@dbx-tools/model 0.9.21 → 0.9.28

package/src/generated/model-rate-limits.json

@@ -0,0 +1,342 @@
+{
+  "generatedAt": 1791144427,
+  "catalogue": {
+    "models": {
+      "bge-large-en": {
+        "inputTokensPerMinute": null,
+        "outputTokensPerMinute": null,
+        "queriesPerHour": 2160000
+      },
+      "claude-4-1-opus": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "claude-4-5-haiku": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "claude-4-5-opus": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "claude-4-5-sonnet": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "claude-4-6-opus": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "claude-4-6-sonnet": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "claude-4-7-opus": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "claude-4-8-opus": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "claude-4-sonnet": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "claude-5-1-fable": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "claude-5-5-opus": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "claude-5-5-sonnet": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "claude-5-fable": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "claude-5-opus": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "claude-5-sonnet": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "deepseek-4-1-flash": {
+        "inputTokensPerMinute": 2000000,
+        "outputTokensPerMinute": 40000,
+        "queriesPerHour": 7200
+      },
+      "deepseek-4-flash-0731": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 10000,
+        "queriesPerHour": null
+      },
+      "deepseek-4-pro-0813": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 4000,
+        "queriesPerHour": 7200
+      },
+      "gemini-2-5-flash": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gemini-2-5-pro": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gemini-3-1-flash-image": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gemini-3-1-flash-lite": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gemini-3-1-pro-preview": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gemini-3-5-flash": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "gemini-3-5-flash-lite": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "gemini-3-6-flash": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "gemini-3-7-flash": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "gemini-3-8-flash": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "gemini-3-flash": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gemini-3-pro-image": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gemma-3-12b": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "glm-5-2": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 7200
+      },
+      "glm-5-3": {
+        "inputTokensPerMinute": 2000000,
+        "outputTokensPerMinute": 40000,
+        "queriesPerHour": 7200
+      },
+      "glm-5-3-flash": {
+        "inputTokensPerMinute": 2000000,
+        "outputTokensPerMinute": 40000,
+        "queriesPerHour": 7200
+      },
+      "gpt-5": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gpt-5-1": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gpt-5-2": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gpt-5-3-codex": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gpt-5-4": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gpt-5-4-mini": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gpt-5-4-nano": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gpt-5-5": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gpt-5-5-pro": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gpt-5-6-luna": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gpt-5-6-sol": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gpt-5-6-terra": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gpt-5-mini": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gpt-5-nano": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 20000,
+        "queriesPerHour": 360000
+      },
+      "gpt-6-1-sol": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "gpt-6-astra": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "gpt-6-luna": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "gpt-6-sol": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "gpt-oss-120b": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "gpt-oss-20b": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "grok-4-6": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "grok-4-7": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "gte-large-en": {
+        "inputTokensPerMinute": null,
+        "outputTokensPerMinute": null,
+        "queriesPerHour": 540000
+      },
+      "inkling": {
+        "inputTokensPerMinute": 200000,
+        "outputTokensPerMinute": 10000,
+        "queriesPerHour": 7200
+      },
+      "kimi-3": {
+        "inputTokensPerMinute": 2000000,
+        "outputTokensPerMinute": 40000,
+        "queriesPerHour": null
+      },
+      "llama-3-1-8b-instruct": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "llama-3-3-70b-instruct": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "llama-4-maverick": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "qwen-3-5-122b-a10b": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      },
+      "qwen-3-5-4b": {
+        "inputTokensPerMinute": 2000000,
+        "outputTokensPerMinute": 40000,
+        "queriesPerHour": 7200
+      },
+      "qwen-3-embedding-0-6b": {
+        "inputTokensPerMinute": null,
+        "outputTokensPerMinute": null,
+        "queriesPerHour": 2160000
+      },
+      "qwen-3-next-80b-a3b-instruct": {
+        "inputTokensPerMinute": 1000000,
+        "outputTokensPerMinute": 100000,
+        "queriesPerHour": 360000
+      }
+    }
+  }
+}

package/src/generated/retired-models.json

@@ -0,0 +1,32 @@
+{
+  "generatedAt": 1791144427,
+  "models": [
+    "Anthropic Claude 3.7 Sonnet",
+    "Anthropic Claude Sonnet 4",
+    "DBRX",
+    "DBRX Instruct",
+    "DeepSeek V4 Pro (0813)",
+    "Gemini 2.5 Flash",
+    "Gemini 2.5 Pro",
+    "Gemini 3 Pro",
+    "Kimi K2.7",
+    "Meta Llama 2 13B",
+    "Meta Llama 2 70B",
+    "Meta Llama 2 7B",
+    "Meta Llama 3 (70B)",
+    "Meta Llama 3 8B",
+    "Meta Llama 3.1 405B",
+    "Meta-Llama-2-70B-Chat",
+    "Mistral 7B",
+    "Mixtral 8x7B",
+    "Mixtral-8x7B Instruct",
+    "MPT 30B",
+    "MPT 30B Instruct",
+    "MPT 7B",
+    "MPT 7B Instruct",
+    "OpenAI GPT-5.1 Codex Max",
+    "OpenAI GPT-5.1 Codex Mini",
+    "OpenAI GPT-5.2 Codex",
+    "Thinking Machine Labs Inkling"
+  ]
+}

package/src/invoke.ts

@@ -18,7 +18,7 @@
  * @module
  */
 
-import { isResponsesOnly as isResponsesOnlyRs } from "@dbx-tools/model-rs";
+import { isResponsesOnly as isResponsesOnlyByPolicy } from "./policy.ts";
 
 /**
  * Path segment appended to a serving endpoint for OpenAI-compatible chat
@@ -83,7 +83,7 @@ export function chatCompletionsUrl(host: string): string {
  * those to {@link responsesUrl} instead of {@link invocationsUrl}.
  */
 export function isResponsesOnly(endpoint: string): boolean {
-  return isResponsesOnlyRs(endpoint);
+  return isResponsesOnlyByPolicy(endpoint);
 }
 
 /**

package/src/metadata.ts

@@ -0,0 +1,189 @@
+/**
+ * Build-generated Databricks model retirement, capability, and rate-limit metadata.
+ *
+ * The committed snapshots are refreshed by `bun run model:metadata`. Runtime
+ * lookups build their normalized sets and maps once per process and perform no
+ * documentation network requests.
+ *
+ * @module
+ */
+import * as functionUtils from "@dbx-tools/shared-core/function-utils";
+import type { ModelStatus, ServingEndpointSummary } from "@dbx-tools/shared-model/contracts";
+
+import capabilitiesSnapshotJson from "./generated/model-capabilities.json" with { type: "json" };
+import rateLimitsSnapshotJson from "./generated/model-rate-limits.json" with { type: "json" };
+import retiredModelsSnapshotJson from "./generated/retired-models.json" with { type: "json" };
+import {
+  MODEL_METADATA_TTL_MS,
+  MODEL_RATE_LIMITS_URL,
+  OPENAI_RESPONSES_MODELS_URL,
+  RETIRED_MODELS_URL,
+  WEB_SEARCH_MODELS_URL,
+  type ModelCapabilitiesSnapshot,
+  type ModelCapabilityCatalogue,
+  type ModelRateLimits,
+  type ModelRateLimitsSnapshot,
+  type RetiredModelsSnapshot,
+} from "./_metadata-contract.ts";
+import { modelSearchQuery } from "./policy.ts";
+
+export {
+  MODEL_METADATA_TTL_MS,
+  MODEL_RATE_LIMITS_URL,
+  OPENAI_RESPONSES_MODELS_URL,
+  RETIRED_MODELS_URL,
+  WEB_SEARCH_MODELS_URL,
+};
+export type {
+  ModelCapabilitiesSnapshot,
+  ModelCapabilityCatalogue,
+  ModelRateLimitCatalogue,
+  ModelRateLimits,
+  ModelRateLimitsSnapshot,
+  RetiredModelsSnapshot,
+} from "./_metadata-contract.ts";
+
+const retiredModelsSnapshot = retiredModelsSnapshotJson as RetiredModelsSnapshot;
+const capabilitiesSnapshot = capabilitiesSnapshotJson as ModelCapabilitiesSnapshot;
+const rateLimitsSnapshot = rateLimitsSnapshotJson as ModelRateLimitsSnapshot;
+
+/** Documented capabilities resolved for one model identity. */
+export interface ModelCapabilities {
+  readonly responses: boolean;
+  readonly imageInput: boolean;
+  readonly applyPatch: boolean;
+  readonly webSearch: boolean;
+}
+
+/** Combined build-generated metadata resolved for one model identity. */
+export interface ModelMetadata {
+  readonly status: ModelStatus;
+  readonly capabilities: ModelCapabilities;
+  readonly rateLimits?: ModelRateLimits;
+}
+
+interface MetadataIndex {
+  readonly retiredNames: readonly string[];
+  readonly retiredKeys: ReadonlySet<string>;
+  readonly capabilities: {
+    readonly [Key in keyof ModelCapabilityCatalogue]: ReadonlySet<string>;
+  };
+  readonly rateLimits: ReadonlyMap<string, ModelRateLimits>;
+}
+
+const metadataIndex = functionUtils.memoize<MetadataIndex>(() => ({
+  retiredNames: Object.freeze([...retiredModelsSnapshot.models]),
+  retiredKeys: new Set(retiredModelsSnapshot.models.map(retiredModelKey).filter(Boolean)),
+  capabilities: {
+    responses: new Set(capabilitiesSnapshot.capabilities.responses),
+    imageInput: new Set(capabilitiesSnapshot.capabilities.imageInput),
+    applyPatch: new Set(capabilitiesSnapshot.capabilities.applyPatch),
+    webSearch: new Set(capabilitiesSnapshot.capabilities.webSearch),
+  },
+  rateLimits: new Map(Object.entries(rateLimitsSnapshot.catalogue.models)),
+}));
+
+/** Return the committed retired-model names. The array is cached and immutable. */
+export function retiredModelNames(): readonly string[] {
+  return metadataIndex().retiredNames;
+}
+
+/** Resolve whether any supplied identity is listed in the retirement snapshot. */
+export function modelStatusFor(
+  ...models: readonly (string | ServingEndpointSummary)[]
+): ModelStatus {
+  const retired = metadataIndex().retiredKeys;
+  const deprecated = models
+    .flatMap((model) => modelIdentities(model))
+    .some((identity) => {
+      const candidate = retiredModelKey(identity);
+      for (const key of retired) {
+        if (candidate === key || candidate.startsWith(`${key}-`)) return true;
+      }
+      return false;
+    });
+  return { deprecated };
+}
+
+/** Resolve documented model capabilities from a name or endpoint summary. */
+export function modelCapabilitiesFor(model: string | ServingEndpointSummary): ModelCapabilities {
+  const capabilities = metadataIndex().capabilities;
+  const keys = modelIdentities(model, false).flatMap((identity) => modelKey(identity) ?? []);
+  return {
+    responses: keys.some((key) => capabilities.responses.has(key)),
+    imageInput: keys.some((key) => capabilities.imageInput.has(key)),
+    applyPatch: keys.some((key) => capabilities.applyPatch.has(key)),
+    webSearch: keys.some((key) => capabilities.webSearch.has(key)),
+  };
+}
+
+/** Resolve published ITPM, OTPM, and QPH limits from a model identity. */
+export function modelRateLimitsFor(
+  model: string | ServingEndpointSummary,
+): ModelRateLimits | undefined {
+  const limits = metadataIndex().rateLimits;
+  for (const identity of modelIdentities(model)) {
+    const key = modelKey(identity);
+    if (!key) continue;
+    const found = limits.get(key);
+    if (found) return found;
+  }
+  return undefined;
+}
+
+/** Resolve retirement, capability, and rate-limit metadata together. */
+export function modelMetadataFor(model: string | ServingEndpointSummary): ModelMetadata {
+  const rateLimits = modelRateLimitsFor(model);
+  return {
+    status: modelStatusFor(model),
+    capabilities: modelCapabilitiesFor(model),
+    ...(rateLimits ? { rateLimits } : {}),
+  };
+}
+
+/** Return generation timestamps for the three committed metadata snapshots. */
+export function modelMetadataGeneratedAt(): Readonly<{
+  retiredModels: number;
+  capabilities: number;
+  rateLimits: number;
+}> {
+  return {
+    retiredModels: retiredModelsSnapshot.generatedAt,
+    capabilities: capabilitiesSnapshot.generatedAt,
+    rateLimits: rateLimitsSnapshot.generatedAt,
+  };
+}
+
+function modelIdentities(
+  model: string | ServingEndpointSummary,
+  includeDisplayName = true,
+): string[] {
+  if (typeof model === "string") return [model];
+  return [
+    model.name,
+    ...(includeDisplayName ? [model.displayName] : []),
+    model.modelServiceName,
+    ...Object.values(model.serviceNames ?? {}),
+  ].filter((value): value is string => Boolean(value));
+}
+
+function modelKey(value: string): string | undefined {
+  return modelSearchQuery(value)?.replaceAll(" ", "-");
+}
+
+const RETIRED_MODEL_PREFIXES = new Set([
+  "ai",
+  "anthropic",
+  "databricks",
+  "dbx",
+  "google",
+  "meta",
+  "openai",
+  "system",
+]);
+
+function retiredModelKey(value: string): string {
+  const tokens = value.toLowerCase().match(/[a-z0-9]+/g) ?? [];
+  while (tokens[0] && RETIRED_MODEL_PREFIXES.has(tokens[0])) tokens.shift();
+  return tokens.join("-");
+}

package/src/model-catalog.ts

@@ -9,7 +9,7 @@
  * `quality` / `speed` / `cost` profile when present, the classified
  * {@link ModelClass}, and (for embedding endpoints) the measured vector
  * `dimension` - and snaps loose, human-typed names to real endpoint ids
- * through the Rust model ranker so tokens like `"claude sonnet"` resolve to
+ * through the TypeScript model ranker so tokens like `"claude sonnet"` resolve to
  * `databricks-claude-sonnet-4-6`.
  *
  * The class stamp and embedding dimension are computed once per cache load:
@@ -27,7 +27,7 @@ import { appkit } from "@dbx-tools/appkit";
 import { errorUtils, log } from "@dbx-tools/shared-core";
 import { model, type ServingEndpointSummary } from "@dbx-tools/shared-model";
 
-import { classifyEndpointClasses, normalizeEndpoints, rankEndpoints } from "./_native.ts";
+import { classifyEndpointClasses, normalizeEndpoints, rankEndpoints } from "./_ranking.ts";
 
 const { ModelClass } = model;
 
@@ -43,7 +43,7 @@ export type WorkspaceClientLike = Pick<appkit.WorkspaceClientLike, "servingEndpo
 /** Default TTL for the in-memory endpoint cache. Matches the Databricks SDK's session lifetime budget. */
 export const DEFAULT_MODEL_CACHE_TTL_MS = 5 * 60 * 1000;
 
-/** Default Rust distance threshold below which a fuzzy match is accepted. */
+/** Default fuzzy distance threshold below which a match is accepted. */
 export const DEFAULT_FUZZY_THRESHOLD = 0.4;
 
 /** Cache key parts under which endpoint listings are stored. */
@@ -129,46 +129,44 @@ export async function listServingEndpointsUncached(
 
 async function fetchEndpoints(client: WorkspaceClientLike): Promise<ServingEndpointSummary[]> {
   const startedAt = Date.now();
-  const out = await listServingEndpointsUncached(client);
-  stampModelClasses(out);
-  await measureEmbeddingDimensions(client, out);
+  const classified = stampModelClasses(await listServingEndpointsUncached(client));
+  const out = await measureEmbeddingDimensions(client, classified);
   logger.debug("listed", { count: out.length, elapsedMs: Date.now() - startedAt });
   return out;
 }
 
 /**
  * Stamp each summary's {@link ServingEndpointSummary.class} from the relative
- * classification of the whole set. Mutates `summaries` in place. Endpoints the
- * classifier doesn't recognize (custom, unscored, non-LLM) are left without a
- * class.
+ * classification of the whole set. Endpoints the classifier doesn't recognize
+ * (custom, unscored, non-LLM) are left without a class.
  */
-function stampModelClasses(summaries: ServingEndpointSummary[]): void {
+function stampModelClasses(
+  summaries: readonly ServingEndpointSummary[],
+): ServingEndpointSummary[] {
   const classOf = classifyEndpointClasses(summaries);
-  for (const summary of summaries) {
+  return summaries.map((summary) => {
     const cls = classOf.get(summary.name);
-    if (cls !== undefined) summary.class = cls;
-  }
+    return cls === undefined ? summary : { ...summary, class: cls };
+  });
 }
 
 /**
  * Measure the embedding vector dimension of every {@link ModelClass.Embedding}
- * endpoint by pinging it once, all in parallel. Mutates the matching summaries
- * in place with the resulting `dimension`. Runs only on a cache miss (it's
+ * endpoint by pinging it once, all in parallel. Runs only on a cache miss (it's
  * called from {@link fetchEndpoints}), so the probe cost is amortized across the
- * cached TTL window. Per-endpoint failures are swallowed (logged at warn) so
- * one unreachable embedding model never fails the listing.
+ * cached TTL window. Per-endpoint failures are swallowed (logged at warn) so one
+ * unreachable embedding model never fails the listing.
  */
 async function measureEmbeddingDimensions(
   client: WorkspaceClientLike,
-  summaries: ServingEndpointSummary[],
-): Promise<void> {
-  await Promise.all(
-    summaries
-      .filter((s) => s.class === ModelClass.Embedding)
-      .map(async (summary) => {
-        const dimension = await pingEmbeddingDimension(client, summary.name);
-        if (dimension !== undefined) summary.dimension = dimension;
-      }),
+  summaries: readonly ServingEndpointSummary[],
+): Promise<ServingEndpointSummary[]> {
+  return Promise.all(
+    summaries.map(async (summary) => {
+      if (summary.class !== ModelClass.Embedding) return summary;
+      const dimension = await pingEmbeddingDimension(client, summary.name);
+      return dimension === undefined ? summary : { ...summary, dimension };
+    }),
   );
 }
 
@@ -215,7 +213,7 @@ export async function clearServingEndpointsCache(
 
 /**
  * Result of fuzzy-resolving a user-supplied model name against the live
- * endpoint list. `score` is the Rust ranker's distance (`0` is exact, `1` is no match);
+ * endpoint list. `score` is the ranker's distance (`0` is exact, `1` is no match);
  * `matched` is `false` when the score exceeds the configured threshold so
  * callers can fall back to the original input (Databricks will then return a
  * clean 404).
@@ -246,7 +244,7 @@ export interface ScoredEndpoint {
  * (lowest score) first, keeping only those within `threshold`:
  *
  * 1. An exact name match short-circuits to a single `score: 0` result.
- * 2. Otherwise Rust tokenizes the input and scores every endpoint through the
+ * 2. Otherwise the shared TypeScript policy tokenizes and scores every endpoint through the
  *    same fuzzy-distance policy used by {@link lookupModels}.
  *
  * Returns `[]` for an empty endpoint list or when `input` tokenizes to nothing,