@anvaka/vue-llm 0.3.1 → 0.3.2

package/dist/pricing/rates.js

@@ -26,6 +26,7 @@ const t = {
     "gpt-5-codex": { input: 1.25, output: 10, cachedInput: 0.125 }
   },
   anthropic: {
+    "claude-opus-4-8": { input: 5, output: 25, cachedInput: 0.5, cacheCreation: 6.25 },
     "claude-opus-4-7": { input: 5, output: 25, cachedInput: 0.5, cacheCreation: 6.25 },
     "claude-opus-4-6": { input: 5, output: 25, cachedInput: 0.5, cacheCreation: 6.25 },
     "claude-sonnet-4-6": { input: 3, output: 15, cachedInput: 0.3, cacheCreation: 3.75 },
@@ -35,6 +36,7 @@ const t = {
   // Bedrock charges the same per-token rates as Anthropic direct for the
   // Claude inference profiles; the model IDs differ (us.anthropic.* prefix).
   bedrock: {
+    "us.anthropic.claude-opus-4-8": { input: 5, output: 25, cachedInput: 0.5, cacheCreation: 6.25 },
     "us.anthropic.claude-opus-4-7": { input: 5, output: 25, cachedInput: 0.5, cacheCreation: 6.25 },
     "us.anthropic.claude-opus-4-6": { input: 5, output: 25, cachedInput: 0.5, cacheCreation: 6.25 },
     "us.anthropic.claude-sonnet-4-6": { input: 3, output: 15, cachedInput: 0.3, cacheCreation: 3.75 },

package/dist/providers/AnthropicProvider.js

@@ -1,19 +1,19 @@
 import { BaseProvider as g } from "./BaseProvider.js";
-class C extends g {
+class b extends g {
   async detectCapabilities() {
     var t, e, o, s;
     ((t = this.config.model) != null && t.includes("claude-3") || (e = this.config.model) != null && e.includes("claude-sonnet") || (o = this.config.model) != null && o.includes("claude-opus") || (s = this.config.model) != null && s.includes("claude-haiku")) && (this.capabilities.add("vision"), this.capabilities.add("tools"));
   }
   prepareRequest(t, e) {
-    const o = _(t), s = e.model || this.config.model || "claude-3-sonnet-20240229", r = {
+    const o = m(t), s = e.model || this.config.model || "claude-3-sonnet-20240229", r = {
       model: s,
       max_tokens: e.maxTokens || 1e3,
       messages: o,
       stream: e.stream || !1
     };
-    s.includes("claude-opus-4-7") || (r.temperature = e.temperature ?? 0.7);
+    _(s) || (r.temperature = e.temperature ?? 0.7);
     const c = t.find((i) => i.role === "system");
-    return c && (r.system = c.content), e.tools && this.capabilities.has("tools") && (r.tools = m(e.tools), e.tool_choice && (r.tool_choice = e.tool_choice)), this.promptCachingEnabled(e) && y(r, e), r;
+    return c && (r.system = c.content), e.tools && this.capabilities.has("tools") && (r.tools = y(e.tools), e.tool_choice && (r.tool_choice = e.tool_choice)), this.promptCachingEnabled(e) && k(r, e), r;
   }
   // Prompt caching is on by default for the Claude/Anthropic-family path
   // (Bedrock inherits this method) because it's strictly cheaper and degrades
@@ -106,6 +106,9 @@ class C extends g {
     return ((e = t.data) == null ? void 0 : e.map((o) => o.id)) || [];
   }
 }
+function _(n) {
+  return /claude-opus-4-[789]/.test(n);
+}
 function h(n) {
   if (!n) return null;
   const t = String(n).toLowerCase();
@@ -116,7 +119,7 @@ function f(n) {
   const t = n.input_tokens ?? 0, e = n.cache_read_input_tokens ?? 0, o = n.cache_creation_input_tokens ?? 0, r = n.input_tokens != null || n.cache_read_input_tokens != null || n.cache_creation_input_tokens != null ? t + e + o : 0, c = n.output_tokens ?? 0, i = { inputTokens: r, outputTokens: c, totalTokens: r + c, raw: n };
   return n.cache_read_input_tokens != null && (i.cachedInputTokens = e), n.cache_creation_input_tokens != null && (i.cacheCreationInputTokens = o), i;
 }
-function _(n) {
+function m(n) {
   const t = [];
   for (const e of n)
     if (e.role !== "system") {
@@ -148,7 +151,7 @@ function _(n) {
     }
   return t;
 }
-function m(n) {
+function y(n) {
   return n.map((t) => t.type === "function" && t.function ? {
     name: t.function.name,
     description: t.function.description,
@@ -156,9 +159,9 @@ function m(n) {
   } : t);
 }
 const l = { type: "ephemeral" };
-function y(n, t) {
+function k(n, t) {
   if (n.system)
-    n.system = k(n.system);
+    n.system = x(n.system);
   else if (Array.isArray(n.tools) && n.tools.length) {
     const e = n.tools.length - 1;
     n.tools[e] = { ...n.tools[e], cache_control: l };
@@ -167,12 +170,12 @@ function y(n, t) {
     const e = n.messages;
     if (Array.isArray(e) && e.length) {
       const o = e[e.length - 1];
-      o.content = x(o.content);
+      o.content = A(o.content);
     }
   }
   return n;
 }
-function k(n) {
+function x(n) {
   let t;
   if (typeof n == "string")
     t = [{ type: "text", text: n }];
@@ -182,7 +185,7 @@ function k(n) {
     return n;
   return t[t.length - 1] = { ...t[t.length - 1], cache_control: l }, t;
 }
-function x(n) {
+function A(n) {
   if (typeof n == "string")
     return [{ type: "text", text: n, cache_control: l }];
   if (Array.isArray(n) && n.length) {
@@ -192,6 +195,7 @@ function x(n) {
   return n;
 }
 export {
-  C as AnthropicProvider,
-  f as normalizeAnthropicUsage
+  b as AnthropicProvider,
+  f as normalizeAnthropicUsage,
+  _ as samplingParamsRemoved
 };

package/dist/providers/BedrockProvider.js

@@ -1,6 +1,7 @@
 import { BaseProvider as y } from "./BaseProvider.js";
 import { AnthropicProvider as w } from "./AnthropicProvider.js";
-const g = [
+const p = [
+  "us.anthropic.claude-opus-4-8",
   "us.anthropic.claude-opus-4-7",
   "us.anthropic.claude-sonnet-4-6",
   "us.anthropic.claude-haiku-4-5-20251001-v1:0",
@@ -9,10 +10,10 @@ const g = [
   "us.anthropic.claude-sonnet-4-5-20250929-v1:0",
   "us.anthropic.claude-opus-4-1-20250805-v1:0"
 ];
-class R extends w {
-  prepareRequest(r, n) {
-    const t = super.prepareRequest(r, n);
-    return delete t.stream, t.anthropic_version = "bedrock-2023-05-31", t;
+class A extends w {
+  prepareRequest(n, t) {
+    const r = super.prepareRequest(n, t);
+    return delete r.stream, r.anthropic_version = "bedrock-2023-05-31", r;
   }
   buildHeaders() {
     return y.prototype.buildHeaders.call(this);
@@ -26,147 +27,170 @@ class R extends w {
   requiresAuth() {
     return !!this.config.apiKey;
   }
-  async makeRequest(r, n, t) {
-    const e = r.model, d = { ...r };
+  async makeRequest(n, t, r) {
+    const e = n.model, d = { ...n };
     delete d.model;
-    const c = `${this.config.baseUrl}/model/${encodeURIComponent(e)}/invoke`, u = n ? { signal: n } : new AbortController();
-    n || (t = t || this.generateRequestId(), this.activeRequests.set(t, u));
+    const i = `${this.config.baseUrl}/model/${encodeURIComponent(e)}/invoke`, l = t ? { signal: t } : new AbortController();
+    t || (r = r || this.generateRequestId(), this.activeRequests.set(r, l));
     try {
-      const s = await fetch(c, {
+      const o = await fetch(i, {
         method: "POST",
         headers: this.buildHeaders(),
         body: JSON.stringify(d),
-        signal: u.signal || n
+        signal: l.signal || t
       });
-      if (!s.ok) {
-        const l = await s.text();
-        throw new Error(`Bedrock API Error (${s.status}): ${l}`);
+      if (!o.ok) {
+        const h = await o.text();
+        throw new Error(`Bedrock API Error (${o.status}): ${h}`);
       }
-      return s.json();
+      return o.json();
     } finally {
-      !n && t && this.activeRequests.delete(t);
+      !t && r && this.activeRequests.delete(r);
     }
   }
-  async makeStreamingRequest(r, n) {
-    const t = r.model, e = { ...r };
+  async makeStreamingRequest(n, t) {
+    const r = n.model, e = { ...n };
     delete e.model;
-    const d = `${this.config.baseUrl}/model/${encodeURIComponent(t)}/invoke-with-response-stream`, c = await fetch(d, {
+    const d = `${this.config.baseUrl}/model/${encodeURIComponent(r)}/invoke-with-response-stream`, i = await fetch(d, {
       method: "POST",
       headers: this.buildHeaders(),
       body: JSON.stringify(e),
-      signal: n
+      signal: t
     });
-    if (!c.ok) {
-      const i = await c.text();
-      throw new Error(`Bedrock API Error (${c.status}): ${i}`);
+    if (!i.ok) {
+      const c = await i.text();
+      throw new Error(`Bedrock API Error (${i.status}): ${c}`);
     }
-    const u = c.body.getReader(), s = new v(), l = new TextEncoder(), f = new ReadableStream({
-      async pull(i) {
+    const l = i.body.getReader(), o = new v(), h = new TextEncoder(), f = new ReadableStream({
+      async pull(c) {
         try {
-          const { done: a, value: m } = await u.read();
+          const { done: a, value: g } = await l.read();
           if (a) {
-            i.close();
+            c.close();
             return;
           }
-          const p = s.feed(m);
-          for (const h of p) {
-            if (h.kind === "error") {
-              i.error(new Error(h.message));
+          const m = o.feed(g);
+          for (const u of m) {
+            if (u.kind === "error") {
+              c.error(new Error(u.message));
               return;
             }
-            i.enqueue(l.encode(`data: ${JSON.stringify(h.payload)}
+            c.enqueue(h.encode(`data: ${JSON.stringify(u.payload)}
 `));
           }
         } catch (a) {
-          i.error(a);
+          c.error(a);
         }
       },
-      cancel(i) {
+      cancel(c) {
         try {
-          u.cancel(i);
+          l.cancel(c);
         } catch {
         }
       }
     });
     return new Response(f, { status: 200, headers: { "Content-Type": "text/event-stream" } });
   }
-  // Bedrock control plane needs SigV4; Bearer auth can't list models.
-  // Return the hardcoded list so the dropdown still populates.
-  async discoverModels() {
-    return g.slice();
+  // Live discovery is opt-in via config.liveModelDiscovery. A Bedrock API key
+  // authenticates against the control plane with Bearer auth (no SigV4), and
+  // the control-plane host returns permissive CORS headers, so a browser fetch
+  // works. When disabled (default) or on any failure, fall back to the static
+  // list — existing apps keep working with zero surprise network calls.
+  async discoverModels(n = 1e4) {
+    if (!this.config.liveModelDiscovery || !this.getModelsEndpoint())
+      return p.slice();
+    try {
+      const t = await super.discoverModels(n);
+      return t.length ? t : p.slice();
+    } catch {
+      return p.slice();
+    }
   }
+  // The control plane lives at the same host with the `-runtime` segment
+  // stripped: bedrock-runtime.{region}.amazonaws.com (invoke) ->
+  // bedrock.{region}.amazonaws.com (ListInferenceProfiles). Returns null when
+  // no distinct control-plane host can be derived (custom/relative baseUrl),
+  // which disables live discovery and leaves the static fallback in place.
   getModelsEndpoint() {
-    return null;
+    const n = this.config.baseUrl || "", t = n.replace("bedrock-runtime.", "bedrock.");
+    return !t || t === n ? null : `${t}/inference-profiles`;
   }
-  parseModelsResponse() {
-    return g.slice();
+  // ListInferenceProfiles returns
+  //   { inferenceProfileSummaries: [{ inferenceProfileId, ... }] }
+  // Keep only Anthropic Claude profiles (us.* and global.*) — those ids are
+  // exactly what InvokeModel expects in the URL path.
+  parseModelsResponse(n) {
+    const t = n == null ? void 0 : n.inferenceProfileSummaries;
+    if (!Array.isArray(t)) return p.slice();
+    const r = t.map((e) => e == null ? void 0 : e.inferenceProfileId).filter((e) => typeof e == "string" && e.includes("anthropic.claude"));
+    return r.length ? r : p.slice();
   }
 }
 class v {
   constructor() {
     this.buf = new Uint8Array(0), this.textDecoder = new TextDecoder();
   }
-  feed(r) {
-    if (r && r.length) {
-      const t = new Uint8Array(this.buf.length + r.length);
-      t.set(this.buf, 0), t.set(r, this.buf.length), this.buf = t;
+  feed(n) {
+    if (n && n.length) {
+      const r = new Uint8Array(this.buf.length + n.length);
+      r.set(this.buf, 0), r.set(n, this.buf.length), this.buf = r;
     }
-    const n = [];
+    const t = [];
     for (; this.buf.length >= 12; ) {
-      const t = new DataView(this.buf.buffer, this.buf.byteOffset, this.buf.byteLength), e = t.getUint32(0, !1), d = t.getUint32(4, !1);
+      const r = new DataView(this.buf.buffer, this.buf.byteOffset, this.buf.byteLength), e = r.getUint32(0, !1), d = r.getUint32(4, !1);
       if (e < 16 || e > 16 * 1024 * 1024)
         throw new Error(`Bedrock event stream: invalid frame length ${e}`);
       if (this.buf.length < e) break;
-      const c = 12, u = c + d, s = u, l = e - 4, f = k(this.buf.subarray(c, u)), i = this.textDecoder.decode(this.buf.subarray(s, l));
+      const i = 12, l = i + d, o = l, h = e - 4, f = k(this.buf.subarray(i, l)), c = this.textDecoder.decode(this.buf.subarray(o, h));
       this.buf = this.buf.subarray(e);
       let a;
       try {
-        a = JSON.parse(i);
+        a = JSON.parse(c);
       } catch {
         a = null;
       }
-      const m = f[":message-type"], p = f[":exception-type"];
-      if (m === "exception" || p) {
-        const h = p || f[":event-type"] || "BedrockStreamException", b = (a == null ? void 0 : a.message) || i || "unknown error";
-        n.push({ kind: "error", message: `${h}: ${b}` });
+      const g = f[":message-type"], m = f[":exception-type"];
+      if (g === "exception" || m) {
+        const u = m || f[":event-type"] || "BedrockStreamException", b = (a == null ? void 0 : a.message) || c || "unknown error";
+        t.push({ kind: "error", message: `${u}: ${b}` });
         continue;
       }
       if (a && typeof a.bytes == "string")
         try {
-          const h = x(a.bytes), b = JSON.parse(h);
-          n.push({ kind: "event", payload: b });
-        } catch (h) {
-          n.push({ kind: "error", message: `Bedrock event-stream decode failed: ${h.message}` });
+          const u = x(a.bytes), b = JSON.parse(u);
+          t.push({ kind: "event", payload: b });
+        } catch (u) {
+          t.push({ kind: "error", message: `Bedrock event-stream decode failed: ${u.message}` });
         }
     }
-    return n;
+    return t;
   }
 }
-function k(o) {
-  const r = {}, n = new DataView(o.buffer, o.byteOffset, o.byteLength), t = new TextDecoder();
+function k(s) {
+  const n = {}, t = new DataView(s.buffer, s.byteOffset, s.byteLength), r = new TextDecoder();
   let e = 0;
-  for (; e < o.length; ) {
-    const d = o[e];
-    if (e += 1, e + d > o.length) break;
-    const c = t.decode(o.subarray(e, e + d));
-    if (e += d, e >= o.length) break;
-    const u = o[e];
-    if (e += 1, u === 7) {
-      if (e + 2 > o.length) break;
-      const s = n.getUint16(e, !1);
-      if (e += 2, e + s > o.length) break;
-      r[c] = t.decode(o.subarray(e, e + s)), e += s;
+  for (; e < s.length; ) {
+    const d = s[e];
+    if (e += 1, e + d > s.length) break;
+    const i = r.decode(s.subarray(e, e + d));
+    if (e += d, e >= s.length) break;
+    const l = s[e];
+    if (e += 1, l === 7) {
+      if (e + 2 > s.length) break;
+      const o = t.getUint16(e, !1);
+      if (e += 2, e + o > s.length) break;
+      n[i] = r.decode(s.subarray(e, e + o)), e += o;
     } else
       break;
   }
-  return r;
+  return n;
 }
-function x(o) {
-  const r = atob(o), n = new Uint8Array(r.length);
-  for (let t = 0; t < r.length; t++) n[t] = r.charCodeAt(t);
-  return new TextDecoder().decode(n);
+function x(s) {
+  const n = atob(s), t = new Uint8Array(n.length);
+  for (let r = 0; r < n.length; r++) t[r] = n.charCodeAt(r);
+  return new TextDecoder().decode(t);
 }
 export {
-  g as BEDROCK_CLAUDE_MODELS,
-  R as BedrockProvider
+  p as BEDROCK_CLAUDE_MODELS,
+  A as BedrockProvider
 };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@anvaka/vue-llm",
-  "version": "0.3.1",
+  "version": "0.3.2",
   "description": "Browser-only LLM client with provider adapters and Vue 3 components",
   "type": "module",
   "license": "MIT",