@premai/api-sdk 1.0.41 → 1.0.42

package/dist/core.browser.cjs

@@ -45,25 +45,127 @@ function getGatewayErrorMessage(err) {
     return err.kind.message;
   return null;
 }
-async function attest(apiKey, options = { enabled: true }) {
-  if (!options.enabled)
-    return null;
+var ATTEST_TTL_MS = 30000;
+var ATTEST_CACHE_MAX = 500;
+var ATTEST_MAX_ATTEMPTS = 4;
+var ATTEST_RETRY_BASE_MS = 250;
+var ATTEST_RETRY_MAX_MS = 2000;
+var TRANSIENT_PATTERNS = [
+  /EOF while parsing/i,
+  /error decoding response body/i,
+  /connection (reset|closed|refused)/i,
+  /socket hang up/i,
+  /ETIMEDOUT/i
+];
+var attestCache = new Map;
+var attestInflight = new Map;
+function attestCacheKey(apiKey, model) {
+  return `${apiKey}|${model ?? ""}`;
+}
+function pruneExpired(now) {
+  for (const [key, entry] of attestCache) {
+    if (entry.expires <= now) {
+      attestCache.delete(key);
+    } else {
+      break;
+    }
+  }
+}
+function isTransientError(err) {
+  const messages = [];
+  if (err instanceof Error) {
+    messages.push(err.message);
+  }
+  if (isAttestationError(err) && Array.isArray(err.cause)) {
+    messages.push(...err.cause);
+  }
+  return messages.some((m) => TRANSIENT_PATTERNS.some((re) => re.test(m)));
+}
+function backoffDelayMs(attempt) {
+  const exp = ATTEST_RETRY_BASE_MS * 2 ** (attempt - 1);
+  const capped = Math.min(exp, ATTEST_RETRY_MAX_MS);
+  const jitter = Math.floor(Math.random() * (capped / 2));
+  return capped + jitter;
+}
+function delay(ms) {
+  return new Promise((resolve) => setTimeout(resolve, ms));
+}
+function safeFree(obj) {
+  if (typeof obj?.free !== "function")
+    return;
+  try {
+    obj.free();
+  } catch {}
+}
+async function attemptAttest(apiKey, options) {
   const prem = await loadPrem();
-  const client = await new prem.ClientBuilder(endpoints.proxy ?? "").with_authorization(apiKey).build();
-  let query = new prem.QueryParams;
-  if (options.model)
-    query = query.with("model", options.model);
+  let client;
+  let attested;
+  let headers;
+  let sessionId;
   try {
-    client.set_query(query);
-    const attested = await client.attest();
-    const headers = attested.headers();
-    const sessionId = attested.headers().gpu()?.get("x-session-id") ?? null;
-    headers.free();
-    attested.free();
-    return sessionId;
+    client = await new prem.ClientBuilder(endpoints.proxy ?? "").with_authorization(apiKey).build();
+    if (options.model) {
+      client.set_query(new prem.QueryParams().with("model", options.model));
+    }
+    attested = await client.attest();
+    headers = attested.headers();
+    sessionId = headers.cpu()?.get("x-session-id") ?? headers.gpu()?.get("x-session-id") ?? null;
   } finally {
-    client.free();
+    safeFree(headers);
+    safeFree(attested);
+    safeFree(client);
+  }
+  if (sessionId === null) {
+    throw new Error("missing x-session-id issued by attestation");
+  }
+  return sessionId;
+}
+async function runAttest(apiKey, options) {
+  let lastErr;
+  for (let attempt = 1;attempt <= ATTEST_MAX_ATTEMPTS; attempt++) {
+    try {
+      return await attemptAttest(apiKey, options);
+    } catch (err) {
+      lastErr = err;
+      if (attempt === ATTEST_MAX_ATTEMPTS || !isTransientError(err)) {
+        throw err;
+      }
+      await delay(backoffDelayMs(attempt));
+    }
   }
+  throw lastErr;
+}
+async function attest(apiKey, options = { enabled: true }) {
+  if (!options.enabled)
+    return null;
+  const key = attestCacheKey(apiKey, options.model);
+  const now = Date.now();
+  const cached = attestCache.get(key);
+  if (cached) {
+    if (cached.expires > now)
+      return cached.sessionId;
+    attestCache.delete(key);
+  }
+  const inflight = attestInflight.get(key);
+  if (inflight) {
+    return inflight;
+  }
+  const work = runAttest(apiKey, options).then((sessionId) => {
+    const insertTime = Date.now();
+    pruneExpired(insertTime);
+    attestCache.set(key, { sessionId, expires: insertTime + ATTEST_TTL_MS });
+    if (attestCache.size > ATTEST_CACHE_MAX) {
+      const oldest = attestCache.keys().next().value;
+      if (oldest)
+        attestCache.delete(oldest);
+    }
+    return sessionId;
+  }).finally(() => {
+    attestInflight.delete(key);
+  });
+  attestInflight.set(key, work);
+  return work;
 }
 
 // src/utils/crypto.ts
@@ -1064,10 +1166,14 @@ function createRvencChatClient(apiKey, encryptionKeys, requestTimeoutMs = DEFAUL
       }
       clearTimeout(timeoutId);
       if (isStreaming) {
-        return await postprocessStreamingResponse(response, encryptedRequest.sharedSecret, encryptedRequest.nonce, maxBufferSize);
-      } else {
-        return await postprocessNonStreamingResponse(response, encryptedRequest.sharedSecret);
+        const contentType = response.headers.get("content-type") ?? "";
+        if (contentType.includes("text/event-stream")) {
+          return await postprocessStreamingResponse(response, encryptedRequest.sharedSecret, encryptedRequest.nonce, maxBufferSize);
+        }
+        const completion = await postprocessNonStreamingResponse(response, encryptedRequest.sharedSecret);
+        return completionToChunkStream(completion);
       }
+      return await postprocessNonStreamingResponse(response, encryptedRequest.sharedSecret);
     } catch (error) {
       clearTimeout(timeoutId);
       if (error instanceof Error && error.name === "AbortError") {
@@ -1078,6 +1184,39 @@ function createRvencChatClient(apiKey, encryptionKeys, requestTimeoutMs = DEFAUL
   };
   return client;
 }
+async function* completionToChunkStream(completion) {
+  const choice = completion.choices[0];
+  const message = choice?.message;
+  const content = typeof message?.content === "string" ? message.content : "";
+  const toolCalls = message?.tool_calls?.filter((tc) => tc.type === "function").map((tc, i) => ({
+    index: i,
+    id: tc.id,
+    type: "function",
+    function: {
+      name: tc.function.name,
+      arguments: tc.function.arguments
+    }
+  }));
+  yield {
+    id: completion.id,
+    object: "chat.completion.chunk",
+    created: completion.created,
+    model: completion.model,
+    choices: [
+      {
+        index: choice?.index ?? 0,
+        delta: {
+          role: "assistant",
+          content,
+          ...toolCalls && toolCalls.length > 0 && { tool_calls: toolCalls }
+        },
+        finish_reason: choice?.finish_reason ?? "stop",
+        logprobs: null
+      }
+    ],
+    usage: completion.usage ?? null
+  };
+}
 async function* createDecryptedStreamGenerator(reader, sharedSecret, nonce, maxBufferSize) {
   const decoder = new TextDecoder;
   let buffer = "";
@@ -1258,7 +1397,7 @@ async function callFileOutputTool(toolName, params, apiKey, dekStore, clientKEK,
   };
   const response = await callToolRequest(toolName, body, apiKey, timeoutMs, attest2);
   const result = await downloadAndDecryptFile(response, dek, apiKey, timeoutMs);
-  if (result && result.fileId) {
+  if (result?.fileId) {
     if (!dekStore.fileDEKs) {
       dekStore.fileDEKs = new Map;
     }
@@ -1315,7 +1454,7 @@ async function callRagTool(toolName, params, apiKey, dekStore, clientKEK, timeou
   }
   const _clientKEK = clientKEK ? hexToBytes6(clientKEK) : getClientKEK();
   const encryptedFileDEKs = fileIds.reduce((acc, fileId) => {
-    const fileDEK = dekStore.fileDEKs.get(fileId);
+    const fileDEK = dekStore.fileDEKs?.get(fileId);
     if (!fileDEK) {
       return acc;
     }

package/dist/core.browser.mjs

@@ -276,25 +276,127 @@ function getGatewayErrorMessage(err) {
     return err.kind.message;
   return null;
 }
-async function attest(apiKey, options = { enabled: true }) {
-  if (!options.enabled)
-    return null;
+var ATTEST_TTL_MS = 30000;
+var ATTEST_CACHE_MAX = 500;
+var ATTEST_MAX_ATTEMPTS = 4;
+var ATTEST_RETRY_BASE_MS = 250;
+var ATTEST_RETRY_MAX_MS = 2000;
+var TRANSIENT_PATTERNS = [
+  /EOF while parsing/i,
+  /error decoding response body/i,
+  /connection (reset|closed|refused)/i,
+  /socket hang up/i,
+  /ETIMEDOUT/i
+];
+var attestCache = new Map;
+var attestInflight = new Map;
+function attestCacheKey(apiKey, model) {
+  return `${apiKey}|${model ?? ""}`;
+}
+function pruneExpired(now) {
+  for (const [key, entry] of attestCache) {
+    if (entry.expires <= now) {
+      attestCache.delete(key);
+    } else {
+      break;
+    }
+  }
+}
+function isTransientError(err) {
+  const messages = [];
+  if (err instanceof Error) {
+    messages.push(err.message);
+  }
+  if (isAttestationError(err) && Array.isArray(err.cause)) {
+    messages.push(...err.cause);
+  }
+  return messages.some((m) => TRANSIENT_PATTERNS.some((re) => re.test(m)));
+}
+function backoffDelayMs(attempt) {
+  const exp = ATTEST_RETRY_BASE_MS * 2 ** (attempt - 1);
+  const capped = Math.min(exp, ATTEST_RETRY_MAX_MS);
+  const jitter = Math.floor(Math.random() * (capped / 2));
+  return capped + jitter;
+}
+function delay(ms) {
+  return new Promise((resolve) => setTimeout(resolve, ms));
+}
+function safeFree(obj) {
+  if (typeof obj?.free !== "function")
+    return;
+  try {
+    obj.free();
+  } catch {}
+}
+async function attemptAttest(apiKey, options) {
   const prem = await loadPrem();
-  const client = await new prem.ClientBuilder(endpoints.proxy ?? "").with_authorization(apiKey).build();
-  let query = new prem.QueryParams;
-  if (options.model)
-    query = query.with("model", options.model);
+  let client;
+  let attested;
+  let headers;
+  let sessionId;
   try {
-    client.set_query(query);
-    const attested = await client.attest();
-    const headers = attested.headers();
-    const sessionId = attested.headers().gpu()?.get("x-session-id") ?? null;
-    headers.free();
-    attested.free();
-    return sessionId;
+    client = await new prem.ClientBuilder(endpoints.proxy ?? "").with_authorization(apiKey).build();
+    if (options.model) {
+      client.set_query(new prem.QueryParams().with("model", options.model));
+    }
+    attested = await client.attest();
+    headers = attested.headers();
+    sessionId = headers.cpu()?.get("x-session-id") ?? headers.gpu()?.get("x-session-id") ?? null;
   } finally {
-    client.free();
+    safeFree(headers);
+    safeFree(attested);
+    safeFree(client);
+  }
+  if (sessionId === null) {
+    throw new Error("missing x-session-id issued by attestation");
   }
+  return sessionId;
+}
+async function runAttest(apiKey, options) {
+  let lastErr;
+  for (let attempt = 1;attempt <= ATTEST_MAX_ATTEMPTS; attempt++) {
+    try {
+      return await attemptAttest(apiKey, options);
+    } catch (err) {
+      lastErr = err;
+      if (attempt === ATTEST_MAX_ATTEMPTS || !isTransientError(err)) {
+        throw err;
+      }
+      await delay(backoffDelayMs(attempt));
+    }
+  }
+  throw lastErr;
+}
+async function attest(apiKey, options = { enabled: true }) {
+  if (!options.enabled)
+    return null;
+  const key = attestCacheKey(apiKey, options.model);
+  const now = Date.now();
+  const cached = attestCache.get(key);
+  if (cached) {
+    if (cached.expires > now)
+      return cached.sessionId;
+    attestCache.delete(key);
+  }
+  const inflight = attestInflight.get(key);
+  if (inflight) {
+    return inflight;
+  }
+  const work = runAttest(apiKey, options).then((sessionId) => {
+    const insertTime = Date.now();
+    pruneExpired(insertTime);
+    attestCache.set(key, { sessionId, expires: insertTime + ATTEST_TTL_MS });
+    if (attestCache.size > ATTEST_CACHE_MAX) {
+      const oldest = attestCache.keys().next().value;
+      if (oldest)
+        attestCache.delete(oldest);
+    }
+    return sessionId;
+  }).finally(() => {
+    attestInflight.delete(key);
+  });
+  attestInflight.set(key, work);
+  return work;
 }
 
 // node_modules/@noble/ciphers/aes.js
@@ -25335,10 +25437,14 @@ function createRvencChatClient(apiKey, encryptionKeys, requestTimeoutMs = DEFAUL
       }
       clearTimeout(timeoutId);
       if (isStreaming) {
-        return await postprocessStreamingResponse(response, encryptedRequest.sharedSecret, encryptedRequest.nonce, maxBufferSize);
-      } else {
-        return await postprocessNonStreamingResponse(response, encryptedRequest.sharedSecret);
+        const contentType = response.headers.get("content-type") ?? "";
+        if (contentType.includes("text/event-stream")) {
+          return await postprocessStreamingResponse(response, encryptedRequest.sharedSecret, encryptedRequest.nonce, maxBufferSize);
+        }
+        const completion = await postprocessNonStreamingResponse(response, encryptedRequest.sharedSecret);
+        return completionToChunkStream(completion);
       }
+      return await postprocessNonStreamingResponse(response, encryptedRequest.sharedSecret);
     } catch (error47) {
       clearTimeout(timeoutId);
       if (error47 instanceof Error && error47.name === "AbortError") {
@@ -25349,6 +25455,39 @@ function createRvencChatClient(apiKey, encryptionKeys, requestTimeoutMs = DEFAUL
   };
   return client;
 }
+async function* completionToChunkStream(completion) {
+  const choice = completion.choices[0];
+  const message = choice?.message;
+  const content = typeof message?.content === "string" ? message.content : "";
+  const toolCalls = message?.tool_calls?.filter((tc) => tc.type === "function").map((tc, i) => ({
+    index: i,
+    id: tc.id,
+    type: "function",
+    function: {
+      name: tc.function.name,
+      arguments: tc.function.arguments
+    }
+  }));
+  yield {
+    id: completion.id,
+    object: "chat.completion.chunk",
+    created: completion.created,
+    model: completion.model,
+    choices: [
+      {
+        index: choice?.index ?? 0,
+        delta: {
+          role: "assistant",
+          content,
+          ...toolCalls && toolCalls.length > 0 && { tool_calls: toolCalls }
+        },
+        finish_reason: choice?.finish_reason ?? "stop",
+        logprobs: null
+      }
+    ],
+    usage: completion.usage ?? null
+  };
+}
 async function* createDecryptedStreamGenerator(reader, sharedSecret, nonce, maxBufferSize) {
   const decoder = new TextDecoder;
   let buffer = "";
@@ -25528,7 +25667,7 @@ async function callFileOutputTool(toolName, params, apiKey, dekStore, clientKEK,
   };
   const response = await callToolRequest(toolName, body, apiKey, timeoutMs, attest2);
   const result = await downloadAndDecryptFile(response, dek, apiKey, timeoutMs);
-  if (result && result.fileId) {
+  if (result?.fileId) {
     if (!dekStore.fileDEKs) {
       dekStore.fileDEKs = new Map;
     }
@@ -25585,7 +25724,7 @@ async function callRagTool(toolName, params, apiKey, dekStore, clientKEK, timeou
   }
   const _clientKEK = clientKEK ? hexToBytes(clientKEK) : getClientKEK();
   const encryptedFileDEKs = fileIds.reduce((acc, fileId) => {
-    const fileDEK = dekStore.fileDEKs.get(fileId);
+    const fileDEK = dekStore.fileDEKs?.get(fileId);
     if (!fileDEK) {
       return acc;
     }