[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8},"actionData","errors","root",{"_107":108,"_109":110},"routes/embed.x.$handle.$name.$",{"_9":10,"_11":12,"_13":14,"_15":16,"_17":18,"_19":20,"_21":-5},"handle","ramyharoon","name","openai-gateway","path","lib/bridge.ts","project",{"_22":48,"_11":12,"_35":49,"_50":-5,"_51":44,"_52":53,"_54":53,"_55":56,"_57":-5,"_58":59,"_60":61,"_62":34,"_63":-5,"_64":-5,"_65":66,"_67":68,"_69":-5,"_70":-5,"_71":-5,"_72":-5,"_73":74,"_75":74,"_76":77,"_78":77,"_79":77,"_80":74,"_81":74,"_82":83,"_84":85,"_86":87,"_88":44},"file",{"_22":23,"_11":24,"_25":26,"_27":28,"_29":30,"_31":32,"_33":34,"_35":36,"_37":38,"_39":40,"_41":42,"_43":44,"_45":-5,"_13":46},"permalink","https://www.val.town/x/ramyharoon/openai-gateway/code/lib/bridge.ts","extension","id","2790bd94-ec1e-4614-a085-ce240493edf3","bridge.ts","fileId","c9e9cd50-a6e9-11f1-8b7e-1607ee4eb77e","parentId","e2a0cbeb-2016-432b-9f07-0defe6ee8dbf","branchId","7bdd8960-6221-4505-900d-d05422b73b8d","branchVersion",45,"latest",true,"createdAt",["D",1788369399905],"version",5,"fileType","script","content","/**\n * lib/bridge.ts — Internal Provider Bridge & Native AI Engine\n *\n * Maps standard OpenAI requests onto Val Town's native AI engine\n * (`std/openai` -> ValTownOpenAI proxy, authenticated via the ambient\n * `valtown` token — no external API key required).\n *\n * Exposes (mounted by main.ts under /api/provider/v1):\n * GET /models -> active native model, OpenAI format, no auth\n * POST /chat/completions -> chat (streaming real SSE or non-streaming)\n *\n * @service the \"port 3040 internal backend service\" of this deployment —\n * no extra process needed: it's a module singleton.\n */\nimport { OpenAI } from \"https://esm.town/v/std/openai/main.ts\";\n\n// ---- Configuration -------------------------------------------------------\n\n/** Native engine is Val Town's proxy; model is configurable. */\nexport const MODEL = (Deno.env.get(\"PROVIDER_MODEL\") || \"gpt-5.4-nano\").trim();\n/** Hard ceiling for output tokens to avoid upstream rejection errors. */\nconst MAX_OUTPUT_TOKENS = 16384;\n\n// ---- Singleton client with auto-healing ----------------------------------\n// If the ambient session token rotates or a transient error occurs, we\n// re-instantiate the client so it re-reads platform credentials.\n\nlet client: OpenAI | null = null;\n\n/** Returns (and lazily creates) the native engine client singleton. */\nexport function getClient(): OpenAI {\n if (!client) client = new OpenAI();\n return client;\n}\n\n/** Force a fresh client on the next call (auto-heal). */\nexport function resetClient(): void {\n client = null;\n}\n\n/**\n * Invoke the engine with an internal rate-limiter and auto-retry on 429.\n * - Throttles to ~9 requests/min so the proxy's ~10/min budget is never\n * exceeded from this gateway, keeping 429s rare.\n * - On a 429 (rate limit) it sleeps with backoff and retries.\n * - Auto-heal: resets the client singleton (re-reading credentials) on error.\n */\nconst MAX_REQ_PER_MIN = 9;\nconst engineTimes: number[] = [];\nfunction sleep(ms: number): Promise {\n return new Promise((r) => setTimeout(r, ms));\n}\nasync function throttleEngine(): Promise {\n const now = Date.now();\n while (engineTimes.length && now - engineTimes[0] >= 60000) engineTimes.shift();\n if (engineTimes.length >= MAX_REQ_PER_MIN) {\n // Wait until the oldest request ages out of the window.\n await sleep(60000 - (now - engineTimes[0]) + 100);\n engineTimes.shift();\n }\n engineTimes.push(Date.now());\n}\n\nasync function callEngine(payload: Record) {\n const attempts = 4;\n for (let i = 0; i < attempts; i++) {\n await throttleEngine();\n try {\n return await getClient().chat.completions.create(payload as never);\n } catch (e) {\n const err = e as { message?: string; status?: number };\n resetClient(); // auto-heal: re-read platform credentials\n // 429 = rate limited: back off and retry.\n if (err.status === 429 && i < attempts - 1) {\n await sleep(2600 * (i + 1));\n continue;\n }\n // One plain retry for transient errors, then surface.\n if (i === 0) continue;\n const status =\n err.status && err.status < 600 && err.status >= 400 ? err.status : 502;\n throw new UpstreamError(err.message || \"native engine failed\", status);\n }\n }\n throw new UpstreamError(\"engine retries exhausted (rate limit)\", 429);\n}\n\nclass UpstreamError extends Error {\n status: number;\n constructor(message: string, status = 502) {\n super(message);\n this.status = status;\n }\n}\n\n// ---- Authentication ------------------------------------------------------\n\n/** The generated gateway key (stored server-side in PROVIDER_API_KEY). */\nexport function apiKey(): string {\n return Deno.env.get(\"PROVIDER_API_KEY\") || \"\";\n}\n\n/** True if the request carries a valid bearer / x-api-key. */\nexport function isAuthed(req: Request): boolean {\n const header = req.headers.get(\"authorization\") || req.headers.get(\"x-api-key\") || \"\";\n const token = header.replace(/^Bearer\\s+/i, \"\").trim();\n return token.length > 0 && token === apiKey();\n}\n\nfunction unauthorized(): Response {\n return Response.json(\n {\n error: {\n message:\n \"Incorrect API key provided. Include `Authorization: Bearer ` or `X-Api-Key: `. Get your key from the gateway UI.\",\n type: \"invalid_request_error\",\n param: null,\n code: \"invalid_api_key\",\n },\n },\n { status: 401 },\n );\n}\n\n// ---- Message sanitization (never throws) ---------------------------------\n\nfunction toText(content: unknown): string {\n try {\n if (content == null) return \"\";\n if (typeof content === \"string\") return content;\n if (Array.isArray(content)) {\n return content\n .map((part) => {\n if (!part || typeof part !== \"object\") return \"\";\n const p = part as { type?: string; text?: string; image_url?: { url?: string } };\n if (p.type === \"text\") return String(p.text ?? \"\");\n if (p.type === \"image_url\") return \"\\[image\\]\";\n return \"\";\n })\n .filter(Boolean)\n .join(\"\\n\");\n }\n return String(content);\n } catch {\n return \"\";\n }\n}\n\n/**\n * Pass through plain system/user/assistant messages; safely flatten\n * tool / function roles and role-object shapes into clean text messages.\n * Guaranteed to never throw.\n */\nexport function sanitizeMessages(messages: unknown): { role: string; content: string }[] {\n try {\n if (!Array.isArray(messages)) return [{ role: \"user\", content: \"Hi\" }];\n const out: { role: string; content: string }[] = [];\n for (const raw of messages) {\n try {\n const msg = (raw ?? {}) as {\n role?: string | { role?: string };\n content?: unknown;\n tool_calls?: { function?: { name?: string } }[];\n name?: string;\n };\n let role = typeof msg.role === \"string\" ? msg.role : \"user\";\n let content = toText(msg.content);\n // Flatten assistant tool_calls into a note so nothing is lost.\n const calls = Array.isArray(msg.tool_calls)\n ? msg.tool_calls.filter(Boolean).map((t) => t?.function?.name || \"tool\")\n : [];\n if (calls.length && !content) {\n content = \"[invoked: \" + calls.join(\", \") + \"]\";\n }\n if (role === \"tool\" || role === \"function\") role = \"user\";\n // Merge adjacent same-role messages so alternating-role rules hold.\n const last = out[out.length - 1];\n if (last && last.role === role) {\n last.content += (last.content ? \"\\n\" : \"\") + content;\n continue;\n }\n out.push({ role, content });\n } catch {\n // drop malformed message\n }\n }\n // Leading assistant messages upset some engines — prepend a user turn.\n if (out.length && out[0].role === \"assistant\") {\n out.unshift({ role: \"user\", content: \"Continue.\" });\n }\n if (!out.length) out.push({ role: \"user\", content: \"Hi\" });\n return out;\n } catch {\n return [{ role: \"user\", content: \"Hi\" }];\n }\n}\n\n// ---- Payload shaping ------------------------------------------------------\n\nfunction clampOutputTokens(v: unknown, fallback = 1024): number {\n let n = Number(v ?? fallback);\n if (!Number.isFinite(n) || n <= 0) n = fallback;\n return Math.min(Math.floor(n), MAX_OUTPUT_TOKENS);\n}\n\nfunction buildPayload(body: Record, model: string) {\n const messages = sanitizeMessages(body.messages);\n const maxTokens = clampOutputTokens(\n body.max_tokens ?? body.max_completion_tokens,\n 2048,\n );\n // Reasoning/thinking models (GPT-5 family, o-series) reject temperature/top_p\n // (only the default (1) is supported) but accept reasoning_effort.\n const isReasoning = /^gpt-5|^o[0-9]|^o[0-9]-mini|reasons/.test(String(model));\n const supportsSampling = !isReasoning && !isClaude;\n // Claude models (via optional Anthropic bridge) also skip sampling params.\n const isClaude = /^claude|^anthropic/.test(String(model));\n\n const payload: Record = {\n model,\n messages,\n max_completion_tokens: maxTokens,\n };\n if (supportsSampling && typeof body.temperature === \"number\") payload.temperature = body.temperature;\n if (supportsSampling && typeof body.top_p === \"number\") payload.top_p = body.top_p;\n if (supportsSampling && typeof body.presence_penalty === \"number\") payload.presence_penalty = body.presence_penalty;\n if (supportsSampling && typeof body.frequency_penalty === \"number\") payload.frequency_penalty = body.frequency_penalty;\n if (!isClaude && typeof body.stop !== \"undefined\") payload.stop = body.stop;\n if (!isClaude && typeof body.user === \"string\") payload.user = body.user;\n if (!isClaude && typeof body.seed === \"number\") payload.seed = body.seed;\n // Reasoning effort is only meaningful (and accepted) on thinking models.\n if (isReasoning && !isClaude) {\n const effort =\n typeof body.reasoning_effort === \"string\" && body.reasoning_effort\n ? body.reasoning_effort\n : typeof body.effort === \"string\" && body.effort\n ? body.effort\n : \"\";\n if (effort) payload.reasoning_effort = effort;\n }\n // Reasoning models burn tokens \"thinking\" before the visible answer — make\n // sure the requested budget leaves room for an actual reply.\n if (isReasoning && maxTokens < 512) payload.max_completion_tokens = 512;\n return { payload, messages, maxTokens, isClaude };\n}\n\n// ---- Handlers -------------------------------------------------------------\n\nexport async function handleModels(): Promise {\n return Response.json({\n object: \"list\",\n data: [\n { id: MODEL, object: \"model\", created: Math.floor(Date.now() / 1000), owned_by: \"val-town\" },\n ],\n });\n}\n\nexport async function handleRootInfo(): Promise {\n const now = Date.now();\n return Response.json({\n service: \"val-town-openai-gateway\",\n status: \"ok\",\n engine: \"std/openai (Val Town native AI)\",\n model: MODEL,\n base_url: \"/api/provider/v1\",\n auth: \"Bearer PROVIDER_API_KEY required on /v1/chat/completions\",\n limits: {\n requests_per_minute: 10,\n max_output_tokens: MAX_OUTPUT_TOKENS,\n expiry: new Date(now).toISOString(),\n },\n endpoints: [\"/health\", \"/v1/models\", \"/v1/chat/completions\"],\n });\n}\n\n/** Public config consumed by the UI (exposes the generated key). */\nexport async function config(): Promise> {\n const key = apiKey();\n return {\n model: MODEL,\n apiKey: key,\n keyRequired: key.length > 0,\n maskedKey: key.length > 6 ? \"sk-\" + key.slice(3, 7) + \"••••\" + key.slice(-4) : key,\n };\n}\n\nexport async function handleChatCompletions(req: Request): Promise {\n if (!isAuthed(req)) return unauthorized();\n\n let body: Record;\n try {\n body = await req.json();\n } catch {\n return Response.json(\n { error: { message: \"Invalid or missing JSON body\", type: \"invalid_request_error\", param: null, code: \"invalid_request\" } },\n { status: 400 },\n );\n }\n if (!Array.isArray(body.messages)) {\n return Response.json(\n { error: { message: \"Field `messages` must be an array\", type: \"invalid_request_error\", param: \"messages\", code: \"invalid_request\" } },\n { status: 400 },\n );\n }\n\n const stream = body.stream === true || body.stream === \"true\";\n const model = typeof body.model === \"string\" && body.model ? String(body.model) : MODEL;\n const { payload, isClaude } = buildPayload(body, model);\n\n if (isClaude) return claudeChat(req, payload, stream);\n if (stream) return streamChat(payload);\n return nonStreamChat(payload);\n}\n\n/**\n * Optional Anthropic bridge: routes claude-* models to Anthropic's\n * OpenAI-compatible endpoint. Requires ANTHROPIC_API_KEY (none provided by\n * the platform — Val Town's native engine only covers OpenAI models).\n */\nasync function claudeChat(req: Request, payload: Record, stream: boolean): Promise {\n const key = Deno.env.get(\"ANTHROPIC_API_KEY\");\n if (!key) {\n return Response.json(\n {\n error: {\n message:\n \"Claude models need the ANTHROPIC_API_KEY environment variable to be set on this gateway before they can run. Add it in the val's env settings to enable Claude.\",\n type: \"invalid_request_error\",\n param: null,\n code: \"missing_anthropic_key\",\n },\n },\n { status: 400 },\n );\n }\n const target = \"https://api.anthropic.com/v1/openai/chat/completions\";\n const outBody: Record = { model: payload.model, messages: payload.messages };\n outBody.max_tokens = payload.max_completion_tokens ?? 2048;\n if (typeof payload.temperature === \"number\") outBody.temperature = payload.temperature;\n if (stream) outBody.stream = true;\n try {\n const upstream = await fetch(target, {\n method: \"POST\",\n headers: { Authorization: `Bearer ${key}`, \"Content-Type\": \"application/json\" },\n body: JSON.stringify(outBody),\n // Propagate client abort to the upstream Anthropic call.\n signal: req.signal,\n });\n const ct = upstream.headers.get(\"content-type\") || \"application/json\";\n return new Response(upstream.body, { status: upstream.status, headers: { \"Content-Type\": ct } });\n } catch (e) {\n const m = e instanceof Error ? e.message : String(e);\n return Response.json(\n { error: { message: \"Claude upstream error: \" + m, type: \"upstream_error\", param: null, code: \"upstream_error\" } },\n { status: 502 },\n );\n }\n}\n\nasync function nonStreamChat(payload: Record): Promise {\n let completion: any;\n try {\n completion = await callEngine(payload);\n } catch (e) {\n const u = e as UpstreamError;\n return Response.json(\n { error: { message: u.message, type: \"upstream_error\", param: null, code: \"upstream_error\" } },\n { status: u.status || 502 },\n );\n }\n const choice = completion?.choices?.[0] ?? {};\n return Response.json({\n id: completion?.id || `chatcmpl-${Date.now()}`,\n object: \"chat.completion\",\n created: completion?.created ?? Math.floor(Date.now() / 1000),\n model: completion?.model || payload.model,\n choices: [\n {\n index: 0,\n message: { role: \"assistant\", content: choice?.message?.content ?? \"\" },\n finish_reason: choice?.finish_reason ?? \"stop\",\n },\n ],\n usage: completion?.usage ?? null,\n });\n}\n\nasync function streamChat(payload: Record): Promise {\n // Start the upstream stream; reset + retry once on transient/token errors.\n // The OpenAI SDK returns an async-iterable stream of chunks for stream:true.\n let upstream: AsyncIterable;\n try {\n upstream = await callEngine({ ...payload, stream: true }) as AsyncIterable;\n } catch (e) {\n const u = e as UpstreamError;\n return Response.json(\n { error: { message: u.message, type: \"upstream_error\", param: null, code: \"upstream_error\" } },\n { status: u.status || 502 },\n );\n }\n\n const encoder = new TextEncoder();\n const sysTime = Math.floor(Date.now() / 1000);\n const model = String(payload.model);\n\n const raw = new ReadableStream({\n async start(controller) {\n try {\n for await (const chunk of upstream) {\n const line = { id: chunk?.id || `chatcmpl-${sysTime}`, object: \"chat.completion.chunk\", created: chunk?.created ?? sysTime, model: chunk?.model || model, choices: chunk?.choices ?? [] };\n controller.enqueue(encoder.encode(\"data: \" + JSON.stringify(line) + \"\\n\\n\"));\n }\n controller.enqueue(encoder.encode(\"data: [DONE]\\n\\n\"));\n } catch (e) {\n const msg = e instanceof Error ? e.message : String(e);\n controller.enqueue(encoder.encode(\"data: \" + JSON.stringify({ error: { message: msg, type: \"stream_error\" } }) + \"\\n\\n\"));\n } finally {\n try { controller.close(); } catch { /* already closed */ }\n }\n },\n cancel() {\n // Client aborted — propagate to upstream if it supports cancellation.\n try { upstream.return?.(undefined); } catch { /* ignore */ }\n },\n });\n\n return new Response(raw, {\n headers: {\n \"Content-Type\": \"text/event-stream\",\n \"Cache-Control\": \"no-cache, no-transform\",\n \"X-Accel-Buffering\": \"no\",\n Connection: \"keep-alive\",\n },\n });\n}\n\n/** Top-level dispatcher for paths mounted under /api/provider/v1. */\nexport async function handleApi(req: Request, sub: string): Promise {\n const method = req.method;\n const path = sub || \"/\";\n\n if (path === \"/\" && method === \"GET\") return handleRootInfo();\n if (path === \"/models\" && method === \"GET\") return handleModels();\n\n if (path === \"/chat/completions\" && method === \"POST\") {\n return handleChatCompletions(req);\n }\n\n return Response.json(\n { error: { message: `No route matched \"${path}\" with method ${method}`, type: \"invalid_request_error\", param: null, code: \"not_found\" } },\n { status: 404 },\n );\n}\n","deleted",false,"label",[47,24],"lib","c5347651-cad0-468b-999b-ff32cc6090ac",["D",1788365801329],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","OpenAI-compatible gateway on Val Town native AI, no keys needed","imageUrl","userId","eddd5cb5-78be-4114-98c9-70030a02faea","pinnedFileId","aeb5b076-a6e9-11f1-8b7e-1607ee4eb77e","pinPreview","pinMigrated","screenshotVersion","likeCount",0,"orgId","01a062e4-8477-75ba-8a1d-30b57d9fb298","tursoName","tursoHostname","tursoMeasuredAt","pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_9":10,"_106":102},"org",{"_9":10,"_99":100,"_101":102,"_58":59,"_103":104,"_105":44},"branch",{"_37":89,"_11":90,"_22":30,"_91":48,"_92":-5,"_93":-5,"_94":-5,"_95":-5,"_96":-5,"_97":12,"_98":10},"isOwned",47,"main","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","displayName","ramy haroon","avatar","https://lh3.googleusercontent.com/a/ACg8ocLJ5VsuqFGypcgPz5AldEDau-BsilRB9eIzAaaz2tVM1ucp3w=s96-c","tier","pro","canRestrictVals","profileImageUrl","ctx",{"_82":-5,"_140":141,"_142":143,"_144":145,"_146":44,"_147":44,"_148":44,"_149":34,"_150":-5,"_151":-7,"_152":-7,"_153":66,"_154":44,"_155":44,"_156":-5,"_157":44,"_158":44,"_159":44,"_160":-5},"ENV",{"_111":112,"_113":114,"_115":90,"_116":117,"_118":119,"_120":121,"_122":123,"_124":125,"_126":127,"_128":129,"_130":131,"_132":133,"_134":135,"_136":137,"_138":139},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","e87eb04e707cf64e7e532c8e8ec395f209e17f15","RENDER_GIT_BRANCH","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","CLOUDFLARE_TURNSTILE_SITE_KEY","0x4AAAAAACtaIZJH7s9rBMZ6","GOOGLE_CLIENT_ID","875989696899-jtsqkhlbdv5jrlsk8631p7ipjg3c6o62.apps.googleusercontent.com","STRIPE_PUBLISHABLE_KEY","pk_live_51N2zvWEpZTTLipIkVEwXrUG9AT0yeOOEYxszarZT1jpsZV58Ly6wlg3SltZpHN26CUalOBmUSY7adQRQiSkCZz2500qSXNkf8Y","API_URL","https://api.val.town","orgs",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]