@agentproto/catalog-sync 0.3.0 → 0.5.0

package/README.md

@@ -66,3 +66,18 @@ Source: `https://openrouter.ai/api/v1/models` (pinned snapshot:
   (P3 spec: "plus cache fields if the source has them") — additive and
   `LLMPricing`-compatible.
 - Unpriced routes (`openrouter/auto`) are skipped.
+
+### `llm:context-windows`
+
+Sources: Anthropic, Groq, xAI, and Moonshot models-list endpoints (pinned
+snapshots: `snapshots/llm-anthropic.json`, `snapshots/llm-groq.json`,
+`snapshots/llm-xai.json`, `snapshots/llm-moonshot.json`). Emits
+`packages/model-catalog/src/llm/context-windows.generated.ts` — a drop-in
+`CONTEXT_WINDOWS: Record<string, ContextWindowEntry>`.
+
+- `contextWindow` maps to each provider's max-input/context-length field.
+- `maxOutput` is captured when the source publishes one.
+- This is **not pricing**; billing data still comes from `llm:openrouter` and
+  hand-maintained pricing entries in `model-catalog`.
+- OpenAI and ZAI/Zhipu are excluded because they don't publish a stable,
+  machine-readable context-window list.

package/dist/bin.mjs

@@ -133,6 +133,252 @@ var llmOpenRouterGenerator = defineGenerator({
   sources,
   generate
 });
+var REQUESTY_MODELS_URL = "https://router.requesty.ai/v1/models";
+var OUTPUT_PATH2 = "packages/model-catalog/src/llm/requesty-routes.generated.ts";
+var ModelSchema2 = z.object({
+  id: z.string(),
+  input_price: z.number().optional(),
+  output_price: z.number().optional(),
+  cached_price: z.number().optional(),
+  supports_caching: z.boolean().optional()
+}).passthrough();
+var SnapshotSchema2 = z.object({
+  data: z.array(ModelSchema2)
+});
+function round62(n) {
+  return Math.round(n * 1e6) / 1e6;
+}
+function per1m2(tokenPrice) {
+  if (tokenPrice === void 0) return void 0;
+  if (!Number.isFinite(tokenPrice)) return void 0;
+  return round62(tokenPrice * 1e6);
+}
+function fmt2(n) {
+  if (!Number.isFinite(n)) return "0";
+  return `${round62(n)}`;
+}
+function vendorOf2(id) {
+  if (!id.includes("/")) return id;
+  return id.split("/")[0] ?? id;
+}
+function serializeEntry2(e) {
+  const fields = [
+    `inputPer1M: ${fmt2(e.inputPer1M)}`,
+    `outputPer1M: ${fmt2(e.outputPer1M)}`
+  ];
+  if (e.cacheReadMultiplier !== void 0) {
+    fields.push(`cacheReadMultiplier: ${fmt2(e.cacheReadMultiplier)}`);
+  }
+  fields.push(`vendor: ${JSON.stringify(e.vendor)}`);
+  fields.push(`provider: ${JSON.stringify(e.provider)}`);
+  return `{
+    ${fields.join(",\n    ")},
+  }`;
+}
+function serializeFile2(entries) {
+  const ids = Object.keys(entries).sort();
+  const providers = Array.from(new Set(ids.map(vendorOf2))).sort();
+  const lines = [
+    "// AUTO-GENERATED by @agentproto/catalog-sync (llm:requesty).",
+    "// Do not edit by hand \u2014 re-run `pnpm --filter @agentproto/catalog-sync generate`.",
+    `// Source: ${REQUESTY_MODELS_URL}`,
+    "// Pricing carries provider USD (inputPer1M / outputPer1M) plus a cache",
+    "// multiplier (cacheReadMultiplier) derived from the source's cached_price",
+    "// per-token field when `supports_caching` is true. Requesty has no",
+    "// cache-write price, so cacheWriteMultiplier is never emitted.",
+    "",
+    'import type { LLMPricing } from "./catalog.js"',
+    "",
+    "export const REQUESTY_ROUTES: Record<string, LLMPricing> = {"
+  ];
+  for (const id of ids) {
+    lines.push(`  ${JSON.stringify(id)}: ${serializeEntry2(entries[id])},`);
+  }
+  lines.push("}");
+  lines.push("");
+  lines.push("/** Provider slugs treated as Requesty routes by the picker. */");
+  lines.push("export const REQUESTY_PROVIDERS: readonly string[] = [");
+  for (const p of providers) {
+    lines.push(`  ${JSON.stringify(p)},`);
+  }
+  lines.push("]");
+  lines.push("");
+  return lines.join("\n");
+}
+async function generate2(ctx) {
+  const src = sources2[0];
+  if (!src) throw new Error("llm:requesty: no source configured");
+  const parsed = SnapshotSchema2.parse(await ctx.fetchSource(src));
+  const entries = {};
+  for (const model of parsed.data) {
+    const inputPer1M = per1m2(model.input_price);
+    const outputPer1M = per1m2(model.output_price);
+    if (inputPer1M === void 0 || outputPer1M === void 0) continue;
+    if (inputPer1M === 0 && outputPer1M === 0) continue;
+    const entry = {
+      inputPer1M,
+      outputPer1M,
+      vendor: vendorOf2(model.id),
+      provider: "requesty"
+    };
+    if (model.supports_caching === true && inputPer1M > 0) {
+      const cachePer1M = per1m2(model.cached_price);
+      if (cachePer1M !== void 0) {
+        entry.cacheReadMultiplier = round62(cachePer1M / inputPer1M);
+      }
+    }
+    entries[model.id] = entry;
+  }
+  return { [OUTPUT_PATH2]: serializeFile2(entries) };
+}
+var sources2 = [
+  {
+    id: "llm-requesty",
+    url: REQUESTY_MODELS_URL,
+    // Live refresh needs the account key (Bearer). Without REQUESTY_API_KEY
+    // set, the framework reuses the committed snapshot.
+    headers: { Authorization: "Bearer env:REQUESTY_API_KEY" }
+  }
+];
+var llmRequestyGenerator = defineGenerator({
+  name: "llm:requesty",
+  modality: "llm",
+  sources: sources2,
+  generate: generate2
+});
+var OUTPUT_PATH3 = "packages/model-catalog/src/llm/context-windows.generated.ts";
+var AnthropicModelSchema = z.object({
+  id: z.string(),
+  display_name: z.string().optional(),
+  max_input_tokens: z.number().optional(),
+  max_tokens: z.number().optional()
+}).passthrough();
+var AnthropicSnapshotSchema = z.object({ data: z.array(AnthropicModelSchema) });
+var GroqModelSchema = z.object({
+  id: z.string(),
+  context_window: z.number().optional(),
+  max_completion_tokens: z.number().optional()
+}).passthrough();
+var GroqSnapshotSchema = z.object({ data: z.array(GroqModelSchema) });
+var XaiModelSchema = z.object({
+  id: z.string(),
+  context_length: z.number().nullable().optional()
+}).passthrough();
+var XaiSnapshotSchema = z.object({ data: z.array(XaiModelSchema) });
+var MoonshotModelSchema = z.object({
+  id: z.string(),
+  context_length: z.number().optional()
+}).passthrough();
+var MoonshotSnapshotSchema = z.object({ data: z.array(MoonshotModelSchema) });
+function serializeEntry3(e) {
+  const fields = [`contextWindow: ${e.contextWindow}`];
+  if (e.maxOutput !== void 0) fields.push(`maxOutput: ${e.maxOutput}`);
+  if (e.displayName !== void 0) fields.push(`displayName: ${JSON.stringify(e.displayName)}`);
+  fields.push(`provider: ${JSON.stringify(e.provider)}`);
+  return `{ ${fields.join(", ")} }`;
+}
+function serializeFile3(entries) {
+  const ids = Object.keys(entries).sort();
+  const lines = [
+    "// AUTO-GENERATED by @agentproto/catalog-sync (llm:context-windows).",
+    "// Do not edit by hand \u2014 re-run `pnpm --filter @agentproto/catalog-sync generate --refresh`.",
+    "// Sources: api.anthropic.com/v1/models, api.groq.com/openai/v1/models,",
+    "//         api.x.ai/v1/models, api.moonshot.ai/v1/models",
+    "// Context window (max input tokens) + max output tokens per live model id.",
+    "// NOT pricing \u2014 see LLM_PRICING_CATALOG / OPENROUTER_ROUTES in catalog.ts.",
+    "",
+    "export interface ContextWindowEntry {",
+    "  contextWindow: number",
+    "  maxOutput?: number",
+    "  displayName?: string",
+    '  provider: "anthropic" | "groq" | "xai" | "moonshot"',
+    "}",
+    "",
+    "export const CONTEXT_WINDOWS: Record<string, ContextWindowEntry> = {"
+  ];
+  for (const id of ids) {
+    lines.push(`  ${JSON.stringify(id)}: ${serializeEntry3(entries[id])},`);
+  }
+  lines.push("}");
+  lines.push("");
+  return lines.join("\n");
+}
+async function generate3(ctx) {
+  const entries = {};
+  const anthropicSrc = sources3.find((s) => s.id === "llm-anthropic");
+  if (anthropicSrc) {
+    const parsed = AnthropicSnapshotSchema.parse(await ctx.fetchSource(anthropicSrc));
+    for (const m of parsed.data) {
+      if (!m.max_input_tokens || !m.max_tokens) continue;
+      entries[m.id] = {
+        contextWindow: m.max_input_tokens,
+        maxOutput: m.max_tokens,
+        ...m.display_name ? { displayName: m.display_name } : {},
+        provider: "anthropic"
+      };
+    }
+  }
+  const groqSrc = sources3.find((s) => s.id === "llm-groq");
+  if (groqSrc) {
+    const parsed = GroqSnapshotSchema.parse(await ctx.fetchSource(groqSrc));
+    for (const m of parsed.data) {
+      if (!m.context_window) continue;
+      entries[m.id] = {
+        contextWindow: m.context_window,
+        ...m.max_completion_tokens ? { maxOutput: m.max_completion_tokens } : {},
+        provider: "groq"
+      };
+    }
+  }
+  const xaiSrc = sources3.find((s) => s.id === "llm-xai");
+  if (xaiSrc) {
+    const parsed = XaiSnapshotSchema.parse(await ctx.fetchSource(xaiSrc));
+    for (const m of parsed.data) {
+      if (!m.context_length) continue;
+      entries[m.id] = { contextWindow: m.context_length, provider: "xai" };
+    }
+  }
+  const moonshotSrc = sources3.find((s) => s.id === "llm-moonshot");
+  if (moonshotSrc) {
+    const parsed = MoonshotSnapshotSchema.parse(await ctx.fetchSource(moonshotSrc));
+    for (const m of parsed.data) {
+      if (!m.context_length) continue;
+      entries[m.id] = { contextWindow: m.context_length, provider: "moonshot" };
+    }
+  }
+  return { [OUTPUT_PATH3]: serializeFile3(entries) };
+}
+var sources3 = [
+  {
+    id: "llm-anthropic",
+    url: "https://api.anthropic.com/v1/models?limit=1000",
+    headers: {
+      "x-api-key": "env:ANTHROPIC_API_KEY",
+      "anthropic-version": "2023-06-01"
+    }
+  },
+  {
+    id: "llm-groq",
+    url: "https://api.groq.com/openai/v1/models",
+    headers: { Authorization: "Bearer env:GROQ_API_KEY" }
+  },
+  {
+    id: "llm-xai",
+    url: "https://api.x.ai/v1/models",
+    headers: { Authorization: "Bearer env:XAI_API_KEY" }
+  },
+  {
+    id: "llm-moonshot",
+    url: "https://api.moonshot.ai/v1/models",
+    headers: { Authorization: "Bearer env:MOONSHOT_API_KEY" }
+  }
+];
+var llmContextWindowsGenerator = defineGenerator({
+  name: "llm:context-windows",
+  modality: "llm",
+  sources: sources3,
+  generate: generate3
+});
 
 // src/generators/serialize-voices.ts
 function serializeVoice(v) {
@@ -176,7 +422,7 @@ var ELEVENLABS_SOURCE = {
   // Live refresh needs the account key; offline reads ignore this.
   headers: { "xi-api-key": "env:ELEVENLABS_API_KEY" }
 };
-var OUTPUT_PATH2 = "packages/catalog-sync/generated/voice-elevenlabs.generated.ts";
+var OUTPUT_PATH4 = "packages/catalog-sync/generated/voice-elevenlabs.generated.ts";
 var voiceElevenlabs = defineGenerator({
   name: "voice:elevenlabs",
   modality: "voice",
@@ -187,7 +433,7 @@ var voiceElevenlabs = defineGenerator({
     );
     const voices = mapElevenLabsVoices(snapshot);
     return {
-      [OUTPUT_PATH2]: serializeVoiceModule(
+      [OUTPUT_PATH4]: serializeVoiceModule(
         "voice:elevenlabs",
         "ELEVENLABS_VOICES",
         voices
@@ -207,7 +453,7 @@ var MINIMAX_SOURCE = {
   },
   body: { voice_type: "system" }
 };
-var OUTPUT_PATH3 = "packages/catalog-sync/generated/voice-minimax.generated.ts";
+var OUTPUT_PATH5 = "packages/catalog-sync/generated/voice-minimax.generated.ts";
 var voiceMinimax = defineGenerator({
   name: "voice:minimax",
   modality: "voice",
@@ -218,7 +464,7 @@ var voiceMinimax = defineGenerator({
     );
     const voices = mapMinimaxVoices(snapshot);
     return {
-      [OUTPUT_PATH3]: serializeVoiceModule(
+      [OUTPUT_PATH5]: serializeVoiceModule(
         "voice:minimax",
         "MINIMAX_VOICES",
         voices
@@ -631,6 +877,8 @@ var refresh = flags.has("--refresh");
 var check = flags.has("--check");
 var generators = [
   llmOpenRouterGenerator,
+  llmRequestyGenerator,
+  llmContextWindowsGenerator,
   voiceElevenlabs,
   voiceMinimax,
   imageReplicate