diff --git a/bun.lock b/bun.lock
index 9d5cf8a5..eb29a379 100644
--- a/bun.lock
+++ b/bun.lock
@@ -6,6 +6,7 @@
       "name": "smshub",
       "dependencies": {
         "@profullstack/affiliates": "^0.3.0",
+        "@profullstack/ai-providers": "0.1.2",
         "@profullstack/coinpay-onboard": "0.4.0",
         "@profullstack/footer": "latest",
         "@profullstack/profile": "0.1.8",
@@ -295,6 +296,8 @@
 
     "@profullstack/affiliates": ["@profullstack/affiliates@0.3.0", "", { "dependencies": { "@profullstack/referrals": "^0.1.0" }, "bin": { "affiliates": "dist/cli.js" } }, "sha512-bp0iREciWetDWrQYOsa6SJRTzviekKwgJQuATT7vql+4equu7pN76q8tb3SP1oBMczydsv6rYUMvPpS6+QT7yQ=="],
 
+    "@profullstack/ai-providers": ["@profullstack/ai-providers@0.1.2", "", { "bin": { "ai-providers": "src/cli.js" } }, "sha512-nOcKaOhfcDM+RNruA4+znva5xXxI4HF77iqUCgieuou2w8yY8LIF2p5ajSVD4kzFtfIUjC7GhGV8eyOrliHg1w=="],
+
     "@profullstack/coinpay-onboard": ["@profullstack/coinpay-onboard@0.4.0", "", { "dependencies": { "@modelcontextprotocol/sdk": "^1.32.2", "zod": "^4.6.7" }, "bin": { "coinpay-onboard": "bin/coinpay-onboard" } }, "sha512-L2P0fxNswZnXt1tvo6t19L/RrJN204vrWnIEY65QI+zriZN7JJPY4Ap6Nv8MLkyA5wDvY88s0+Dgjf9Myw/tUA=="],
 
     "@profullstack/emailer": ["@profullstack/emailer@1.0.3", "", {}, "sha512-VyAtCcfnSI6LOwJvWogCr3ZNmEsaZw+iuWrqnosLsdcKsbRyCrCmXigMNZJZEjlgYlR+lK//PwNPG+xtCFvWAw=="],
diff --git a/package.json b/package.json
index d75ba203..e6faa22c 100644
--- a/package.json
+++ b/package.json
@@ -25,6 +25,7 @@
   },
   "dependencies": {
     "@profullstack/affiliates": "^0.3.0",
+    "@profullstack/ai-providers": "0.1.2",
     "@profullstack/coinpay-onboard": "0.4.0",
     "@profullstack/footer": "latest",
     "@profullstack/profile": "0.1.8",
diff --git a/src/app/privacy/page.tsx b/src/app/privacy/page.tsx
index 134fd173..f648b7d1 100644
--- a/src/app/privacy/page.tsx
+++ b/src/app/privacy/page.tsx
@@ -33,7 +33,7 @@ export default async function PrivacyPage() {
         <ul className="list-disc list-inside space-y-1 text-gray-300">
           <li>Send and receive SMS messages on your behalf through connected providers</li>
           <li>Display your conversations, contacts, and message history</li>
-          <li>Provide AI-powered reply suggestions (when enabled, using OpenAI)</li>
+          <li>Provide AI-powered reply suggestions (when enabled, using c0mpute, with OpenAI as a fallback)</li>
           <li>Send push notifications for new messages</li>
           <li>Improve the platform and fix bugs</li>
           <li>Communicate with you about your account and service updates</li>
@@ -60,7 +60,8 @@ export default async function PrivacyPage() {
         </p>
         <p>
           We do not sell, rent, or share your personal data or message content with third parties,
-          except as necessary to provide the service (e.g., sending messages through Twilio/Telnyx).
+          except as necessary to provide the service (e.g., sending messages through Twilio/Telnyx,
+          or generating an AI reply suggestion you asked for).
         </p>
         <p>
           No mobile information will be shared with third parties or affiliates for marketing or
@@ -73,7 +74,8 @@ export default async function PrivacyPage() {
         <p>{name} integrates with the following third-party services:</p>
         <ul className="list-disc list-inside space-y-1 text-gray-300">
           <li><strong>Twilio / Telnyx / phonenumbers.bot:</strong> SMS sending and receiving</li>
-          <li><strong>OpenAI:</strong> AI reply suggestions (opt-in, per conversation)</li>
+          <li><strong>c0mpute:</strong> AI reply suggestions (opt-in, per conversation). The recent messages in that conversation are processed by nodes on a peer-to-peer inference network, which may be run by independent operators.</li>
+          <li><strong>OpenAI:</strong> Fallback for AI reply suggestions when c0mpute is unavailable</li>
           <li><strong>Supabase:</strong> Database, authentication, and real-time messaging</li>
           <li><strong>Sentry:</strong> Error tracking and performance monitoring</li>
           <li><strong>Expo:</strong> Mobile app push notifications</li>
diff --git a/src/lib/ai/__tests__/auto-reply.test.ts b/src/lib/ai/__tests__/auto-reply.test.ts
index eb892988..267049e1 100644
--- a/src/lib/ai/__tests__/auto-reply.test.ts
+++ b/src/lib/ai/__tests__/auto-reply.test.ts
@@ -1,11 +1,14 @@
 import { describe, it, expect, vi, beforeEach, afterEach } from "vitest";
-import { suggestReply } from "../auto-reply";
 
 describe("suggestReply", () => {
   const originalEnv = process.env;
+  let suggestReply: typeof import("../auto-reply").suggestReply;
 
-  beforeEach(() => {
-    process.env = { ...originalEnv, OPENAI_API_KEY: "test-key" };
+  beforeEach(async () => {
+    process.env = { ...originalEnv, OPENAI_API_KEY: "test-key", C0MPUTE_API_KEY: "" };
+    // A fresh router each test: a failure parks its provider.
+    vi.resetModules();
+    ({ suggestReply } = await import("../auto-reply"));
   });
 
   afterEach(() => {
@@ -80,4 +83,45 @@ describe("suggestReply", () => {
       })
     ).rejects.toThrow("No suggestion generated");
   });
+
+  describe("with a c0mpute key", () => {
+    const message = [{ direction: "inbound" as const, body: "Hello", created_at: "2024-01-01T00:00:00Z" }];
+
+    beforeEach(() => {
+      process.env.C0MPUTE_API_KEY = "ik_live_test";
+    });
+
+    it("asks c0mpute first and never sends the phone number", async () => {
+      const fetchMock = vi.spyOn(global, "fetch").mockImplementation((async (url: string) =>
+        url.endsWith("/api/chat")
+          ? Response.json({ jobId: "j", source: "p2p", model: "m" })
+          : new Response(`event: done\ndata: ${JSON.stringify({ text: "<think>hm</think>On my way." })}\n\n`)) as typeof fetch);
+
+      const result = await suggestReply({ messages: message, contactPhone: "+15551234567" });
+
+      expect(result.suggestion).toBe("On my way.");
+      const calls = fetchMock.mock.calls.map(([url]) => String(url));
+      expect(calls.every((url) => url.includes("infernetprotocol.com"))).toBe(true);
+      expect(String(fetchMock.mock.calls[0][1]?.body)).not.toContain("5551234567");
+    });
+
+    it("works on c0mpute alone", async () => {
+      delete process.env.OPENAI_API_KEY;
+      vi.spyOn(global, "fetch").mockImplementation((async (url: string) =>
+        url.endsWith("/api/chat")
+          ? Response.json({ jobId: "j", source: "p2p", model: "m" })
+          : new Response(`event: done\ndata: ${JSON.stringify({ text: "Sure." })}\n\n`)) as typeof fetch);
+
+      expect((await suggestReply({ messages: message })).suggestion).toBe("Sure.");
+    });
+
+    it("falls back to OpenAI, and a dead OpenAI still throws", async () => {
+      vi.spyOn(global, "fetch").mockImplementation((async (url: string) =>
+        url.includes("openai.com")
+          ? new Response("Incorrect API key", { status: 401 })
+          : new Response("{}", { status: 503 })) as typeof fetch);
+
+      await expect(suggestReply({ messages: message })).rejects.toThrow("OpenAI API error: 401");
+    });
+  });
 });
diff --git a/src/lib/ai/auto-reply.ts b/src/lib/ai/auto-reply.ts
index e7435890..087173e2 100644
--- a/src/lib/ai/auto-reply.ts
+++ b/src/lib/ai/auto-reply.ts
@@ -1,6 +1,10 @@
 /**
- * AI Auto-Reply using OpenAI API (gpt-4o-mini)
+ * AI Auto-Reply: c0mpute first, OpenAI (gpt-4o-mini) behind it.
  */
+import { createAiProviders, type AiRouter } from "@profullstack/ai-providers";
+
+// c0mpute takes 30 s or more, and someone is waiting on the suggestion.
+const C0MPUTE_TIMEOUT_MS = 100_000;
 
 export interface Message {
   direction: "inbound" | "outbound";
@@ -18,31 +22,26 @@ export interface SuggestReplyResult {
   suggestion: string;
 }
 
-/**
- * Generate a reply suggestion using OpenAI's gpt-4o-mini model.
- */
-export async function suggestReply(
-  options: SuggestReplyOptions
-): Promise<SuggestReplyResult> {
-  const apiKey = process.env.OPENAI_API_KEY;
-  if (!apiKey) {
-    throw new Error("OPENAI_API_KEY is not configured");
-  }
-
-  const { messages, contactName, contactPhone } = options;
+const c0mputeOn = (): boolean =>
+  Boolean(process.env.C0MPUTE_API_KEY) && !/^(1|true|yes|on)$/i.test(process.env.C0MPUTE_DISABLED ?? "");
 
-  const conversationContext = messages
-    .slice(-20) // Last 20 messages for context
-    .map(
-      (m) =>
-        `${m.direction === "inbound" ? (contactName || contactPhone || "Contact") : "You"}: ${m.body}`
-    )
-    .join("\n");
+let shared: { keys: string; router: AiRouter } | undefined;
 
-  const systemPrompt = `You are an SMS reply assistant. Generate a brief, natural reply suggestion for an SMS conversation. Keep it concise (1-2 sentences max) and appropriate for text messaging. Do not use emojis unless the conversation style uses them. Reply with ONLY the suggested message text, nothing else.`;
-
-  const userPrompt = `Here is the recent SMS conversation with ${contactName || contactPhone || "a contact"}:\n\n${conversationContext}\n\nSuggest a reply:`;
+/** One router per set of keys, so a parked provider stays parked across calls. */
+function router(): AiRouter {
+  const keys = `${c0mputeOn()}|${process.env.OPENAI_API_KEY ?? ""}`;
+  if (shared?.keys !== keys) {
+    shared = {
+      keys,
+      router: createAiProviders({
+        env: { ...process.env, C0MPUTE_DISABLED: c0mputeOn() ? "" : "1" },
+      }),
+    };
+  }
+  return shared.router;
+}
 
+async function openaiReply(apiKey: string, systemPrompt: string, userPrompt: string): Promise<string> {
   const response = await fetch("https://api.openai.com/v1/chat/completions", {
     method: "POST",
     headers: {
@@ -66,11 +65,51 @@ export async function suggestReply(
   }
 
   const data = await response.json();
-  const suggestion = data.choices?.[0]?.message?.content?.trim();
+  return data.choices?.[0]?.message?.content ?? "";
+}
 
-  if (!suggestion) {
-    throw new Error("No suggestion generated");
+/**
+ * Generate a reply suggestion, c0mpute first with OpenAI as the fallback.
+ */
+export async function suggestReply(
+  options: SuggestReplyOptions
+): Promise<SuggestReplyResult> {
+  const apiKey = process.env.OPENAI_API_KEY;
+  if (!apiKey && !c0mputeOn()) {
+    throw new Error("OPENAI_API_KEY is not configured");
   }
 
+  const { messages, contactName } = options;
+  // Never the phone number: the model has no use for it, and c0mpute nodes are
+  // run by strangers.
+  const contact = contactName || "Contact";
+
+  const conversationContext = messages
+    .slice(-20) // Last 20 messages for context
+    .map((m) => `${m.direction === "inbound" ? contact : "You"}: ${m.body}`)
+    .join("\n");
+
+  const systemPrompt = `You are an SMS reply assistant. Generate a brief, natural reply suggestion for an SMS conversation. Keep it concise (1-2 sentences max) and appropriate for text messaging. Do not use emojis unless the conversation style uses them. Reply with ONLY the suggested message text, nothing else.`;
+
+  const userPrompt = `Here is the recent SMS conversation with ${contactName || "a contact"}:\n\n${conversationContext}\n\nSuggest a reply:`;
+
+  const { value: suggestion } = await router().run(async (provider) => {
+    const text =
+      provider.name === "openai"
+        ? await openaiReply(apiKey as string, systemPrompt, userPrompt)
+        : (
+            await provider.generate({
+              system: systemPrompt,
+              user: userPrompt,
+              maxTokens: 150,
+              temperature: 0.7,
+              timeoutMs: C0MPUTE_TIMEOUT_MS,
+            })
+          ).text;
+    const reply = text.replace(/<think>[\s\S]*?<\/think>/gi, "").trim();
+    if (!reply) throw new Error("No suggestion generated");
+    return reply;
+  });
+
   return { suggestion };
 }