@opennodes/registry 0.1.1 → 0.1.2

package/src/importers.js

@@ -1,346 +1,346 @@
-// DS-IMP-01: seed importers so search is useful before any real registration.
-// Imported records enter as unverified, unsigned, source='import', and are
-// excluded from verified tiers until claimed by their operator.
-import { validateCard } from '@opennodes/core';
-
-const slug = (s) => String(s).toLowerCase().replace(/[^a-z0-9._-]+/g, '-').replace(/^[^a-z0-9]+/, '').slice(0, 100);
-const vendorSlug = (s) => slug(s).replace(/[._]/g, '-');
-
-/** Map input/output modality arrays (the shape both HF and OpenRouter use) to the ONP enum. */
-export function mapModalities(inputs = ['text'], outputs = ['text']) {
-  const out = new Set(outputs);
-  if (out.has('image')) return 'image';
-  if (out.has('video')) return 'video';
-  if (out.has('audio')) return 'audio';
-  if (out.has('embedding') || out.has('embeddings')) return 'embedding';
-  return inputs.some((i) => i !== 'text') ? 'multimodal' : 'text';
-}
-
-const PIPELINE_TAG_MODALITY = {
-  'text-generation': 'text', 'text2text-generation': 'text', 'conversational': 'text',
-  'image-text-to-text': 'multimodal', 'visual-question-answering': 'multimodal',
-  'text-to-image': 'image', 'image-to-image': 'image',
-  'text-to-video': 'video', 'image-to-video': 'video', 'image-text-to-video': 'video',
-  'text-to-speech': 'audio', 'automatic-speech-recognition': 'audio', 'audio-to-audio': 'audio',
-  'feature-extraction': 'embedding', 'sentence-similarity': 'embedding',
-};
-
-/** Per-token price strings/values -> USD per MTok. Values >= 0.01 are assumed per-MTok already. */
-const perMtok = (v) => {
-  const n = Number(v);
-  if (!Number.isFinite(n) || n <= 0) return 0;
-  return Math.round((n < 0.01 ? n * 1_000_000 : n) * 1e6) / 1e6;
-};
-
-async function saveImportedCard(store, nodeId, origin, sourceRef, card) {
-  const errors = validateCard(card);
-  if (errors.length) return errors[0];
-  await store.upsertNode(nodeId, origin, sourceRef, 'indexed', 'import');
-  await store.saveRevision(nodeId, card.revision, card, false);
-  return null;
-}
-
-/**
- * Import a models.dev-shaped catalog: { providerId: { id, name, api?, doc?, models: { key: model } } }.
- * Each provider becomes a synthetic node `import.modelsdev.{provider}`; each model an offering.
- * Returns { providers, offerings, skipped }.
- */
-export async function importModelsDev(store, catalog) {
-  let providers = 0, offerings = 0;
-  const skipped = [];
-  for (const [providerId, provider] of Object.entries(catalog ?? {})) {
-    const models = provider?.models ?? {};
-    const offeringList = [];
-    for (const [modelKey, model] of Object.entries(models)) {
-      const outputs = model?.modalities?.output ?? ['text'];
-      const modality = outputs.includes('text') ? 'text' : outputs.includes('image') ? 'image' : 'text';
-      offeringList.push({
-        offering_id: slug(modelKey) || 'model',
-        modality,
-        model: {
-          name: model?.name ?? modelKey,
-          family: slug(providerId),
-          artifact: `urn:proprietary:${slug(providerId).replace(/[._]/g, '-')}:${slug(model?.id ?? modelKey).replace(/\//g, '-')}`,
-        },
-        serving: {
-          ...(model?.limit?.context ? { context_window: Math.floor(model.limit.context) } : {}),
-          ...(model?.limit?.output ? { max_output_tokens: Math.floor(model.limit.output) } : {}),
-          ...(model?.tool_call ? { supports: ['tool_calls'] } : {}),
-        },
-        binding: { profile: 'onp.openai.chat/v1', model_id: model?.id ?? modelKey },
-        pricing: {
-          currency: 'USD',
-          input_per_mtok: Number(model?.cost?.input ?? 0),
-          output_per_mtok: Number(model?.cost?.output ?? 0),
-          schemes: ['prepaid'],
-        },
-      });
-    }
-    if (!offeringList.length) { skipped.push(providerId); continue; }
-
-    const nodeId = `import.modelsdev.${slug(providerId).replace(/[._]/g, '-')}`;
-    const base = provider?.api ?? provider?.doc ?? 'https://models.dev';
-    const card = {
-      onp: '0.1',
-      revision: new Date().toISOString(),
-      node: {
-        id: nodeId,
-        name: provider?.name ?? providerId,
-        description: `Imported from models.dev; metadata only, unclaimed by the operator.`,
-        operator: { name: provider?.name ?? providerId },
-      },
-      endpoints: { openai: base, health: `${base.replace(/\/$/, '')}/health` },
-      offerings: offeringList,
-      payment: { schemes: [{ scheme: 'prepaid' }] },
-    };
-    const errors = validateCard(card);
-    if (errors.length) { skipped.push(`${providerId}: ${errors[0]}`); continue; }
-
-    await store.upsertNode(nodeId, new URL(base).origin, `import:modelsdev:${providerId}`, 'indexed', 'import');
-    await store.saveRevision(nodeId, card.revision, card, false);
-    providers++;
-    offerings += offeringList.length;
-  }
-  return { providers, offerings, skipped };
-}
-
-export const HF_ROUTER_URL = 'https://router.huggingface.co/v1/models';
-export const HF_HUB_URL = 'https://huggingface.co/api/models?inference_provider=all&limit=1000&expand[]=inferenceProviderMapping&expand[]=pipeline_tag&expand[]=safetensors';
-
-/**
- * Import Hugging Face Inference Providers: `router` is the /v1/models response
- * (live serving: per-provider status/capabilities/perf), `hub` the Hub API list
- * (params_b from safetensors, pipeline_tag). One synthetic node per provider
- * (`import.hf.{provider}`); artifact `hf:{repo}` is authoritative.
- */
-export async function importHuggingFace(store, router, hub = []) {
-  const hubById = new Map((hub ?? []).map((m) => [m.id, m]));
-  const byProvider = new Map(); // provider -> offerings[]
-
-  for (const model of router?.data ?? []) {
-    const hubEntry = hubById.get(model.id);
-    const paramsB = hubEntry?.safetensors?.total
-      ? Math.round(hubEntry.safetensors.total / 1e7) / 100 : undefined;
-    const inputs = model.architecture?.input_modalities ?? ['text'];
-    const outputs = model.architecture?.output_modalities ?? ['text'];
-    const modality = hubEntry?.pipeline_tag && PIPELINE_TAG_MODALITY[hubEntry.pipeline_tag]
-      ? PIPELINE_TAG_MODALITY[hubEntry.pipeline_tag]
-      : mapModalities(inputs, outputs);
-
-    for (const p of model.providers ?? []) {
-      if (p.status !== 'live') continue;
-      const supports = ['streaming'];
-      if (p.supports_tools) supports.push('tool_calls');
-      if (p.supports_structured_output) supports.push('json_mode');
-      if (inputs.includes('image') && outputs.includes('text')) supports.push('vision');
-      const offering = {
-        offering_id: slug(model.id),
-        modality,
-        modalities: { input: inputs, output: outputs }, // Draft 0.2 field, additive
-        model: {
-          name: model.id,
-          family: vendorSlug(model.owned_by ?? model.id.split('/')[0]),
-          artifact: `hf:${model.id}`,
-          ...(paramsB ? { params_b: paramsB } : {}),
-        },
-        serving: {
-          supports,
-          ...(p.context_length ? { context_window: Math.floor(p.context_length) } : {}),
-          ...(Number.isFinite(p.first_token_latency_ms)
-            ? { expected_ttft_ms: { p50: Math.round(p.first_token_latency_ms), basis: 'claimed' } } : {}),
-          ...(Number.isFinite(p.throughput)
-            ? { expected_tps: { p50: Math.round(p.throughput), basis: 'claimed' } } : {}),
-        },
-        binding: { profile: 'onp.openai.chat/v1', model_id: `${model.id}:${p.provider}` },
-        pricing: {
-          currency: 'USD',
-          input_per_mtok: perMtok(p.pricing?.input),
-          output_per_mtok: perMtok(p.pricing?.output),
-          schemes: p.is_free ? ['free'] : ['prepaid'],
-        },
-      };
-      if (!byProvider.has(p.provider)) byProvider.set(p.provider, []);
-      byProvider.get(p.provider).push(offering);
-    }
-  }
-
-  let providers = 0, offerings = 0;
-  const skipped = [];
-  for (const [provider, offeringList] of byProvider) {
-    const nodeId = `import.hf.${vendorSlug(provider)}`;
-    const card = {
-      onp: '0.1',
-      revision: new Date().toISOString(),
-      node: {
-        id: nodeId,
-        name: `${provider} (via Hugging Face Inference Providers)`,
-        description: 'Imported from the Hugging Face Inference Providers router; metadata only, unclaimed by the operator.',
-        operator: { name: provider },
-      },
-      endpoints: { openai: 'https://router.huggingface.co/v1', health: 'https://router.huggingface.co/health' },
-      offerings: offeringList,
-      payment: { schemes: [{ scheme: 'prepaid', signup: 'https://huggingface.co/settings/billing' }] },
-    };
-    const err = await saveImportedCard(store, nodeId, 'https://router.huggingface.co', `import:hf:${provider}`, card);
-    if (err) { skipped.push(`${provider}: ${err}`); continue; }
-    providers++;
-    offerings += offeringList.length;
-  }
-  return { providers, offerings, skipped };
-}
-
-export const OLLAMA_LIBRARY_URL = 'https://ollama.com/library';
-
-/** Parse ollama.com/library HTML into entries. Resilient to styling churn:
- *  anchors on /library/{name}, spans classified by content (sizes look like "27b"/"270m"). */
-export function parseOllamaLibrary(html) {
-  const entries = [];
-  const blocks = String(html).split('href="/library/').slice(1);
-  for (const block of blocks) {
-    const name = block.match(/^([a-z0-9._-]+)"/)?.[1];
-    if (!name) continue;
-    const chunk = block.slice(0, 6000);
-    const description = chunk.match(/<p[^>]*class="[^"]*max-w-lg[^"]*"[^>]*>\s*([^<]{1,400}?)\s*<\/p>/)?.[1]?.trim() ?? null;
-    const spans = [...chunk.matchAll(/<span[^>]*class="[^"]*inline-flex[^"]*"[^>]*>\s*([^<]{1,40}?)\s*<\/span>/g)].map((m) => m[1].trim());
-    const sizes = spans.filter((s) => /^\d+(\.\d+)?[mb]$/i.test(s));
-    const capabilities = spans.filter((s) => !/^\d+(\.\d+)?[mb]$/i.test(s));
-    const pulls = chunk.match(/<span[^>]*>\s*([\d.]+[KMB]?)\s*<\/span>\s*<span[^>]*>&nbsp;Pulls/)?.[1] ?? null;
-    entries.push({ name, description, capabilities, sizes, pulls });
-  }
-  return entries;
-}
-
-const sizeToParamsB = (s) => {
-  const m = String(s).match(/^(\d+(?:\.\d+)?)([mb])$/i);
-  if (!m) return null;
-  return m[2].toLowerCase() === 'b' ? Number(m[1]) : Math.round(Number(m[1]) / 10) / 100;
-};
-
-/**
- * Import the Ollama library as locally-runnable offerings on a synthetic node whose
- * endpoint is the user's OWN machine (http://127.0.0.1:11434/v1) — free, local, and
- * directly usable from the model-page playground once `ollama run <name>` has pulled it.
- */
-export async function importOllamaLibrary(store, entries) {
-  const offeringList = [];
-  for (const e of entries ?? []) {
-    if (!e?.name) continue;
-    const caps = (e.capabilities ?? []).map((c) => String(c).toLowerCase());
-    const modality = caps.includes('embedding') ? 'embedding' : (caps.includes('vision') ? 'multimodal' : 'text');
-    const paramsCandidates = (e.sizes ?? []).map(sizeToParamsB).filter((n) => n != null);
-    const supports = ['streaming'];
-    if (caps.includes('tools')) supports.push('tool_calls');
-    if (caps.includes('vision')) supports.push('vision');
-    offeringList.push({
-      offering_id: slug(e.name),
-      modality,
-      model: {
-        name: e.name,
-        family: vendorSlug(e.name.split(/[.:0-9]/)[0] || e.name),
-        artifact: `urn:proprietary:ollama:${vendorSlug(e.name)}`,
-        ...(paramsCandidates.length ? { params_b: Math.max(...paramsCandidates) } : {}),
-        ...(e.description ? { description: String(e.description).slice(0, 500) } : {}),
-      },
-      serving: { supports },
-      binding: {
-        profile: modality === 'embedding' ? 'onp.openai.embeddings/v1' : 'onp.openai.chat/v1',
-        model_id: e.name,
-      },
-      pricing: { currency: 'USD', input_per_mtok: 0, output_per_mtok: 0, schemes: ['free'] },
-      local: {
-        runtime: 'ollama',
-        run: `ollama run ${e.name}`,
-        ...(e.pulls ? { pulls: e.pulls } : {}),
-        ...(e.sizes?.length ? { sizes: e.sizes } : {}),
-      },
-    });
-  }
-  if (!offeringList.length) return { providers: 0, offerings: 0, skipped: ['no entries'] };
-  const nodeId = 'import.ollama.library';
-  const card = {
-    onp: '0.1',
-    revision: new Date().toISOString(),
-    node: {
-      id: nodeId,
-      name: 'Ollama library (runs on your machine)',
-      description: 'Locally-runnable models from the Ollama library. The endpoint is your own machine: install Ollama, `ollama run <model>`, and these offerings serve at localhost — free, private, no network.',
-      operator: { name: 'Ollama (community library)' },
-    },
-    endpoints: { openai: 'http://127.0.0.1:11434/v1', health: 'http://127.0.0.1:11434' },
-    offerings: offeringList,
-    payment: { schemes: [{ scheme: 'free' }] },
-  };
-  const err = await saveImportedCard(store, nodeId, 'https://ollama.com', 'import:ollama-library', card);
-  if (err) return { providers: 0, offerings: 0, skipped: [err] };
-  return { providers: 1, offerings: offeringList.length, skipped: [] };
-}
-
-export const OPENROUTER_URL = 'https://openrouter.ai/api/v1/models';
-
-/**
- * Import the OpenRouter catalog as one synthetic node (`import.openrouter.catalog`):
- * per-token price strings -> per-MTok; `hugging_face_id` gives the artifact join
- * where present, else a proprietary URN under the model author's namespace.
- */
-export async function importOpenRouter(store, catalog) {
-  const offeringList = [];
-  for (const model of catalog?.data ?? []) {
-    const inputs = model.architecture?.input_modalities ?? ['text'];
-    const outputs = model.architecture?.output_modalities ?? ['text'];
-    const supports = ['streaming'];
-    const params = model.supported_parameters ?? [];
-    if (params.includes('tools')) supports.push('tool_calls');
-    if (params.includes('response_format') || params.includes('structured_outputs')) supports.push('json_mode');
-    if (inputs.includes('image') && outputs.includes('text')) supports.push('vision');
-    const author = vendorSlug(model.id.split('/')[0] || 'unknown');
-    offeringList.push({
-      offering_id: slug(model.id),
-      modality: mapModalities(inputs, outputs),
-      modalities: { input: inputs, output: outputs },
-      model: {
-        name: model.name ?? model.id,
-        family: author,
-        artifact: model.hugging_face_id
-          ? `hf:${model.hugging_face_id}`
-          : `urn:proprietary:${author}:${slug(model.id.split('/').slice(1).join('-') || model.id)}`,
-        ...(model.description ? { description: String(model.description).slice(0, 500) } : {}),
-        ...(Number.isFinite(model.created)
-          ? { released: new Date(model.created * 1000).toISOString().slice(0, 10) } : {}),
-      },
-      serving: {
-        supports,
-        ...(model.context_length ? { context_window: Math.floor(model.context_length) } : {}),
-        ...(model.top_provider?.max_completion_tokens
-          ? { max_output_tokens: Math.floor(model.top_provider.max_completion_tokens) } : {}),
-      },
-      binding: { profile: 'onp.openai.chat/v1', model_id: model.id },
-      pricing: {
-        currency: 'USD',
-        input_per_mtok: perMtok(model.pricing?.prompt),
-        output_per_mtok: perMtok(model.pricing?.completion),
-        ...(model.pricing?.input_cache_read
-          ? { cached_input_per_mtok: perMtok(model.pricing.input_cache_read) } : {}),
-        schemes: Number(model.pricing?.prompt) === 0 ? ['free'] : ['prepaid'],
-      },
-    });
-  }
-  if (!offeringList.length) return { providers: 0, offerings: 0, skipped: ['empty catalog'] };
-
-  const nodeId = 'import.openrouter.catalog';
-  const card = {
-    onp: '0.1',
-    revision: new Date().toISOString(),
-    node: {
-      id: nodeId,
-      name: 'OpenRouter catalog',
-      description: 'Imported from the public OpenRouter models API; metadata only, unclaimed by the operator.',
-      operator: { name: 'OpenRouter' },
-    },
-    endpoints: { openai: 'https://openrouter.ai/api/v1', health: 'https://openrouter.ai/api/v1/models' },
-    offerings: offeringList,
-    payment: { schemes: [{ scheme: 'prepaid', signup: 'https://openrouter.ai/keys' }] },
-  };
-  const err = await saveImportedCard(store, nodeId, 'https://openrouter.ai', 'import:openrouter', card);
-  if (err) return { providers: 0, offerings: 0, skipped: [err] };
-  return { providers: 1, offerings: offeringList.length, skipped: [] };
-}
+// DS-IMP-01: seed importers so search is useful before any real registration.
+// Imported records enter as unverified, unsigned, source='import', and are
+// excluded from verified tiers until claimed by their operator.
+import { validateCard } from '@opennodes/core';
+
+const slug = (s) => String(s).toLowerCase().replace(/[^a-z0-9._-]+/g, '-').replace(/^[^a-z0-9]+/, '').slice(0, 100);
+const vendorSlug = (s) => slug(s).replace(/[._]/g, '-');
+
+/** Map input/output modality arrays (the shape both HF and OpenRouter use) to the ONP enum. */
+export function mapModalities(inputs = ['text'], outputs = ['text']) {
+  const out = new Set(outputs);
+  if (out.has('image')) return 'image';
+  if (out.has('video')) return 'video';
+  if (out.has('audio')) return 'audio';
+  if (out.has('embedding') || out.has('embeddings')) return 'embedding';
+  return inputs.some((i) => i !== 'text') ? 'multimodal' : 'text';
+}
+
+const PIPELINE_TAG_MODALITY = {
+  'text-generation': 'text', 'text2text-generation': 'text', 'conversational': 'text',
+  'image-text-to-text': 'multimodal', 'visual-question-answering': 'multimodal',
+  'text-to-image': 'image', 'image-to-image': 'image',
+  'text-to-video': 'video', 'image-to-video': 'video', 'image-text-to-video': 'video',
+  'text-to-speech': 'audio', 'automatic-speech-recognition': 'audio', 'audio-to-audio': 'audio',
+  'feature-extraction': 'embedding', 'sentence-similarity': 'embedding',
+};
+
+/** Per-token price strings/values -> USD per MTok. Values >= 0.01 are assumed per-MTok already. */
+const perMtok = (v) => {
+  const n = Number(v);
+  if (!Number.isFinite(n) || n <= 0) return 0;
+  return Math.round((n < 0.01 ? n * 1_000_000 : n) * 1e6) / 1e6;
+};
+
+async function saveImportedCard(store, nodeId, origin, sourceRef, card) {
+  const errors = validateCard(card);
+  if (errors.length) return errors[0];
+  await store.upsertNode(nodeId, origin, sourceRef, 'indexed', 'import');
+  await store.saveRevision(nodeId, card.revision, card, false);
+  return null;
+}
+
+/**
+ * Import a models.dev-shaped catalog: { providerId: { id, name, api?, doc?, models: { key: model } } }.
+ * Each provider becomes a synthetic node `import.modelsdev.{provider}`; each model an offering.
+ * Returns { providers, offerings, skipped }.
+ */
+export async function importModelsDev(store, catalog) {
+  let providers = 0, offerings = 0;
+  const skipped = [];
+  for (const [providerId, provider] of Object.entries(catalog ?? {})) {
+    const models = provider?.models ?? {};
+    const offeringList = [];
+    for (const [modelKey, model] of Object.entries(models)) {
+      const outputs = model?.modalities?.output ?? ['text'];
+      const modality = outputs.includes('text') ? 'text' : outputs.includes('image') ? 'image' : 'text';
+      offeringList.push({
+        offering_id: slug(modelKey) || 'model',
+        modality,
+        model: {
+          name: model?.name ?? modelKey,
+          family: slug(providerId),
+          artifact: `urn:proprietary:${slug(providerId).replace(/[._]/g, '-')}:${slug(model?.id ?? modelKey).replace(/\//g, '-')}`,
+        },
+        serving: {
+          ...(model?.limit?.context ? { context_window: Math.floor(model.limit.context) } : {}),
+          ...(model?.limit?.output ? { max_output_tokens: Math.floor(model.limit.output) } : {}),
+          ...(model?.tool_call ? { supports: ['tool_calls'] } : {}),
+        },
+        binding: { profile: 'onp.openai.chat/v1', model_id: model?.id ?? modelKey },
+        pricing: {
+          currency: 'USD',
+          input_per_mtok: Number(model?.cost?.input ?? 0),
+          output_per_mtok: Number(model?.cost?.output ?? 0),
+          schemes: ['prepaid'],
+        },
+      });
+    }
+    if (!offeringList.length) { skipped.push(providerId); continue; }
+
+    const nodeId = `import.modelsdev.${slug(providerId).replace(/[._]/g, '-')}`;
+    const base = provider?.api ?? provider?.doc ?? 'https://models.dev';
+    const card = {
+      onp: '0.1',
+      revision: new Date().toISOString(),
+      node: {
+        id: nodeId,
+        name: provider?.name ?? providerId,
+        description: `Imported from models.dev; metadata only, unclaimed by the operator.`,
+        operator: { name: provider?.name ?? providerId },
+      },
+      endpoints: { openai: base, health: `${base.replace(/\/$/, '')}/health` },
+      offerings: offeringList,
+      payment: { schemes: [{ scheme: 'prepaid' }] },
+    };
+    const errors = validateCard(card);
+    if (errors.length) { skipped.push(`${providerId}: ${errors[0]}`); continue; }
+
+    await store.upsertNode(nodeId, new URL(base).origin, `import:modelsdev:${providerId}`, 'indexed', 'import');
+    await store.saveRevision(nodeId, card.revision, card, false);
+    providers++;
+    offerings += offeringList.length;
+  }
+  return { providers, offerings, skipped };
+}
+
+export const HF_ROUTER_URL = 'https://router.huggingface.co/v1/models';
+export const HF_HUB_URL = 'https://huggingface.co/api/models?inference_provider=all&limit=1000&expand[]=inferenceProviderMapping&expand[]=pipeline_tag&expand[]=safetensors';
+
+/**
+ * Import Hugging Face Inference Providers: `router` is the /v1/models response
+ * (live serving: per-provider status/capabilities/perf), `hub` the Hub API list
+ * (params_b from safetensors, pipeline_tag). One synthetic node per provider
+ * (`import.hf.{provider}`); artifact `hf:{repo}` is authoritative.
+ */
+export async function importHuggingFace(store, router, hub = []) {
+  const hubById = new Map((hub ?? []).map((m) => [m.id, m]));
+  const byProvider = new Map(); // provider -> offerings[]
+
+  for (const model of router?.data ?? []) {
+    const hubEntry = hubById.get(model.id);
+    const paramsB = hubEntry?.safetensors?.total
+      ? Math.round(hubEntry.safetensors.total / 1e7) / 100 : undefined;
+    const inputs = model.architecture?.input_modalities ?? ['text'];
+    const outputs = model.architecture?.output_modalities ?? ['text'];
+    const modality = hubEntry?.pipeline_tag && PIPELINE_TAG_MODALITY[hubEntry.pipeline_tag]
+      ? PIPELINE_TAG_MODALITY[hubEntry.pipeline_tag]
+      : mapModalities(inputs, outputs);
+
+    for (const p of model.providers ?? []) {
+      if (p.status !== 'live') continue;
+      const supports = ['streaming'];
+      if (p.supports_tools) supports.push('tool_calls');
+      if (p.supports_structured_output) supports.push('json_mode');
+      if (inputs.includes('image') && outputs.includes('text')) supports.push('vision');
+      const offering = {
+        offering_id: slug(model.id),
+        modality,
+        modalities: { input: inputs, output: outputs }, // Draft 0.2 field, additive
+        model: {
+          name: model.id,
+          family: vendorSlug(model.owned_by ?? model.id.split('/')[0]),
+          artifact: `hf:${model.id}`,
+          ...(paramsB ? { params_b: paramsB } : {}),
+        },
+        serving: {
+          supports,
+          ...(p.context_length ? { context_window: Math.floor(p.context_length) } : {}),
+          ...(Number.isFinite(p.first_token_latency_ms)
+            ? { expected_ttft_ms: { p50: Math.round(p.first_token_latency_ms), basis: 'claimed' } } : {}),
+          ...(Number.isFinite(p.throughput)
+            ? { expected_tps: { p50: Math.round(p.throughput), basis: 'claimed' } } : {}),
+        },
+        binding: { profile: 'onp.openai.chat/v1', model_id: `${model.id}:${p.provider}` },
+        pricing: {
+          currency: 'USD',
+          input_per_mtok: perMtok(p.pricing?.input),
+          output_per_mtok: perMtok(p.pricing?.output),
+          schemes: p.is_free ? ['free'] : ['prepaid'],
+        },
+      };
+      if (!byProvider.has(p.provider)) byProvider.set(p.provider, []);
+      byProvider.get(p.provider).push(offering);
+    }
+  }
+
+  let providers = 0, offerings = 0;
+  const skipped = [];
+  for (const [provider, offeringList] of byProvider) {
+    const nodeId = `import.hf.${vendorSlug(provider)}`;
+    const card = {
+      onp: '0.1',
+      revision: new Date().toISOString(),
+      node: {
+        id: nodeId,
+        name: `${provider} (via Hugging Face Inference Providers)`,
+        description: 'Imported from the Hugging Face Inference Providers router; metadata only, unclaimed by the operator.',
+        operator: { name: provider },
+      },
+      endpoints: { openai: 'https://router.huggingface.co/v1', health: 'https://router.huggingface.co/health' },
+      offerings: offeringList,
+      payment: { schemes: [{ scheme: 'prepaid', signup: 'https://huggingface.co/settings/billing' }] },
+    };
+    const err = await saveImportedCard(store, nodeId, 'https://router.huggingface.co', `import:hf:${provider}`, card);
+    if (err) { skipped.push(`${provider}: ${err}`); continue; }
+    providers++;
+    offerings += offeringList.length;
+  }
+  return { providers, offerings, skipped };
+}
+
+export const OLLAMA_LIBRARY_URL = 'https://ollama.com/library';
+
+/** Parse ollama.com/library HTML into entries. Resilient to styling churn:
+ *  anchors on /library/{name}, spans classified by content (sizes look like "27b"/"270m"). */
+export function parseOllamaLibrary(html) {
+  const entries = [];
+  const blocks = String(html).split('href="/library/').slice(1);
+  for (const block of blocks) {
+    const name = block.match(/^([a-z0-9._-]+)"/)?.[1];
+    if (!name) continue;
+    const chunk = block.slice(0, 6000);
+    const description = chunk.match(/<p[^>]*class="[^"]*max-w-lg[^"]*"[^>]*>\s*([^<]{1,400}?)\s*<\/p>/)?.[1]?.trim() ?? null;
+    const spans = [...chunk.matchAll(/<span[^>]*class="[^"]*inline-flex[^"]*"[^>]*>\s*([^<]{1,40}?)\s*<\/span>/g)].map((m) => m[1].trim());
+    const sizes = spans.filter((s) => /^\d+(\.\d+)?[mb]$/i.test(s));
+    const capabilities = spans.filter((s) => !/^\d+(\.\d+)?[mb]$/i.test(s));
+    const pulls = chunk.match(/<span[^>]*>\s*([\d.]+[KMB]?)\s*<\/span>\s*<span[^>]*>&nbsp;Pulls/)?.[1] ?? null;
+    entries.push({ name, description, capabilities, sizes, pulls });
+  }
+  return entries;
+}
+
+const sizeToParamsB = (s) => {
+  const m = String(s).match(/^(\d+(?:\.\d+)?)([mb])$/i);
+  if (!m) return null;
+  return m[2].toLowerCase() === 'b' ? Number(m[1]) : Math.round(Number(m[1]) / 10) / 100;
+};
+
+/**
+ * Import the Ollama library as locally-runnable offerings on a synthetic node whose
+ * endpoint is the user's OWN machine (http://127.0.0.1:11434/v1) — free, local, and
+ * directly usable from the model-page playground once `ollama run <name>` has pulled it.
+ */
+export async function importOllamaLibrary(store, entries) {
+  const offeringList = [];
+  for (const e of entries ?? []) {
+    if (!e?.name) continue;
+    const caps = (e.capabilities ?? []).map((c) => String(c).toLowerCase());
+    const modality = caps.includes('embedding') ? 'embedding' : (caps.includes('vision') ? 'multimodal' : 'text');
+    const paramsCandidates = (e.sizes ?? []).map(sizeToParamsB).filter((n) => n != null);
+    const supports = ['streaming'];
+    if (caps.includes('tools')) supports.push('tool_calls');
+    if (caps.includes('vision')) supports.push('vision');
+    offeringList.push({
+      offering_id: slug(e.name),
+      modality,
+      model: {
+        name: e.name,
+        family: vendorSlug(e.name.split(/[.:0-9]/)[0] || e.name),
+        artifact: `urn:proprietary:ollama:${vendorSlug(e.name)}`,
+        ...(paramsCandidates.length ? { params_b: Math.max(...paramsCandidates) } : {}),
+        ...(e.description ? { description: String(e.description).slice(0, 500) } : {}),
+      },
+      serving: { supports },
+      binding: {
+        profile: modality === 'embedding' ? 'onp.openai.embeddings/v1' : 'onp.openai.chat/v1',
+        model_id: e.name,
+      },
+      pricing: { currency: 'USD', input_per_mtok: 0, output_per_mtok: 0, schemes: ['free'] },
+      local: {
+        runtime: 'ollama',
+        run: `ollama run ${e.name}`,
+        ...(e.pulls ? { pulls: e.pulls } : {}),
+        ...(e.sizes?.length ? { sizes: e.sizes } : {}),
+      },
+    });
+  }
+  if (!offeringList.length) return { providers: 0, offerings: 0, skipped: ['no entries'] };
+  const nodeId = 'import.ollama.library';
+  const card = {
+    onp: '0.1',
+    revision: new Date().toISOString(),
+    node: {
+      id: nodeId,
+      name: 'Ollama library (runs on your machine)',
+      description: 'Locally-runnable models from the Ollama library. The endpoint is your own machine: install Ollama, `ollama run <model>`, and these offerings serve at localhost — free, private, no network.',
+      operator: { name: 'Ollama (community library)' },
+    },
+    endpoints: { openai: 'http://127.0.0.1:11434/v1', health: 'http://127.0.0.1:11434' },
+    offerings: offeringList,
+    payment: { schemes: [{ scheme: 'free' }] },
+  };
+  const err = await saveImportedCard(store, nodeId, 'https://ollama.com', 'import:ollama-library', card);
+  if (err) return { providers: 0, offerings: 0, skipped: [err] };
+  return { providers: 1, offerings: offeringList.length, skipped: [] };
+}
+
+export const OPENROUTER_URL = 'https://openrouter.ai/api/v1/models';
+
+/**
+ * Import the OpenRouter catalog as one synthetic node (`import.openrouter.catalog`):
+ * per-token price strings -> per-MTok; `hugging_face_id` gives the artifact join
+ * where present, else a proprietary URN under the model author's namespace.
+ */
+export async function importOpenRouter(store, catalog) {
+  const offeringList = [];
+  for (const model of catalog?.data ?? []) {
+    const inputs = model.architecture?.input_modalities ?? ['text'];
+    const outputs = model.architecture?.output_modalities ?? ['text'];
+    const supports = ['streaming'];
+    const params = model.supported_parameters ?? [];
+    if (params.includes('tools')) supports.push('tool_calls');
+    if (params.includes('response_format') || params.includes('structured_outputs')) supports.push('json_mode');
+    if (inputs.includes('image') && outputs.includes('text')) supports.push('vision');
+    const author = vendorSlug(model.id.split('/')[0] || 'unknown');
+    offeringList.push({
+      offering_id: slug(model.id),
+      modality: mapModalities(inputs, outputs),
+      modalities: { input: inputs, output: outputs },
+      model: {
+        name: model.name ?? model.id,
+        family: author,
+        artifact: model.hugging_face_id
+          ? `hf:${model.hugging_face_id}`
+          : `urn:proprietary:${author}:${slug(model.id.split('/').slice(1).join('-') || model.id)}`,
+        ...(model.description ? { description: String(model.description).slice(0, 500) } : {}),
+        ...(Number.isFinite(model.created)
+          ? { released: new Date(model.created * 1000).toISOString().slice(0, 10) } : {}),
+      },
+      serving: {
+        supports,
+        ...(model.context_length ? { context_window: Math.floor(model.context_length) } : {}),
+        ...(model.top_provider?.max_completion_tokens
+          ? { max_output_tokens: Math.floor(model.top_provider.max_completion_tokens) } : {}),
+      },
+      binding: { profile: 'onp.openai.chat/v1', model_id: model.id },
+      pricing: {
+        currency: 'USD',
+        input_per_mtok: perMtok(model.pricing?.prompt),
+        output_per_mtok: perMtok(model.pricing?.completion),
+        ...(model.pricing?.input_cache_read
+          ? { cached_input_per_mtok: perMtok(model.pricing.input_cache_read) } : {}),
+        schemes: Number(model.pricing?.prompt) === 0 ? ['free'] : ['prepaid'],
+      },
+    });
+  }
+  if (!offeringList.length) return { providers: 0, offerings: 0, skipped: ['empty catalog'] };
+
+  const nodeId = 'import.openrouter.catalog';
+  const card = {
+    onp: '0.1',
+    revision: new Date().toISOString(),
+    node: {
+      id: nodeId,
+      name: 'OpenRouter catalog',
+      description: 'Imported from the public OpenRouter models API; metadata only, unclaimed by the operator.',
+      operator: { name: 'OpenRouter' },
+    },
+    endpoints: { openai: 'https://openrouter.ai/api/v1', health: 'https://openrouter.ai/api/v1/models' },
+    offerings: offeringList,
+    payment: { schemes: [{ scheme: 'prepaid', signup: 'https://openrouter.ai/keys' }] },
+  };
+  const err = await saveImportedCard(store, nodeId, 'https://openrouter.ai', 'import:openrouter', card);
+  if (err) return { providers: 0, offerings: 0, skipped: [err] };
+  return { providers: 1, offerings: offeringList.length, skipped: [] };
+}