@sriinnu/kosha-discovery 0.6.0 → 1.0.0

package/dist/cli-commands.js

@@ -5,17 +5,21 @@
  * Formatting utilities are imported from `./cli-format.js` to keep
  * presentation logic separate from command orchestration.
  *
+ * This module also serves as the barrel re-export for commands that were
+ * split into dedicated files (`cli-cmd-model.ts`, `cli-cmd-query.ts`,
+ * `cli-help.ts`) to keep each file under 450 LOC.
+ *
  * @module cli-commands
  */
-import { BOLD, CYAN, DIM, GREEN, MAGENTA, RED, YELLOW, c, formatContextWindow, formatNumber, formatPrice, formatTimestamp, line, renderTable, } from "./cli-format.js";
-import { createRequire } from "node:module";
-const _require = createRequire(import.meta.url);
-const VERSION = _require("../package.json").version;
+import { BOLD, CYAN, DIM, GREEN, RED, YELLOW, c, formatContextWindow, formatPrice, formatPricingTier, formatRelativeTime, formatTimestamp, line, renderTable, } from "./cli-format.js";
+// ---------------------------------------------------------------------------
+//  Shared helpers — exported so split modules can import them
+// ---------------------------------------------------------------------------
 /**
  * Standard column layout reused by both `cmdList` and `cmdSearch`.
  * Extracted to avoid duplicating the same six-column definition.
  */
-const MODEL_TABLE_COLUMNS = [
+export const MODEL_TABLE_COLUMNS = [
     { header: "Provider", width: 12 },
     { header: "Model", width: 38 },
     { header: "Mode", width: 10 },
@@ -23,6 +27,22 @@ const MODEL_TABLE_COLUMNS = [
     { header: "$/M in", width: 8, align: "right" },
     { header: "$/M out", width: 8, align: "right" },
 ];
+/**
+ * Extended pricing column layout used when `--pricing` flag is passed.
+ * Replaces context-window with tier, cache, and batch columns.
+ */
+const PRICING_TABLE_COLUMNS = [
+    { header: "Provider", width: 12 },
+    { header: "Model", width: 34 },
+    { header: "Mode", width: 10 },
+    { header: "Tier", width: 6 },
+    { header: "$/M in", width: 8, align: "right" },
+    { header: "$/M out", width: 8, align: "right" },
+    { header: "Cache R", width: 8, align: "right" },
+    { header: "Cache W", width: 8, align: "right" },
+    { header: "Batch in", width: 8, align: "right" },
+    { header: "Batch out", width: 9, align: "right" },
+];
 /**
  * Return a human-readable label for a provider's credential source.
  * Providers that need no remote auth (e.g. local Ollama) show `"none (local)"`.
@@ -30,7 +50,7 @@ const MODEL_TABLE_COLUMNS = [
  * @param provider  The provider info object.
  * @returns         A short label suitable for table display.
  */
-function formatCredentialSource(provider) {
+export function formatCredentialSource(provider) {
     if (!provider.credentialSource || provider.credentialSource === "none") {
         return c(DIM, "none (local)");
     }
@@ -38,14 +58,37 @@ function formatCredentialSource(provider) {
 }
 /**
  * Ensure the registry has at least one provider loaded.
- * If empty (first run or stale cache), triggers a full discovery pass.
+ *
+ * The registry's `discover()` transparently hydrates from the on-disk cache
+ * at `~/.kosha/cache` when that cache is still within TTL, and only hits
+ * provider APIs on a cold start or a stale cache. I look at the resulting
+ * `discoveredAt` timestamp to tell the user which path actually ran — the
+ * old "No cached data. Running discovery..." message was misleading because
+ * it fired on every invocation regardless of whether the cache was used.
+ *
+ * Honesty here matters: users need to trust that `kosha list` is fast
+ * because the cache exists, and that `kosha update` is the way to refresh.
  *
  * @param registry  The shared model registry instance.
  */
-async function ensureDiscovered(registry) {
-    if (registry.providers_list().length === 0) {
-        console.log(c(DIM, "No cached data. Running discovery..."));
-        await registry.discover();
+export async function ensureDiscovered(registry) {
+    if (registry.providers_list().length > 0)
+        return;
+    const beforeMs = Date.now();
+    await registry.discover();
+    const { discoveredAt } = registry.toJSON();
+    const providers = registry.providers_list();
+    const totalModels = providers.reduce((sum, p) => sum + p.models.length, 0);
+    // If the timestamp on the hydrated registry is older than "basically now",
+    // we know the disk cache was loaded rather than a live discovery running.
+    const usedCache = discoveredAt > 0 && beforeMs - discoveredAt > 2_000;
+    if (usedCache) {
+        const age = formatRelativeTime(discoveredAt, beforeMs);
+        console.log(c(DIM, `Loaded ${totalModels} models from cache (${age}). Run "kosha update" to refresh.`));
+    }
+    else {
+        console.log(c(DIM, `Discovered ${totalModels} models from ${providers.length} providers.`));
+        console.log(c(DIM, `Saved to ~/.kosha/cache and exported manifest to ~/.kosha/registry.json`));
     }
 }
 /**
@@ -53,7 +96,7 @@ async function ensureDiscovered(registry) {
  * @param m  A model card object.
  * @returns  An array of formatted cell strings.
  */
-function modelRow(m) {
+export function modelRow(m) {
     return [
         c(CYAN, m.provider), m.id, m.mode,
         formatContextWindow(m.contextWindow),
@@ -61,13 +104,23 @@ function modelRow(m) {
         formatPrice(m.pricing?.outputPerMillion),
     ];
 }
-function parseNumberFlag(value) {
-    if (typeof value !== "string")
-        return undefined;
-    const n = Number(value);
-    if (!Number.isFinite(n))
-        return undefined;
-    return n;
+/**
+ * Build an extended pricing-table row from a model card.
+ * Includes tier label, cache read/write, and batch pricing columns.
+ * @param m  A model card with full pricing detail.
+ * @returns  An array of formatted cell strings for the pricing column layout.
+ */
+function pricingRow(m) {
+    return [
+        c(CYAN, m.provider), m.id, m.mode,
+        formatPricingTier(m.pricing),
+        formatPrice(m.pricing?.inputPerMillion),
+        formatPrice(m.pricing?.outputPerMillion),
+        formatPrice(m.pricing?.cacheReadPerMillion),
+        formatPrice(m.pricing?.cacheWritePerMillion),
+        formatPrice(m.pricing?.batchInputPerMillion),
+        formatPrice(m.pricing?.batchOutputPerMillion),
+    ];
 }
 /** Print a provider summary line (used by discover & refresh). */
 function printProviderSummary(providers) {
@@ -107,6 +160,7 @@ export async function cmdDiscover(registry, flags) {
         return;
     }
     printProviderSummary(providers);
+    console.log(c(DIM, `\nCached to ~/.kosha/cache  ·  Manifest: ~/.kosha/registry.json`));
 }
 // ── latest ───────────────────────────────────────────────────────────────
 /**
@@ -133,225 +187,72 @@ export async function cmdLatest(registry, flags) {
     printProviderSummary(result.providers);
     console.log(c(DIM, `\nFetched at: ${formatTimestamp(result.discoveredAt)}`));
 }
-// ── list ─────────────────────────────────────────────────────────────────
+// ── enrich ───────────────────────────────────────────────────────────────
 /**
- * List all known models in a formatted table.
- * Supports filtering by `--provider`, `--origin`, `--mode`, and `--capability`.
+ * Re-run LiteLLM enrichment on cached models without re-discovering providers.
  *
- * @param registry  The model registry to query.
- * @param flags     CLI flags (supports `--provider`, `--origin`, `--mode`, `--capability`, `--json`).
- */
-export async function cmdList(registry, flags) {
-    const provider = typeof flags.provider === "string" ? flags.provider : undefined;
-    const originProvider = typeof flags.origin === "string" ? flags.origin : undefined;
-    const mode = typeof flags.mode === "string" ? flags.mode : undefined;
-    const capability = typeof flags.capability === "string" ? flags.capability : undefined;
-    await ensureDiscovered(registry);
-    const models = registry.models({ provider, originProvider, mode, capability });
-    if (flags.json) {
-        console.log(JSON.stringify(models, null, 2));
-        return;
-    }
-    if (models.length === 0) {
-        console.log(c(YELLOW, "No models found matching the given filters."));
-        return;
-    }
-    console.log(renderTable(MODEL_TABLE_COLUMNS, models.map(modelRow)));
-    const providerCount = new Set(models.map((m) => m.provider)).size;
-    console.log(c(DIM, line("\u2500", 90)));
-    console.log(`${c(BOLD, String(models.length))} models from ${c(BOLD, String(providerCount))} providers`);
-}
-// ── roles ────────────────────────────────────────────────────────────────
-/**
- * Show a provider -> model -> roles matrix.
+ * This is the lightweight alternative to `kosha refresh` when you only need
+ * updated pricing data — no provider API calls, just a single fetch from the
+ * litellm community catalogue plus a cache + manifest update.
  *
- * Useful for assistants that need to answer: "which providers/models can do X?"
+ * @param registry  The model registry with cached data.
+ * @param flags     CLI flags (supports `--json`).
  */
-export async function cmdRoles(registry, flags) {
-    const provider = typeof flags.provider === "string" ? flags.provider : undefined;
-    const originProvider = typeof flags.origin === "string" ? flags.origin : undefined;
-    const mode = typeof flags.mode === "string" ? flags.mode : undefined;
-    const capability = typeof flags.capability === "string" ? flags.capability : undefined;
-    const role = typeof flags.role === "string" ? flags.role : undefined;
-    await ensureDiscovered(registry);
-    const providers = registry.providerRoles({ provider, originProvider, mode, capability, role });
-    if (flags.json) {
-        console.log(JSON.stringify({
-            providers,
-            count: providers.length,
-            modelCount: providers.reduce((sum, p) => sum + p.models.length, 0),
-            missingCredentials: registry.missingCredentialPrompts(providers.map((p) => p.id)),
-        }, null, 2));
-        return;
-    }
-    if (providers.length === 0) {
-        console.log(c(YELLOW, "No provider/model roles found for the given filters."));
-        return;
-    }
-    const columns = [
-        { header: "Provider", width: 12 },
-        { header: "Model", width: 40 },
-        { header: "Mode", width: 10 },
-        { header: "Roles", width: 56 },
-    ];
-    const rows = providers.flatMap((providerInfo) => providerInfo.models.map((model) => [
-        c(CYAN, providerInfo.id),
-        model.id,
-        model.mode,
-        model.roles.join(", "),
-    ]));
-    console.log(renderTable(columns, rows));
-    const missing = registry.missingCredentialPrompts(providers.map((p) => p.id));
-    if (missing.length > 0) {
-        console.log(c(DIM, "\nMissing provider credentials:"));
-        for (const prompt of missing) {
-            console.log(`  ${c(YELLOW, prompt.providerId)}: ${prompt.message}`);
-        }
+export async function cmdEnrich(registry, flags) {
+    console.log(c(DIM, "Re-enriching cached models with LiteLLM pricing data..."));
+    const result = await registry.enrichOnly();
+    if (result === null) {
+        console.error(c(YELLOW, "No cached data found. Run `kosha discover` first."));
+        process.exit(1);
     }
-}
-// ── cheapest ─────────────────────────────────────────────────────────────
-/**
- * Return cheapest model candidates for a requested role/capability.
- */
-export async function cmdCheapest(registry, flags) {
-    const provider = typeof flags.provider === "string" ? flags.provider : undefined;
-    const originProvider = typeof flags.origin === "string" ? flags.origin : undefined;
-    const mode = typeof flags.mode === "string" ? flags.mode : undefined;
-    const capability = typeof flags.capability === "string" ? flags.capability : undefined;
-    const role = typeof flags.role === "string" ? flags.role : undefined;
-    await ensureDiscovered(registry);
-    const result = registry.cheapestModels({
-        provider,
-        originProvider,
-        mode,
-        capability,
-        role,
-        limit: parseNumberFlag(flags.limit),
-        priceMetric: typeof flags["price-metric"] === "string"
-            ? flags["price-metric"]
-            : undefined,
-        inputWeight: parseNumberFlag(flags["input-weight"]),
-        outputWeight: parseNumberFlag(flags["output-weight"]),
-        includeUnpriced: flags["include-unpriced"] === true,
-    });
     if (flags.json) {
         console.log(JSON.stringify(result, null, 2));
         return;
     }
-    if (result.matches.length === 0) {
-        console.log(c(YELLOW, "No priced models found for the requested filters."));
-        if (result.missingCredentials.length > 0) {
-            console.log(c(DIM, "\nMissing provider credentials:"));
-            for (const prompt of result.missingCredentials) {
-                console.log(`  ${c(YELLOW, prompt.providerId)}: ${prompt.message}`);
-            }
-        }
-        return;
-    }
-    const columns = [
-        { header: "Provider", width: 12 },
-        { header: "Model", width: 38 },
-        { header: "Mode", width: 10 },
-        { header: "Metric", width: 8 },
-        { header: "Score", width: 10, align: "right" },
-        { header: "$/M in", width: 8, align: "right" },
-        { header: "$/M out", width: 8, align: "right" },
-    ];
-    const rows = result.matches.map((match) => [
-        c(CYAN, match.model.provider),
-        match.model.id,
-        match.model.mode,
-        match.priceMetric,
-        match.score === undefined ? "\u2014" : formatPrice(match.score),
-        formatPrice(match.model.pricing?.inputPerMillion),
-        formatPrice(match.model.pricing?.outputPerMillion),
-    ]);
-    console.log(renderTable(columns, rows));
-    console.log(c(DIM, `\n${result.pricedCandidates}/${result.candidates} candidates had usable pricing.`));
-    if (result.missingCredentials.length > 0) {
-        console.log(c(DIM, "\nMissing provider credentials:"));
-        for (const prompt of result.missingCredentials) {
-            console.log(`  ${c(YELLOW, prompt.providerId)}: ${prompt.message}`);
-        }
+    const totalModels = result.modelCount;
+    const cacheUpdated = result.cachePricingUpdated;
+    const batchUpdated = result.batchPricingUpdated;
+    console.log(c(DIM, `\nEnriched ${c(BOLD, String(totalModels))} models:`));
+    if (cacheUpdated > 0) {
+        console.log(c(DIM, `  Cache pricing updated: ${cacheUpdated} models`));
     }
-}
-// ── capabilities ─────────────────────────────────────────────────────────
-/**
- * Show an aggregated capability overview across all discovered models.
- * Each row shows a capability, how many models/providers support it,
- * and an example model ID for quick reference.
- *
- * @param registry  The model registry to query.
- * @param flags     CLI flags (supports `--provider`, `--json`).
- */
-export async function cmdCapabilities(registry, flags) {
-    const provider = typeof flags.provider === "string" ? flags.provider : undefined;
-    await ensureDiscovered(registry);
-    const caps = registry.capabilities({ provider });
-    if (flags.json) {
-        console.log(JSON.stringify(caps, null, 2));
-        return;
+    if (batchUpdated > 0) {
+        console.log(c(DIM, `  Batch pricing updated:  ${batchUpdated} models`));
     }
-    if (caps.length === 0) {
-        console.log(c(YELLOW, "No capabilities found."));
-        return;
+    if (cacheUpdated === 0 && batchUpdated === 0) {
+        console.log(c(DIM, "  No new pricing data found."));
     }
-    const columns = [
-        { header: "Capability", width: 20 },
-        { header: "Models", width: 8, align: "right" },
-        { header: "Providers", width: 10, align: "right" },
-        { header: "Example Model", width: 38 },
-    ];
-    const rows = caps.map((cap) => [
-        c(CYAN, cap.capability),
-        String(cap.modelCount),
-        String(cap.providerCount),
-        cap.exampleModelId ?? c(DIM, "—"),
-    ]);
-    const totalModels = registry.models({ provider }).length;
-    console.log(renderTable(columns, rows));
-    console.log(c(DIM, `\n${caps.length} capabilities across ${totalModels} models`));
+    console.log(c(DIM, `\nCache and manifest updated at ~/.kosha/`));
 }
-// ── capable ──────────────────────────────────────────────────────────────
+// ── list ─────────────────────────────────────────────────────────────────
 /**
- * Show models that support a given capability/role.
- * Normalizes the query so aliases work: "embeddings", "stt", "tools", "vision".
+ * List all known models in a formatted table.
+ * Supports filtering by `--provider`, `--origin`, `--mode`, and `--capability`.
  *
  * @param registry  The model registry to query.
- * @param query     The capability query (e.g. "vision", "embeddings", "tools").
- * @param flags     CLI flags (supports `--provider`, `--origin`, `--json`, `--limit`).
+ * @param flags     CLI flags (supports `--provider`, `--origin`, `--mode`, `--capability`, `--pricing`, `--json`).
  */
-export async function cmdCapable(registry, query, flags) {
-    if (!query) {
-        console.error(c(RED, "Usage: kosha capable <capability>"));
-        process.exit(1);
-    }
+export async function cmdList(registry, flags) {
     const provider = typeof flags.provider === "string" ? flags.provider : undefined;
     const originProvider = typeof flags.origin === "string" ? flags.origin : undefined;
     const mode = typeof flags.mode === "string" ? flags.mode : undefined;
-    const limit = parseNumberFlag(flags.limit);
+    const capability = typeof flags.capability === "string" ? flags.capability : undefined;
+    const showPricing = flags.pricing === true;
     await ensureDiscovered(registry);
-    const normalized = registry.normalizeRoleToken(query);
-    const allModels = registry.models({ provider, originProvider, mode });
-    let models = allModels.filter((m) => registry.modelSupportsRole(m, normalized));
-    if (limit !== undefined && limit > 0) {
-        models = models.slice(0, limit);
-    }
+    const models = registry.models({ provider, originProvider, mode, capability });
     if (flags.json) {
         console.log(JSON.stringify(models, null, 2));
         return;
     }
     if (models.length === 0) {
-        console.log(c(YELLOW, `No models found with capability "${query}"${normalized !== query ? ` (normalized: "${normalized}")` : ""}`));
+        console.log(c(YELLOW, "No models found matching the given filters."));
         return;
     }
-    const header = normalized !== query
-        ? `Models with capability ${c(CYAN, normalized)} ${c(DIM, `(from "${query}")`)}`
-        : `Models with capability ${c(CYAN, normalized)}`;
-    console.log(`\n${c(BOLD, header)}\n`);
-    console.log(renderTable(MODEL_TABLE_COLUMNS, models.map(modelRow)));
+    const listColumns = showPricing ? PRICING_TABLE_COLUMNS : MODEL_TABLE_COLUMNS;
+    const listRowFn = showPricing ? pricingRow : modelRow;
+    console.log(renderTable(listColumns, models.map(listRowFn)));
     const providerCount = new Set(models.map((m) => m.provider)).size;
-    console.log(c(DIM, line("\u2500", 90)));
+    console.log(c(DIM, line("\u2500", showPricing ? 110 : 90)));
     console.log(`${c(BOLD, String(models.length))} models from ${c(BOLD, String(providerCount))} providers`);
 }
 // ── search ───────────────────────────────────────────────────────────────
@@ -361,7 +262,7 @@ export async function cmdCapable(registry, query, flags) {
  *
  * @param registry  The model registry to search.
  * @param query     The search term (substring).
- * @param flags     CLI flags (supports `--origin`, `--json`).
+ * @param flags     CLI flags (supports `--origin`, `--pricing`, `--json`).
  */
 export async function cmdSearch(registry, query, flags) {
     if (!query) {
@@ -371,6 +272,7 @@ export async function cmdSearch(registry, query, flags) {
     await ensureDiscovered(registry);
     const needle = query.toLowerCase();
     const originProvider = typeof flags.origin === "string" ? flags.origin : undefined;
+    const showPricing = flags.pricing === true;
     // Apply optional origin filter first, then substring-match across id, name, and aliases
     const matches = registry.models({ originProvider }).filter((m) => m.id.toLowerCase().includes(needle) ||
         m.name.toLowerCase().includes(needle) ||
@@ -383,89 +285,11 @@ export async function cmdSearch(registry, query, flags) {
         console.log(c(YELLOW, `No models matching "${query}"`));
         return;
     }
-    console.log(renderTable(MODEL_TABLE_COLUMNS, matches.map(modelRow)));
+    const searchColumns = showPricing ? PRICING_TABLE_COLUMNS : MODEL_TABLE_COLUMNS;
+    const searchRowFn = showPricing ? pricingRow : modelRow;
+    console.log(renderTable(searchColumns, matches.map(searchRowFn)));
     console.log(c(DIM, `\n${matches.length} result${matches.length !== 1 ? "s" : ""} for "${query}"`));
 }
-// ── model (detail view) ─────────────────────────────────────────────────
-/**
- * Show detailed information for a single model identified by id or alias.
- *
- * If an exact match is not found, falls back to fuzzy substring search.
- * Single fuzzy match is shown automatically; multiple matches produce a
- * disambiguation list (capped at 5 suggestions).
- *
- * @param registry   The model registry to look up.
- * @param idOrAlias  A model id, display name, or alias.
- * @param flags      CLI flags (supports `--json`).
- */
-export async function cmdModel(registry, idOrAlias, flags) {
-    if (!idOrAlias) {
-        console.error(c(RED, "Usage: kosha model <id|alias>"));
-        process.exit(1);
-    }
-    await ensureDiscovered(registry);
-    const model = registry.model(idOrAlias);
-    if (!model) {
-        // Fuzzy fallback — substring match across id, name, and aliases so users
-        // can type fragments like "sonnet" without knowing the full canonical ID.
-        const needle = idOrAlias.toLowerCase();
-        const fuzzy = registry.models().filter((m) => m.id.toLowerCase().includes(needle) ||
-            m.name.toLowerCase().includes(needle) ||
-            m.aliases.some((a) => a.toLowerCase().includes(needle)));
-        if (fuzzy.length === 1)
-            return cmdModel(registry, fuzzy[0].id, flags);
-        if (fuzzy.length > 1) {
-            console.log(c(YELLOW, `No exact match for "${idOrAlias}". Did you mean:`));
-            for (const m of fuzzy.slice(0, 5))
-                console.log(`  ${c(CYAN, m.id)} (${m.provider})`);
-            return;
-        }
-        console.error(c(RED, `Model not found: "${idOrAlias}"`));
-        process.exit(1);
-    }
-    if (flags.json) {
-        console.log(JSON.stringify(model, null, 2));
-        return;
-    }
-    const providerName = registry.provider(model.provider)?.name ?? model.provider;
-    const pricingStr = model.pricing
-        ? `${formatPrice(model.pricing.inputPerMillion)} / ${formatPrice(model.pricing.outputPerMillion)} per million tokens (in/out)`
-        : c(DIM, "unknown");
-    // Cache pricing is optional — only displayed when the model has it
-    const cacheStr = model.pricing?.cacheReadPerMillion !== undefined
-        ? `\nCache Pricing: ${formatPrice(model.pricing.cacheReadPerMillion)} read / ${formatPrice(model.pricing.cacheWritePerMillion)} write per million tokens`
-        : "";
-    // Reasoning pricing is optional — shown when providers expose it.
-    const reasoningStr = model.pricing?.reasoningInputPerMillion !== undefined || model.pricing?.reasoningOutputPerMillion !== undefined
-        ? `\nReasoning Pricing: ${formatPrice(model.pricing?.reasoningInputPerMillion)} in / ${formatPrice(model.pricing?.reasoningOutputPerMillion)} out per million tokens`
-        : "";
-    const originPricingStr = model.originPricing
-        ? `\nOrigin Pricing: ${formatPrice(model.originPricing.inputPerMillion)} / ${formatPrice(model.originPricing.outputPerMillion)} per million tokens (in/out)` +
-            ((model.originPricing.reasoningInputPerMillion !== undefined || model.originPricing.reasoningOutputPerMillion !== undefined)
-                ? `\nOrigin Reasoning Pricing: ${formatPrice(model.originPricing.reasoningInputPerMillion)} in / ${formatPrice(model.originPricing.reasoningOutputPerMillion)} out per million tokens`
-                : "")
-        : "";
-    // Origin provider line: only shown when distinct from the serving-layer provider
-    const originLine = model.originProvider && model.originProvider !== model.provider
-        ? `\n${c(BOLD, "Origin Provider:")} ${c(CYAN, model.originProvider)}`
-        : "";
-    // Region and project ID are only shown when present (Bedrock / Vertex models)
-    const regionLine = model.region ? `\n${c(BOLD, "Region:")} ${model.region}` : "";
-    const projectLine = model.projectId ? `\n${c(BOLD, "Project ID:")} ${model.projectId}` : "";
-    console.log(`
-${c(BOLD, "Model:")} ${model.id}
-${c(BOLD, "Name:")} ${model.name}
-${c(BOLD, "Provider:")} ${providerName}${originLine}${regionLine}${projectLine}
-${c(BOLD, "Mode:")} ${model.mode}
-${c(BOLD, "Aliases:")} ${model.aliases.length > 0 ? model.aliases.join(", ") : c(DIM, "none")}
-${c(BOLD, "Context Window:")} ${model.contextWindow > 0 ? formatNumber(model.contextWindow) + " tokens" : c(DIM, "unknown")}
-${c(BOLD, "Max Output:")} ${model.maxOutputTokens > 0 ? formatNumber(model.maxOutputTokens) + " tokens" : c(DIM, "unknown")}${model.dimensions ? `\n${c(BOLD, "Dimensions:")} ${formatNumber(model.dimensions)}` : ""}
-${c(BOLD, "Capabilities:")} ${model.capabilities.join(", ")}
-${c(BOLD, "Pricing:")} ${pricingStr}${cacheStr}${reasoningStr}${originPricingStr}
-${c(BOLD, "Source:")} ${model.source}
-${c(BOLD, "Discovered:")} ${formatTimestamp(model.discoveredAt)}
-`.trim());
-}
 // ── providers ────────────────────────────────────────────────────────────
 /**
  * List all known providers with their authentication status and model counts.
@@ -501,92 +325,6 @@ export async function cmdProviders(registry, flags) {
     });
     console.log(renderTable(columns, rows));
 }
-// ── resolve ──────────────────────────────────────────────────────────────
-/**
- * Resolve a model alias to its canonical provider ID.
- * Unknown aliases are returned as-is with a yellow warning.
- *
- * @param registry  The model registry containing alias mappings.
- * @param alias     The alias string to resolve.
- * @param flags     CLI flags (supports `--json`).
- */
-export async function cmdResolve(registry, alias, flags) {
-    if (!alias) {
-        console.error(c(RED, "Usage: kosha resolve <alias>"));
-        process.exit(1);
-    }
-    await ensureDiscovered(registry);
-    const resolved = registry.resolve(alias);
-    if (flags.json) {
-        console.log(JSON.stringify({ alias, resolved }, null, 2));
-        return;
-    }
-    if (resolved === alias) {
-        // resolve() returns input unchanged when no alias mapping exists
-        console.log(c(YELLOW, `"${alias}" is not a known alias (returned as-is)`));
-    }
-    else {
-        console.log(`${c(DIM, alias)} ${c(DIM, "\u2192")} ${c(CYAN, resolved)}`);
-    }
-}
-// ── routes ───────────────────────────────────────────────────────────────
-/**
- * Show every provider route through which a model can be accessed.
- *
- * A "route" is a serving-layer entry whose normalized model ID matches the
- * given identifier. Output includes direct/preferred markers, origin provider,
- * model version hint, and serving base URL.
- *
- * Pricing columns show `—` when data is unavailable.
- *
- * @param registry   The model registry to query.
- * @param modelId    Canonical model ID or alias to look up.
- * @param flags      CLI flags (supports `--json`).
- */
-export async function cmdRoutes(registry, modelId, flags) {
-    if (!modelId) {
-        console.error(c(RED, "Usage: kosha routes <model-id|alias>"));
-        process.exit(1);
-    }
-    await ensureDiscovered(registry);
-    const routes = registry.modelRouteInfo(modelId);
-    if (flags.json) {
-        console.log(JSON.stringify(routes, null, 2));
-        return;
-    }
-    if (routes.length === 0) {
-        console.error(c(RED, `No routes found for model: "${modelId}"`));
-        process.exit(1);
-    }
-    const columns = [
-        { header: "Pref", width: 4 },
-        { header: "Provider", width: 14 },
-        { header: "Model ID", width: 42 },
-        { header: "Region", width: 14 },
-        { header: "Origin", width: 12 },
-        { header: "Ver", width: 12 },
-        { header: "Base URL", width: 34 },
-        { header: "$/M in", width: 8, align: "right" },
-        { header: "$/M out", width: 8, align: "right" },
-    ];
-    const rows = routes.map((route) => [
-        route.isPreferred ? c(GREEN, "*") : route.isDirect ? c(CYAN, "\u00B7") : " ",
-        c(CYAN, route.provider),
-        route.model.id,
-        route.model.region ?? c(DIM, "—"),
-        route.originProvider ?? c(DIM, "—"),
-        route.version ?? c(DIM, "—"),
-        route.baseUrl ?? c(DIM, "—"),
-        formatPrice(route.model.pricing?.inputPerMillion),
-        formatPrice(route.model.pricing?.outputPerMillion),
-    ]);
-    console.log(`\n${c(BOLD, routes[0].model.name)} ${c(DIM, `(${routes.length} route${routes.length !== 1 ? "s" : ""})`)}\n`);
-    console.log(renderTable(columns, rows));
-    const preferred = routes.find((route) => route.isPreferred);
-    if (preferred) {
-        console.log(c(DIM, `\nPreferred route: ${preferred.provider} (${preferred.baseUrl ?? "base URL unknown"})`));
-    }
-}
 // ── refresh ──────────────────────────────────────────────────────────────
 /**
  * Force a full re-discovery of all providers, bypassing the cache.
@@ -607,8 +345,7 @@ export async function cmdRefresh(registry, flags) {
     }
     const providers = registry.providers_list();
     printProviderSummary(providers);
-    // Overwrite the last summary line to include "Refreshed:" prefix
-    // (printProviderSummary already printed the count, so we add context)
+    console.log(c(DIM, `\nManifest exported to ~/.kosha/registry.json`));
 }
 // ── serve ────────────────────────────────────────────────────────────────
 /**
@@ -628,129 +365,10 @@ export async function cmdServe(flags) {
     const { startServer } = await import("./server.js");
     await startServer(port);
 }
-// ── help & version ───────────────────────────────────────────────────────
-/** Print the full CLI usage / help text to stdout. */
-export function showHelp() {
-    console.log(`
-${c(BOLD, "kosha")} ${c(DIM, "\u2014 AI Model & Provider Discovery Registry")}
-
-${c(BOLD, "USAGE")}
-  kosha <command> [options]
-
-${c(BOLD, "COMMANDS")}
-  ${c(CYAN, "discover")}                      Discover all providers and models
-  ${c(CYAN, "list")}                          List all known models
-    --provider <name>             Filter by serving-layer provider
-    --origin <name>               Filter by origin/creator provider (e.g. anthropic)
-    --mode <mode>                 Filter by mode (chat, embedding, image, audio)
-    --capability <cap>            Filter by capability (vision, function_calling, etc.)
-  ${c(CYAN, "search")} <query>                Search models by name/ID (fuzzy match)
-    --origin <name>               Restrict search to a specific origin provider
-  ${c(CYAN, "model")} <id|alias>              Show detailed info for one model
-  ${c(CYAN, "roles")}                         Show provider -> model -> roles matrix
-    --role <role>                 Filter by task role (e.g. embeddings, image, tool_use)
-    --provider <name>             Filter by serving-layer provider
-    --origin <name>               Filter by model creator provider
-    --mode <mode>                 Filter by mode (chat, embedding, image, audio, moderation)
-    --capability <cap>            Filter by capability tag
-  ${c(CYAN, "capabilities")} ${c(DIM, "(caps)")}             Show all capabilities across the ecosystem
-    --provider <name>             Scope to one provider
-  ${c(CYAN, "capable")} <capability>            List models with a given capability
-    --provider <name>             Filter by serving-layer provider
-    --origin <name>               Filter by origin/creator provider
-    --mode <mode>                 Filter by mode (chat, embedding, image, audio)
-    --limit <n>                   Maximum models to show
-  ${c(CYAN, "cheapest")}                      Find cheapest eligible models
-    --role <role>                 Task role, e.g. embeddings or image
-    --capability <cap>            Capability filter (vision, embedding, function_calling)
-    --mode <mode>                 Mode filter
-    --limit <n>                   Maximum matches to return (default 5)
-    --price-metric <metric>       input | output | blended
-    --input-weight <n>            Weight for blended metric input price
-    --output-weight <n>           Weight for blended metric output price
-    --include-unpriced            Include unpriced models after ranked matches
-  ${c(CYAN, "routes")} <id|alias>             Show all provider routes for a model
-  ${c(CYAN, "providers")}                     List all providers and their status
-  ${c(CYAN, "resolve")} <alias>               Resolve an alias to canonical model ID
-  ${c(CYAN, "latest")}                        Force-fetch latest model/provider details
-    --provider <name>             Scope latest fetch to one provider
-  ${c(CYAN, "refresh")}                       Force re-discover all providers (bypass cache)
-    --provider <name>             Refresh only one provider
-  ${c(CYAN, "serve")} [--port 3000]           Start HTTP API server
-
-${c(BOLD, "OPTIONS")}
-  --json                          Output as JSON (works with any command)
-  --help                          Show this help message
-  --version                       Show version
-
-${c(BOLD, "EXAMPLES")}
-  ${c(DIM, "$")} kosha discover
-  ${c(DIM, "$")} kosha list --provider anthropic
-  ${c(DIM, "$")} kosha list --origin anthropic
-  ${c(DIM, "$")} kosha list --mode embedding --json
-  ${c(DIM, "$")} kosha search gemini
-  ${c(DIM, "$")} kosha search claude --origin anthropic
-  ${c(DIM, "$")} kosha model sonnet
-  ${c(DIM, "$")} kosha roles --role embeddings
-  ${c(DIM, "$")} kosha capabilities
-  ${c(DIM, "$")} kosha capable vision
-  ${c(DIM, "$")} kosha capable embeddings --limit 5
-  ${c(DIM, "$")} kosha cheapest --role image --limit 3
-  ${c(DIM, "$")} kosha routes claude-opus-4-6
-  ${c(DIM, "$")} kosha routes gpt-4o --json
-  ${c(DIM, "$")} kosha providers
-  ${c(DIM, "$")} kosha latest
-  ${c(DIM, "$")} kosha latest --provider openai --json
-  ${c(DIM, "$")} kosha resolve haiku
-  ${c(DIM, "$")} kosha refresh --provider anthropic
-  ${c(DIM, "$")} kosha serve --port 8080
-`.trim());
-}
-/** Print the CLI version string to stdout. */
-export function showVersion() {
-    console.log(`kosha-discovery v${VERSION}`);
-}
-/**
- * Display a branded splash screen when `kosha` is invoked with no arguments.
- *
- * Shows the Kosha logo, tagline, version, and quick-start commands.
- * Uses MAGENTA branding with a clean, minimal layout.
- */
-export function showSplash() {
-    const brandWord = `${c(CYAN, "k")}${c(GREEN, "o")}${c(YELLOW, "s")}${c(MAGENTA, "h")}${c(RED, "a")}`;
-    const mascot1 = `${c(CYAN, " /\\_/\\ ")} ${c(DIM, "assistant mascot")}`;
-    const mascot2 = `${c(CYAN, "( o.o )")} ${c(DIM, "ready to route")}`;
-    const mascot3 = `${c(CYAN, " > ^ < ")} ${c(DIM, "providers + models")}`;
-    console.log(`
-${c(MAGENTA, "  ╔═══════════════════════════════════════════════════╗")}
-${c(MAGENTA, "  ║")}                                                   ${c(MAGENTA, "║")}
-${c(MAGENTA, "  ║")}   ${c(BOLD, "  █▄▀ █▀█ █▀ █ █ ▄▀█")}                         ${c(MAGENTA, "║")}
-${c(MAGENTA, "  ║")}   ${c(BOLD, "  █ █ █▄█ ▄█ █▀█ █▀█")}    ${c(DIM, "कोश — treasury")}        ${c(MAGENTA, "║")}
-${c(MAGENTA, "  ║")}                                                   ${c(MAGENTA, "║")}
-${c(MAGENTA, "  ║")}   ${brandWord} ${c(DIM, "AI Model & Provider Discovery Registry")}  ${c(MAGENTA, "║")}
-${c(MAGENTA, "  ║")}   ${c(DIM, `v${VERSION}`)}                                          ${c(MAGENTA, "║")}
-${c(MAGENTA, "  ║")}   ${mascot1}                                ${c(MAGENTA, "║")}
-${c(MAGENTA, "  ║")}   ${mascot2}                                ${c(MAGENTA, "║")}
-${c(MAGENTA, "  ║")}   ${mascot3}                                ${c(MAGENTA, "║")}
-${c(MAGENTA, "  ║")}                                                   ${c(MAGENTA, "║")}
-${c(MAGENTA, "  ╚═══════════════════════════════════════════════════╝")}
-
-  ${c(BOLD, "Quick start:")}
-
-    ${c(CYAN, "kosha discover")}       Scan all providers for models
-    ${c(CYAN, "kosha list")}           List all discovered models
-    ${c(CYAN, "kosha search")} ${c(DIM, "<q>")}     Search by name or ID
-    ${c(CYAN, "kosha model")} ${c(DIM, "<id>")}     Detailed info for one model
-    ${c(CYAN, "kosha capabilities")}   What capabilities exist?
-    ${c(CYAN, "kosha capable")} ${c(DIM, "<cap>")}  Models with a given capability
-    ${c(CYAN, "kosha roles")}          Provider -> model -> roles matrix
-    ${c(CYAN, "kosha cheapest")}       Cheapest models for a role
-    ${c(CYAN, "kosha routes")} ${c(DIM, "<id>")}    All provider routes for a model
-    ${c(CYAN, "kosha providers")}      Show provider status
-    ${c(CYAN, "kosha latest")}         Force-fetch latest provider/model details
-    ${c(CYAN, "kosha serve")}          Start the HTTP API server
-
-  ${c(DIM, "Run")} ${c(CYAN, "kosha --help")} ${c(DIM, "for full usage.")}
-`);
-}
+// ---------------------------------------------------------------------------
+//  Re-exports from split modules
+// ---------------------------------------------------------------------------
+export { cmdModel, cmdResolve, cmdRoutes } from "./cli-cmd-model.js";
+export { cmdRoles, cmdCheapest, cmdCapabilities, cmdCapable } from "./cli-cmd-query.js";
+export { showHelp, showVersion, showSplash } from "./cli-help.js";
 //# sourceMappingURL=cli-commands.js.map