@devries-cameron20 / t3code Mirror · read-only
Fork 0

The best way to control Claude Code, Codex and more from any computer or phone

Mirrored from https://github.com/Camo67/t3code.git

Synced Nov 21, 07:15 UTC

main
ts 81 lines 2.97 KB
Raw
1 import * as Cloudflare from "@alchemy.run/cloudflare";
2 import type { RuntimeContext } from "alchemy/RuntimeContext";
3 import { LanguageModel as AiLanguageModel } from "effect/ai";
4 import * as Effect from "effect/Effect";
5 import { HttpServerRequest } from "effect/http/HttpServerRequest";
6 import * as HttpServerResponse from "effect/http/HttpServerResponse";
7 import * as Stream from "effect/Stream";
8
9 // `@cf/meta/llama-3.1-8b-instruct` was deprecated by Cloudflare on
10 // 2026-05-30 (the API answers 410), so use the supported fast 3.3 model.
11 const MODEL = "@cf/meta/llama-3.3-70b-instruct-fp8-fast";
12
13 export default class AiBindingTestWorker extends Cloudflare.Worker<AiBindingTestWorker>()(
14 "AiBindingTestWorker",
15 {
16 main: import.meta.url,
17 },
18 Effect.gen(function* () {
19 const ai = yield* Cloudflare.Workers.AI();
20
21 const languageModel = ai.model({
22 model: MODEL,
23 parameters: { temperature: 0.7, maxTokens: 1024 },
24 });
25
26 return {
27 fetch: Effect.gen(function* () {
28 const request = yield* HttpServerRequest;
29 const ctx = yield* Effect.context<RuntimeContext>();
30 const url = new URL(request.url, "http://worker");
31 const prompt =
32 url.searchParams.get("prompt") ?? "Say the single word 'pong' and nothing else.";
33
34 if (url.pathname === "/run") {
35 // Raw `ai.run` against the binding — no gateway involved.
36 const result = yield* ai
37 .run(MODEL, {
38 messages: [{ role: "user", content: prompt }],
39 max_tokens: 128,
40 })
41 .pipe(Effect.orDie);
42 return yield* HttpServerResponse.json(result);
43 }
44
45 if (url.pathname === "/models") {
46 const models = yield* ai.models({ search: "llama-3.3" }).pipe(Effect.orDie);
47 return yield* HttpServerResponse.json({
48 count: models.length,
49 names: models.map((m) => m.name),
50 });
51 }
52
53 if (url.pathname === "/generate") {
54 const response = yield* AiLanguageModel.generateText({ prompt }).pipe(Effect.orDie);
55 return yield* HttpServerResponse.json({
56 text: response.text,
57 finishReason: response.finishReason,
58 usage: {
59 inputTokens: response.usage.inputTokens.total,
60 outputTokens: response.usage.outputTokens.total,
61 },
62 });
63 }
64
65 if (url.pathname === "/stream") {
66 const encoder = new TextEncoder();
67 const body = AiLanguageModel.streamText({ prompt }).pipe(
68 Stream.map((part) => encoder.encode(`data: ${JSON.stringify(part)}\n\n`)),
69 Stream.provide(languageModel),
70 Stream.provideContext(ctx),
71 );
72 return HttpServerResponse.stream(body, {
73 headers: { "content-type": "text/event-stream" },
74 });
75 }
76
77 return HttpServerResponse.text("ok");
78 }).pipe(Effect.provide(languageModel)),
79 };
80 }).pipe(Effect.provide(Cloudflare.Workers.AIBinding)),
81 ) {}