
+
+    const data = (await response.json()) as OllamaTagsResponse
+    const models = data.models ?? []
+
+    log.info("ollama detected", { url, modelCount: models.length, ms: Date.now() - start })
+
+    return {
+      running: true,
+      url,
+      models,
+    }
+  } catch (error) {
+    const err = error as Error
+    if (err.name === "AbortError") {
+      log.info("ollama timeout", { url })
+    } else {
+      log.info("ollama not running", { url, error: err.message })
+    }
+    return { running: false, url, models: [] }
+  }
+}
+
+export function parseModelName(fullName: string): { model: string; tag?: string } {
+  const colonIndex = fullName.lastIndexOf(":")
+  if (colonIndex === -1) {
+    return { model: fullName }
+  }
+  return {
+    model: fullName.substring(0, colonIndex),
+    tag: fullName.substring(colonIndex + 1),
+  }
+}
diff --git a/packages/opencode/src/provider/provider.ts b/packages/opencode/src/provider/provider.ts
index 6d5c9d1ad37..960f433acc9 100644
--- a/packages/opencode/src/provider/provider.ts
+++ b/packages/opencode/src/provider/provider.ts
@@ -45,6 +45,7 @@ import { fromNodeProviderChain } from "@aws-sdk/credential-providers"
 import { GoogleAuth } from "google-auth-library"
 import { ProviderTransform } from "./transform"
 import { Installation } from "../installation"
+import { detect as detectOllama, parseModelName } from "./ollama"
 
 export namespace Provider {
   const log = Log.create({ service: "provider" })
@@ -180,6 +181,22 @@ export namespace Provider {
         options: {},
       }
     },
+    ollama: async (input) => {
+      const hasKey = await (async () => {
+        const env = Env.all()
+        if (input.env.some((item) => env[item])) return true
+        if (await Auth.get(input.id)) return true
+        return true
+      })()
+
+      return {
+        autoload: hasKey,
+        async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
+          return sdk.chat(modelID)
+        },
+        options: { apiKey: "ollama" },
+      }
+    },
     azure: async () => {
       return {
         autoload: false,
@@ -1007,6 +1024,102 @@ export namespace Provider {
       mergeProvider(providerID, partial)
     }
 
+    // Patterns to detect reasoning-capable Ollama models
+    const OLLAMA_REASONING_PATTERNS = [
+      /qwen[_\-]?3/i, // qwen3, qwen-3, qwen_3
+      /phi[_\-]?4/i, // phi4, phi-4
+      /gemma[_\-]?3/i, // gemma3, gemma-3
+      /llama[_\-]?3/i, // llama3, llama-3
+      /r1$/i, // deepseek-r1, etc.
+      /qwq/i, // QwQ
+      /deepseek/i, // DeepSeek family
+      /gpt-?oss/i, // GPT-OSS
+    ]
+
+    function isOllamaReasoningModel(modelName: string, family?: string, families?: string[]): boolean {
+      const searchText = `${modelName} ${family ?? ""} ${families?.join(" ") ?? ""}`
+      return OLLAMA_REASONING_PATTERNS.some(pattern => pattern.test(searchText))
+    }
+
+    // Auto-detect Ollama if not already configured
+    const ollamaConfigured = providers["ollama"] || configProviders.some(([id]) => id === "ollama")
+    if (!ollamaConfigured) {
+      const ollama = await detectOllama()
+      if (ollama.running && ollama.models.length > 0) {
+        const ollamaProviderID = "ollama"
+        const ollamaModels: Record<string, Model> = {}
+
+        for (const ollamaModel of ollama.models) {
+          const { model, tag } = parseModelName(ollamaModel.name)
+          const modelID = tag ? `${model}:${tag}` : model
+
+          // Detect if this is a reasoning model based on patterns
+          const isReasoning = isOllamaReasoningModel(
+            model,
+            ollamaModel.details?.family,
+            ollamaModel.details?.families,
+          )
+
+          // Check for config overrides - allow forcing reasoning on/off
+          const configModel = config.provider?.ollama?.models?.[modelID]
+          const configForceReasoning = configModel?.reasoning // undefined = auto, true = force on, false = force off
+          const finalReasoning = configForceReasoning !== undefined ? configForceReasoning : isReasoning
+
+          ollamaModels[modelID] = {
+            id: modelID,
+            providerID: ollamaProviderID,
+            name: model,
+            family: ollamaModel.details?.family ?? model,
+            api: {
+              id: modelID,
+              url: ollama.url,
+              npm: "@ai-sdk/openai-compatible",
+            },
+            status: "active",
+            capabilities: {
+              temperature: true,
+              reasoning: finalReasoning,
+              attachment: false,
+              toolcall: true,
+              input: { text: true, audio: false, image: false, video: false, pdf: false },
+              output: { text: true, audio: false, image: false, video: false, pdf: false },
+              interleaved: finalReasoning
+                ? configModel?.interleaved ?? { field: "reasoning_content" }
+                : false,
+            },
+            cost: { input: 0, output: 0, cache: { read: 0, write: 0 } },
+            options:
+              finalReasoning
+                ? { reasoningEffort: configModel?.options?.reasoningEffort ?? "medium", ...configModel?.options }
+                : {},
+            limit: {
+              context:
+                configModel?.limit?.context ??
+                (finalReasoning ? 200000 : ollamaModel.details?.parameter_size ? 128000 : 8192),
+              output: configModel?.limit?.output ?? (finalReasoning ? 32768 : 8192),
+            },
+            headers: {},
+            release_date: "",
+            variants: {},
+          }
+          ollamaModels[modelID].variants = mapValues(
+            ProviderTransform.variants(ollamaModels[modelID]),
+            (v) => v,
+          )
+        }
+
+        providers[ollamaProviderID] = {
+          id: ollamaProviderID,
+          name: "Ollama",
+          source: "api",
+          env: [],
+          options: { baseURL: `${ollama.url}/v1`, apiKey: "ollama" },
+          models: ollamaModels,
+        }
+        log.info("ollama auto-detected", { modelCount: ollamaModels.length })
+      }
+    }
+
     for (const [providerID, provider] of Object.entries(providers)) {
       if (!isProviderAllowed(providerID)) {
         delete providers[providerID]
@@ -1016,6 +1129,28 @@ export namespace Provider {
       const configProvider = config.provider?.[providerID]
 
       for (const [modelID, model] of Object.entries(provider.models)) {
+        // Apply reasoning detection for Ollama models (both auto-detected and config-loaded)
+        if (providerID === "ollama") {
+          const isReasoning = isOllamaReasoningModel(modelID)
+          const configModel = config.provider?.ollama?.models?.[modelID]
+          const configForceReasoning = configModel?.reasoning
+          const finalReasoning = configForceReasoning !== undefined ? configForceReasoning : isReasoning
+
+          if (finalReasoning) {
+            model.capabilities.reasoning = true
+            if (!model.capabilities.interleaved) {
+              model.capabilities.interleaved = configModel?.interleaved ?? { field: "reasoning_content" }
+            }
+            model.options = { reasoningEffort: configModel?.options?.reasoningEffort ?? "medium", ...model.options }
+            if (!model.limit.output || model.limit.output < 32768) {
+              model.limit.output = configModel?.limit?.output ?? 32768
+            }
+            if (!model.limit.context || model.limit.context < 200000) {
+              model.limit.context = configModel?.limit?.context ?? 200000
+            }
+          }
+        }
+
         model.api.id = model.api.id ?? model.id ?? modelID
         if (modelID === "gpt-5-chat-latest" || (providerID === "openrouter" && modelID === "openai/gpt-5-chat"))
           delete provider.models[modelID]
diff --git a/packages/opencode/src/provider/transform.ts b/packages/opencode/src/provider/transform.ts
index 6980be05188..d195100f8bf 100644
--- a/packages/opencode/src/provider/transform.ts
+++ b/packages/opencode/src/provider/transform.ts
@@ -345,8 +345,14 @@ export namespace ProviderTransform {
       id.includes("kimi") ||
       // TODO: Remove this after models.dev data is fixed to use "kimi-k2.5" instead of "k2p5"
       id.includes("k2p5")
-    )
+    ) {
+      // Skip variants for certain providers that need special handling in options()
+      // but allow variants for Ollama models since they use @ai-sdk/openai-compatible
+      if (model.providerID === "ollama") {
+        return Object.fromEntries(WIDELY_SUPPORTED_EFFORTS.map((effort) => [effort, { reasoningEffort: effort }]))
+      }
       return {}
+    }
 
     // see: https://docs.x.ai/docs/guides/reasoning#control-how-hard-the-model-thinks
     if (id.includes("grok") && id.includes("grok-3-mini")) {
@@ -756,6 +762,28 @@ export namespace ProviderTransform {
       result["enable_thinking"] = true
     }
 
+    // Enable thinking for Ollama reasoning models via the "think" parameter
+    // Ollama supports thinking for: DeepSeek R1, DeepSeek v3.1, Qwen 3, GPT-OSS
+    // Most models accept true/false, GPT-OSS accepts low/medium/high
+    if (
+      input.model.providerID === "ollama" &&
+      input.model.capabilities.reasoning &&
+      input.model.api.npm === "@ai-sdk/openai-compatible"
+    ) {
+      // Check if it's a GPT-OSS model (supports thinking levels)
+      if (input.model.id.toLowerCase().includes("gpt-oss")) {
+        result["think"] = input.model.options?.reasoningEffort ?? "medium"
+      } else {
+        // For most models, enable thinking when not disabled
+        const effort = input.model.options?.reasoningEffort
+        if (effort === "none") {
+          result["think"] = false
+        } else {
+          result["think"] = true
+        }
+      }
+    }
+
     if (input.model.api.id.includes("gpt-5") && !input.model.api.id.includes("gpt-5-chat")) {
       if (!input.model.api.id.includes("gpt-5-pro")) {
         result["reasoningEffort"] = "medium"