diff --git a/packages/app/src/adapters/claude-adapter.ts b/packages/app/src/adapters/claude-adapter.ts
new file mode 100644
index 00000000..c65c3ee9
--- /dev/null
+++ b/packages/app/src/adapters/claude-adapter.ts
@@ -0,0 +1,106 @@
+import type { AIAdapter, ChatMessage, ChatOptions } from './types'
+import { getApiKey } from '@/utils/key-vault'
+
+const CLAUDE_PATH = '/api/claude/v1/messages'
+
+export const claudeAdapter: AIAdapter = {
+  id: 'claude',
+  name: 'Claude (Max)',
+  supportsStreaming: true,
+  supportsVision: true,
+  supportsTools: true,
+
+  models() {
+    return [
+      { id: 'claude-haiku-4.5', name: 'Claude 4.5 Haiku' },
+      { id: 'claude-sonnet-4', name: 'Claude Sonnet 4' },
+      { id: 'claude-opus-4', name: 'Claude Opus 4' },
+    ]
+  },
+
+  async chat(messages, options, onToken, onError) {
+    const headers: Record<string, string> = { 'Content-Type': 'application/json' }
+
+    const vaultKey = await getApiKey('claude')
+    if (vaultKey) {
+      headers['x-api-key'] = vaultKey
+    }
+
+    const apiMessages = messages
+      .filter(m => m.role !== 'system')
+      .map(m => ({ role: m.role, content: m.content }))
+
+    const res = await fetch(CLAUDE_PATH, {
+      method: 'POST',
+      headers,
+      body: JSON.stringify({
+        model: options.model,
+        system: options.systemPrompt,
+        messages: apiMessages,
+        stream: true,
+        webSearch: options.webSearch ?? false,
+      }),
+      signal: options.signal,
+    })
+
+    if (!res.ok) {
+      const body = await res.text().catch(() => 'Could not read error body')
+      onError(`Claude proxy error ${res.status}: ${body}`)
+      return
+    }
+
+    await readSSE(res, (data) => {
+      const parsed = JSON.parse(data)
+      if (parsed.type === 'content_block_delta' && parsed.delta?.text) {
+        onToken(parsed.delta.text)
+      } else if (parsed.type === 'error') {
+        onError(parsed.error?.message ?? 'Claude stream error')
+      }
+    }, onError, options.signal)
+  },
+}
+
+async function readSSE(
+  res: Response,
+  onData: (data: string) => void,
+  onError: (err: string) => void,
+  signal?: AbortSignal,
+): Promise<void> {
+  const reader = res.body?.getReader()
+  if (!reader) {
+    onError('No response body')
+    return
+  }
+
+  const decoder = new TextDecoder()
+  let buffer = ''
+
+  try {
+    while (true) {
+      if (signal?.aborted) {
+        reader.cancel()
+        return
+      }
+      const { done, value } = await reader.read()
+      if (done) break
+
+      buffer += decoder.decode(value, { stream: true })
+      const lines = buffer.split('\n')
+      buffer = lines.pop() ?? ''
+
+      for (const line of lines) {
+        const trimmed = line.trim()
+        if (!trimmed || !trimmed.startsWith('data: ')) continue
+        const payload = trimmed.slice(6)
+        if (payload === '[DONE]') return
+        try {
+          onData(payload)
+        } catch {
+          // skip malformed chunks
+        }
+      }
+    }
+  } finally {
+    reader.cancel().catch(() => {})
+  }
+}
diff --git a/packages/app/src/adapters/ollama-adapter.ts b/packages/app/src/adapters/ollama-adapter.ts
new file mode 100644
index 00000000..2aadc1d6
--- /dev/null
+++ b/packages/app/src/adapters/ollama-adapter.ts
@@ -0,0 +1,94 @@
+import type { AIAdapter, ChatOptions } from './types'
+
+const OLLAMA_BASE = 'http://localhost:11434'
+
+export const ollamaAdapter: AIAdapter = {
+  id: 'ollama',
+  name: 'Ollama (Local)',
+  supportsStreaming: true,
+  supportsVision: false,
+  supportsTools: false,
+
+  models() {
+    return [
+      { id: 'llama3.2', name: 'Llama 3.2' },
+      { id: 'mistral', name: 'Mistral' },
+      { id: 'gemma2', name: 'Gemma 2' },
+      { id: 'qwen2.5', name: 'Qwen 2.5' },
+    ]
+  },
+
+  async chat(messages, options, onToken, onError) {
+    const ollamaMessages = messages.map(m => ({
+      role: m.role,
+      content: m.content,
+    }))
+
+    if (options.systemPrompt) {
+      ollamaMessages.unshift({ role: 'system', content: options.systemPrompt })
+    }
+
+    let res: Response
+    try {
+      res = await fetch(`${OLLAMA_BASE}/api/chat`, {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify({
+          model: options.model,
+          messages: ollamaMessages,
+          stream: true,
+        }),
+        signal: options.signal,
+      })
+    } catch {
+      onError('Cannot connect to Ollama. Is it running on localhost:11434?')
+      return
+    }
+
+    if (!res.ok) {
+      const body = await res.text().catch(() => 'Could not read error body')
+      onError(`Ollama error ${res.status}: ${body}`)
+      return
+    }
+
+    // Ollama uses newline-delimited JSON (not SSE)
+    const reader = res.body?.getReader()
+    if (!reader) {
+      onError('No response body')
+      return
+    }
+
+    const decoder = new TextDecoder()
+    let buffer = ''
+
+    try {
+      while (true) {
+        if (options.signal?.aborted) {
+          reader.cancel()
+          return
+        }
+        const { done, value } = await reader.read()
+        if (done) break
+
+        buffer += decoder.decode(value, { stream: true })
+        const lines = buffer.split('\n')
+        buffer = lines.pop() ?? ''
+
+        for (const line of lines) {
+          if (!line.trim()) continue
+          try {
+            const parsed = JSON.parse(line)
+            if (parsed.message?.content) {
+              onToken(parsed.message.content)
+            }
+            if (parsed.done) return
+          } catch {
+            // skip malformed lines
+          }
+        }
+      }
+    } finally {
+      reader.cancel().catch(() => {})
+    }
+  },
+}
diff --git a/packages/app/src/adapters/openrouter-adapter.ts b/packages/app/src/adapters/openrouter-adapter.ts
new file mode 100644
index 00000000..ac2ee05b
--- /dev/null
+++ b/packages/app/src/adapters/openrouter-adapter.ts
@@ -0,0 +1,113 @@
+import type { AIAdapter, ChatOptions } from './types'
+import { getApiKey } from '@/utils/key-vault'
+
+const OPENROUTER_PATH = '/api/openrouter'
+
+export const openrouterAdapter: AIAdapter = {
+  id: 'openrouter',
+  name: 'OpenRouter',
+  supportsStreaming: true,
+  supportsVision: false,
+  supportsTools: false,
+
+  models() {
+    return [
+      { id: 'meta-llama/llama-4-maverick', name: 'Llama 4 Maverick' },
+      { id: 'qwen/qwen3-235b-a22b-thinking-2507', name: 'Qwen3 235B Thinking' },
+      { id: 'mistralai/mistral-small-3.1-24b-instruct:free', name: 'Mistral Small 3.1 (free)' },
+      { id: 'google/gemma-3-27b-it:free', name: 'Gemma 3 27B (free)' },
+    ]
+  },
+
+  async chat(messages, options, onToken, onError) {
+    const orMessages = messages.map(m => ({
+      role: m.role as 'user' | 'assistant' | 'system',
+      content: m.content,
+    }))
+
+    // Prepend system prompt as system message
+    if (options.systemPrompt) {
+      orMessages.unshift({ role: 'system', content: options.systemPrompt })
+    }
+
+    const headers: Record<string, string> = {
+      'Content-Type': 'application/json',
+      'HTTP-Referer': window.location.origin,
+      'X-Title': 'AIUI',
+    }
+
+    const vaultKey = await getApiKey('openrouter')
+    if (vaultKey) {
+      headers['Authorization'] = `Bearer ${vaultKey}`
+    }
+
+    const res = await fetch(OPENROUTER_PATH, {
+      method: 'POST',
+      headers,
+      body: JSON.stringify({
+        model: options.model,
+        messages: orMessages,
+        stream: true,
+      }),
+      signal: options.signal,
+    })
+
+    if (!res.ok) {
+      const body = await res.text().catch(() => 'Could not read error body')
+      onError(`OpenRouter error ${res.status}: ${body}`)
+      return
+    }
+
+    await readSSE(res, (data) => {
+      if (data === '[DONE]') return
+      const parsed = JSON.parse(data)
+      const delta = parsed.choices?.[0]?.delta?.content
+      if (delta) onToken(delta)
+    }, onError, options.signal)
+  },
+}
+
+async function readSSE(
+  res: Response,
+  onData: (data: string) => void,
+  onError: (err: string) => void,
+  signal?: AbortSignal,
+): Promise<void> {
+  const reader = res.body?.getReader()
+  if (!reader) {
+    onError('No response body')
+    return
+  }
+
+  const decoder = new TextDecoder()
+  let buffer = ''
+
+  try {
+    while (true) {
+      if (signal?.aborted) {
+        reader.cancel()
+        return
+      }
+      const { done, value } = await reader.read()
+      if (done) break
+
+      buffer += decoder.decode(value, { stream: true })
+      const lines = buffer.split('\n')
+      buffer = lines.pop() ?? ''
+
+      for (const line of lines) {
+        const trimmed = line.trim()
+        if (!trimmed || !trimmed.startsWith('data: ')) continue
+        const payload = trimmed.slice(6)
+        if (payload === '[DONE]') return
+        try {
+          onData(payload)
+        } catch {
+          // skip malformed chunks
+        }
+      }
+    }
+  } finally {
+    reader.cancel().catch(() => {})
+  }
+}
diff --git a/packages/app/src/adapters/types.ts b/packages/app/src/adapters/types.ts
new file mode 100644
index 00000000..7f33a839
--- /dev/null
+++ b/packages/app/src/adapters/types.ts
@@ -0,0 +1,34 @@
+/**
+ * Unified AI adapter interface for normalizing different provider APIs.
+ */
+
+export interface ChatMessage {
+  role: 'user' | 'assistant' | 'system'
+  content: string
+}
+
+export interface ChatOptions {
+  model: string
+  systemPrompt?: string
+  webSearch?: boolean
+  signal?: AbortSignal
+}
+
+export interface AIAdapter {
+  id: string
+  name: string
+  supportsStreaming: boolean
+  supportsVision: boolean
+  supportsTools: boolean
+
+  /** List available models for this adapter */
+  models(): { id: string; name: string }[]
+
+  /** Stream chat completion, yielding text tokens */
+  chat(
+    messages: ChatMessage[],
+    options: ChatOptions,
+    onToken: (text: string) => void,
+    onError: (err: string) => void,
+  ): Promise<void>
+}