@d-ai/magpie 0.1.1 → 0.1.2

package/README.md

@@ -1,16 +1,18 @@
 # D.AI for magpie / OpenCode
 
-Use your [D.AI](https://dai.itbug.shop) models — GPT-6, Claude Opus 5.5, Grok 4.7,
-DeepSeek V4.1, GLM 5.3, Gemini 3.8 — from any agent that speaks through
-[magpie](https://usemagpie.ai) or OpenCode.
+Use your [D.AI](https://dai.itbug.shop) models — GPT-6, Claude, Grok, DeepSeek,
+GLM and Gemini — from any agent that speaks through
+[magpie](https://usemagpie.ai) or OpenCode. The same plugin exposes a read-only
+catalog of D.AI image, video and audio models to Magpie's plugin-tools MCP.
 
 One `ddsk_…` API key is one account, and magpie keeps several.
 
 ## Add it
 
 ```sh
-magpie plugin add mdddj/d-ai-magpie       # from GitHub (the unofficial list installs this)
+magpie plugin add mdddj/d-ai-magpie       # from GitHub
 magpie plugin add @d-ai/magpie            # from npm, once published
+magpie plugin add ./sdks/d-ai-magpie      # from a folder, for development
 magpie plugin login dai                   # asks for the API key
 magpie provider test dai                  # one tiny call per API
 ```
@@ -22,17 +24,13 @@ The key comes from the D.AI console (`/#/apikeys`). Custom deployments can point
 at another host with `magpie plugin options dai '{"baseUrl":"https://dai.example"}'`
 or `DAI_BASE_URL`.
 
-`DAI_API_KEY=ddsk_… ./e2e.sh` runs the same checks in a throwaway `HOME`, so it
-never touches the magpie you actually use. Add `--live` to make one real call
-on each API.
+## Chat models
 
-## What it exposes
-
-15 token-metered chat models, in two APIs:
+The live catalog currently publishes 17 token-metered chat models, in two APIs:
 
 | API | Models | Path |
 | --- | --- | --- |
-| OpenAI chat completions | `deepseek-v4.1-flash`, `gemini-3.8-flash`, `glm-5.3-flash`, `gpt-5.6-luna`, `gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-6-astra`, `gpt-6-sol`, `grok-4.6`, `grok-4.7` | `/v1/chat/completions` |
+| OpenAI chat completions | `deepseek-v4.1-flash`, `deepseek-v4.1-flash-1017`, `gemini-3.8-flash`, `glm-5.3-flash`, `gpt-5.6-luna/sol/terra`, `gpt-6-astra/sol`, `gpt-6.1-sol`, `grok-4.6/4.7` | `/v1/chat/completions` |
 | Anthropic Messages | `claude-fable-5`, `claude-opus-4-7`, `claude-opus-4-8`, `claude-opus-5`, `claude-opus-5-5` | `/v1/messages` |
 
 The model list and the prices come from D.AI's public catalog on every start
@@ -42,15 +40,48 @@ new model appears without a plugin update. A built-in list is used when D.AI
 cannot be reached while the provider is first configured.
 
 Every model reports its context window, its output cap, its USD/1M-token prices
-and its thinking levels. Image input is marked for the families D.AI serves with
-vision (GPT-5/6, Claude, Gemini, Grok, DeepSeek, GLM).
+and its thinking levels. Image input is marked for GPT-5/6, Claude, DeepSeek and
+GLM, matching the modality table used by `sdks/pi`.
+
+## Media catalog
+
+The plugin registers one free, read-only Magpie tool, `media_catalog`. Add
+Magpie's plugin-tools MCP server to an agent:
+
+```sh
+claude mcp add --transport http magpie-tools http://127.0.0.1:3425/mcp/magpie/tools
+```
+
+Then ask for models by kind, capability, model code or search text:
+
+```json
+{ "kind": "image", "query": "gpt-image", "limit": 10 }
+{ "kind": "video", "capabilityCode": "gemini_video" }
+{ "kind": "image", "capabilityCode": "midjourney", "serviceCode": "midjourney_imagine" }
+{ "kind": "audio", "capabilityCode": "suno_speech", "model": "suno-speech" }
+```
+
+It reads `GET /api/system/public/gateway-models`, so it does not need an API key
+and does not spend wallet balance. Each result carries the executable
+`submitPath`, the optional `fetchPath`, `resultKind`, the complete parameter
+schema and the public `tierRates`. Some models (Midjourney, HappyHorse, Gemini
+Video and FlowMusic) have several operation variants: filter with
+`serviceCode` to choose one.
+
+The short plugin name depends on how it was installed: a folder named
+`d-ai-magpie` exposes `d-ai-magpie_media_catalog`; the published
+`@d-ai/magpie` package exposes `magpie_media_catalog`.
+
+Paid generation and account-scoped job tools are intentionally not included.
+Magpie's tool process does not receive the signed-in `dai` provider's API key,
+so adding generation here would require a separate tool-auth design.
 
 ## Usage
 
-`magpie quota` shows the account behind the key: the wallet balance, the tier
-and its cycle, how much of the balance this key alone may spend, and how many
-calls it made today. That comes from `GET /api/ai/account`, an endpoint
-authenticated by the `ddsk_` key itself.
+`magpie quota` shows the account behind the key: the wallet balance and pending
+reservation, the tier and its cycle, included USD, the key's own reservation and
+daily limits, and how many calls it made today. That comes from
+`GET /api/ai/account`, an endpoint authenticated by the `ddsk_` key itself.
 
 A D.AI key has no plan window — it spends the account's shared wallet — so the
 card reads as a balance, not a quota window. Two keys of one account show the
@@ -58,12 +89,16 @@ same balance but their own reservations and call counts.
 
 ## Notes
 
-- Only chat models. D.AI's image, video, music and ASR models are separate
-  capabilities with their own endpoints; agents here want a chat completion.
+- Chat models use the provider API. Image, video and audio models are separate
+  gateway capabilities; `media_catalog` returns their actual submit/fetch
+  paths without pretending they are chat completions. ASR is not in the public
+  media gateway today; use the authenticated `/api/audio/asr/v1/transcribe`
+  endpoint when a deployment enables Fun-ASR.
 - The usage card needs `GET /api/ai/account` on the D.AI deployment. Against an
   older one (404) the plugin falls back to the key's call log and shows call
   counts instead of a balance.
 - The plugin never rewrites request bodies. Thinking levels are offered as
   model variants, and the gateway passes `reasoning_effort` through.
-- Requests go out with `Authorization: Bearer <key>` plus `x-api-key`, so
-  either D.AI route works.
+- Requests go out with `Authorization: Bearer <key>` plus `x-api-key`, and carry
+  `x-dai-client: magpie/0.1.2`. Gateway requests also carry the stable Magpie
+  session as `x-dai-session`.

package/index.mjs

@@ -13,9 +13,9 @@
 // used. The older JWT-only endpoints (`/api/usd/me`, `/api/user/usage-stats`)
 // cannot serve a plugin, and the key's call log only counts calls, not money.
 //
-// Facts the tables below encode (checked 2026-10-15):
+// Facts the tables below encode (checked 2026-10-22):
 //   - The capability codes carrying token-metered chat are
-//     `openai_chat_completions` (10 models) and `claude_messages` (5 models).
+//     `openai_chat_completions` (12 models) and `claude_messages` (5 models).
 //   - `/v1/models` takes the API key as `Authorization: Bearer` or `x-api-key`
 //     and lists only the models that key may use.
 //   - `prepay_completion_token_limit` is D.AI's completion cap and the value a
@@ -26,14 +26,21 @@
 //     until someone checks it. D.AI's own catalog is not consulted for either:
 //     it carries no per-model reasoning or modality metadata.
 //
-// Sections: constants · helpers · catalog · plugin.
+// The same module also gives a read-only media catalog tool. Magpie tools run
+// from a separate MCP server and do not receive the provider account's key, so
+// paid generation and account-scoped job tools are intentionally absent.
+//
+// Sections: constants · helpers · chat catalog · media catalog · plugin.
 
 const PROVIDER = "dai";
+const CLIENT = "magpie/0.1.2";
 const DEFAULT_BASE = "https://dai.itbug.shop";
 const OPENAI_NPM = "@ai-sdk/openai-compatible";
 const ANTHROPIC_NPM = "@ai-sdk/anthropic";
 const CATALOG_TTL_MS = 5 * 60 * 1000;
 const REQUEST_TIMEOUT_MS = 20_000;
+const MEDIA_CATALOG_LIMIT = 20;
+const MEDIA_KINDS = new Set(["image", "video", "audio"]);
 
 /** Capability codes whose models are token-metered chat models. */
 const CHAT_CAPABILITIES = new Set(["openai_chat_completions", "claude_messages"]);
@@ -47,8 +54,10 @@ const VARIANTS = {
 	// GPT-6 cannot turn thinking off.
 	"gpt-6-astra": ["low", "medium", "high", "xhigh", "max"],
 	"gpt-6-sol": ["low", "medium", "high", "xhigh", "max"],
+	"gpt-6.1-sol": ["low", "medium", "high", "xhigh", "max"],
 	// River rejects minimal, medium and xhigh on DeepSeek-V4.1-Flash; `none` is off.
 	"deepseek-v4.1-flash": ["none", "low", "high", "max"],
+	"deepseek-v4.1-flash-1017": ["none", "low", "high", "max"],
 	// GLM-5.3-Flash cannot disable reasoning; River also rejects minimal/medium/xhigh.
 	"glm-5.3-flash": ["low", "high", "max"],
 	// Claude: the adaptive-effort models take xhigh/max, Opus 5.5 the full range.
@@ -60,10 +69,9 @@ const VARIANTS = {
 };
 
 /**
- * Models that take images as well as text, same set pi's tables mark.
- * Gemini, Grok and GPT-5.6 each carry `cache_write_usd_per_1m_tokens: null`
- * on D.AI today, so a null there reads as "no cache write price", not as a
- * missing check: only listed, non-zero rates become a model's cache price.
+ * Models that take images as well as text. D.AI serves GPT-5/6, Claude,
+ * DeepSeek and GLM with image input. Gemini and Grok are text-only through
+ * this gateway until pi's modality table marks them otherwise.
  */
 const VISION = /^(?:gpt-5|gpt-6|deepseek-|glm-|claude-)/u;
 
@@ -74,6 +82,7 @@ const VISION = /^(?:gpt-5|gpt-6|deepseek-|glm-|claude-)/u;
  */
 const FALLBACK_ROWS = [
 	["deepseek-v4.1-flash", 1_048_576, 128_000, 0.15, 0.6, 0.003],
+	["deepseek-v4.1-flash-1017", 1_048_576, 128_000, 0.15, 0.6, 0.003],
 	["gemini-3.8-flash", 1_050_000, 128_000, 0.5, 3, 0.05],
 	["glm-5.3-flash", 1_048_576, 128_000, 0.15, 0.5, 0.03],
 	["gpt-5.6-luna", 1_050_000, 128_000, 0.2, 1.2, 0.02],
@@ -81,6 +90,7 @@ const FALLBACK_ROWS = [
 	["gpt-5.6-terra", 1_050_000, 128_000, 2, 12, 0.2],
 	["gpt-6-astra", 1_050_000, 128_000, 10, 50, 1],
 	["gpt-6-sol", 1_050_000, 128_000, 2, 10, 0.2],
+	["gpt-6.1-sol", 1_050_000, 128_000, 2, 10, 0.1],
 	["grok-4.6", 500_000, 128_000, 1.5, 5, 0.4],
 	["grok-4.7", 500_000, 128_000, 1.5, 5, 0.4],
 	["claude-fable-5", 1_000_000, 128_000, 20, 100, 2],
@@ -133,9 +143,13 @@ function variantsOf(id) {
 		VARIANTS[id] ??
 		(/^gpt-5\.6/u.test(id)
 			? VARIANTS["gpt-5.6-luna"]
-			: /^gpt-6/u.test(id)
+			: /^gpt-6\.1/u.test(id)
+				? VARIANTS["gpt-6.1-sol"]
+				: /^gpt-6/u.test(id)
 				? VARIANTS["gpt-6-astra"]
-				: undefined);
+				: /^deepseek-v4\.1-flash/u.test(id)
+					? VARIANTS["deepseek-v4.1-flash"]
+					: undefined);
 	if (!levels) return undefined;
 	return Object.fromEntries(levels.map((level) => [level, {}]));
 }
@@ -155,13 +169,18 @@ function buildModel(id, base, rate = {}) {
 		input: numberOr(rate.input_usd_per_1m_tokens, 0),
 		output: numberOr(rate.output_usd_per_1m_tokens, 0),
 	};
-	const cacheRead = positiveNumber(rules.cache_read_usd_per_1m_tokens);
+	const cacheRead = cacheRate(id, rate);
 	if (cacheRead !== undefined) cost.cache_read = cacheRead;
 	const cacheWrite = positiveNumber(rules.cache_write_usd_per_1m_tokens);
 	if (cacheWrite !== undefined) cost.cache_write = cacheWrite;
 	return {
 		id,
 		name: id,
+		// Magpie's model parser reads these top-level fields. Keep
+		// capabilities as well for OpenCode readers that use that shape.
+		reasoning: Boolean(variants),
+		attachment: vision,
+		tool_call: true,
 		api: { id, npm, url: apiBase(base) },
 		provider: { npm, api: apiBase(base) },
 		limit: {
@@ -169,7 +188,7 @@ function buildModel(id, base, rate = {}) {
 			output: Math.max(1, numberOr(rate.prepay_completion_token_limit, 128_000)),
 		},
 		capabilities: {
-			reasoning: Boolean(variants) || vision,
+			reasoning: Boolean(variants),
 			toolcall: true,
 			attachment: vision,
 			input: vision ? { text: true, image: true } : { text: true },
@@ -180,6 +199,17 @@ function buildModel(id, base, rate = {}) {
 	};
 }
 
+/**
+ * Claude settlement ignores D.AI's published cache rates and weights cached
+ * tokens against the input rate, matching sdks/pi/src/discover.ts.
+ */
+function cacheRate(id, rate) {
+	const published = positiveNumber(rate.billing_rules?.cache_read_usd_per_1m_tokens);
+	if (!isClaudeModel(id)) return published;
+	const input = numberOr(rate.input_usd_per_1m_tokens, 0);
+	return (input * 2) / 20;
+}
+
 function fallbackCatalog(base) {
 	const models = {};
 	for (const [id, context, output, input, out, cacheRead] of FALLBACK_ROWS) {
@@ -194,14 +224,15 @@ function fallbackCatalog(base) {
 	return models;
 }
 
-// ---- catalog -------------------------------------------------------------------
+// ---- chat catalog --------------------------------------------------------------
 
-async function fetchJson(url, { apiKey, signal } = {}) {
+async function fetchJson(url, { apiKey, signal, headers: extraHeaders } = {}) {
 	const controller = new AbortController();
 	const onAbort = () => controller.abort();
 	const timer = setTimeout(() => controller.abort(), REQUEST_TIMEOUT_MS);
 	signal?.addEventListener("abort", onAbort, { once: true });
-	const headers = new Headers();
+	const headers = new Headers(extraHeaders);
+	headers.set("x-dai-client", CLIENT);
 	if (apiKey) {
 		headers.set("Authorization", `Bearer ${apiKey}`);
 		headers.set("x-api-key", apiKey);
@@ -224,16 +255,8 @@ async function fetchJson(url, { apiKey, signal } = {}) {
  * full catalog in place rather than hiding every model.
  */
 async function fetchCatalog(base, { key, signal } = {}) {
-	const docs = await fetchJson(`${base}/api/system/public/capability-docs`, { signal });
-	const models = {};
-	for (const capability of docs?.data ?? []) {
-		if (!CHAT_CAPABILITIES.has(capability?.capability_code)) continue;
-		for (const model of capability?.models ?? []) {
-			const id = textOf(model?.model_code);
-			if (!id || models[id]) continue;
-			models[id] = buildModel(id, base, model?.token_rate ?? {});
-		}
-	}
+	const docs = await fetchCapabilityDocs(base, signal);
+	const models = chatModelsFromDocs(docs, base);
 	if (!key || !Object.keys(models).length) return models;
 	try {
 		const listed = await fetchJson(`${apiBase(base)}/models`, { apiKey: key, signal });
@@ -242,11 +265,36 @@ async function fetchCatalog(base, { key, signal } = {}) {
 		);
 		if (!allowed.size) return models;
 		return Object.fromEntries(Object.entries(models).filter(([id]) => allowed.has(id)));
-	} catch {
+	} catch (error) {
+		// A refused key must reach Magpie as a sign-in failure. A network or
+		// server failure keeps the last good catalog instead.
+		if (error?.status === 401 || error?.status === 403) {
+			throw Object.assign(new Error("D.AI refused the API key; sign in again"), {
+				status: error.status,
+				signIn: "expired",
+			});
+		}
 		return models;
 	}
 }
 
+async function fetchCapabilityDocs(base, signal) {
+	return fetchJson(`${base}/api/system/public/capability-docs`, { signal });
+}
+
+function chatModelsFromDocs(docs, base) {
+	const models = {};
+	for (const capability of docs?.data ?? []) {
+		if (!CHAT_CAPABILITIES.has(capability?.capability_code)) continue;
+		for (const model of capability?.models ?? []) {
+			const id = textOf(model?.model_code);
+			if (!id || models[id]) continue;
+			models[id] = buildModel(id, base, model?.token_rate ?? {});
+		}
+	}
+	return models;
+}
+
 // ---- the usage card ------------------------------------------------------------
 
 /**
@@ -257,7 +305,11 @@ function balanceCard(account, key) {
 	const balance = account?.balance ?? {};
 	const self = account?.api_key ?? {};
 	const tier = account?.tier ?? {};
-	const parts = [`$${usd(optionalNumber(balance.total) ?? 0)} left`];
+	const total = optionalNumber(balance.total);
+	const pending = optionalNumber(balance.pending_reserved);
+	const calls = optionalNumber(account?.tokens?.today?.calls);
+	const parts = [`$${usd(total ?? 0)} left`];
+	if (pending && pending > 0) parts.push(`$${usd(pending)} pending`);
 	if (textOf(tier.code)) {
 		parts.push(
 			[
@@ -271,6 +323,8 @@ function balanceCard(account, key) {
 				.join(" · "),
 		);
 	}
+	const included = optionalNumber(tier.included_usd_total);
+	if (included && included > 0) parts.push(`$${usd(included)} included`);
 	// Only a key with a budget envelope has a reservation; the shared pool has none.
 	if (self.envelope_active) {
 		const reserved =
@@ -278,12 +332,19 @@ function balanceCard(account, key) {
 			(optionalNumber(self.reserved_recharge_usd) ?? 0);
 		if (reserved > 0) parts.push(`this key $${usd(reserved)}`);
 	}
-	const calls = optionalNumber(account?.tokens?.today?.calls);
+	const requestLimit = optionalNumber(self.daily_request_limit);
+	const usdLimit = optionalNumber(self.daily_usd_limit);
+	const limits = [
+		requestLimit === undefined ? undefined : `${requestLimit} requests/day`,
+		usdLimit === undefined ? undefined : `$${usd(usdLimit)}/day`,
+	].filter(Boolean);
+	if (limits.length) parts.push(`limits ${limits.join(", ")}`);
 	if (calls !== undefined) {
 		parts.push(`${calls} call${calls === 1 ? "" : "s"} today`);
 	}
 	return {
 		plan: textOf(tier.code) || keyLabel(key),
+		balance: total === undefined ? "" : `$${usd(total)}`,
 		windows: [{ name: "D.AI balance", display: parts.join(" · ") }],
 	};
 }
@@ -318,6 +379,154 @@ async function callLogCard(base, key) {
 	};
 }
 
+// ---- media catalog -------------------------------------------------------------
+
+/**
+ * The flat gateway catalog is the source of truth for executable media
+ * models. Unlike capability-docs it carries submit/fetch paths, parameters,
+ * and the public tier prices a caller needs before submitting a job.
+ */
+async function fetchGatewayModels(base, signal) {
+	return fetchJson(`${base}/api/system/public/gateway-models`, { signal });
+}
+
+function mediaTierRates(value) {
+	return (Array.isArray(value) ? value : []).map((rate) => ({
+		tierCode: textOf(rate?.tier_code),
+		tierName: textOf(rate?.tier_name),
+		chargeType: textOf(rate?.charge_type),
+		usdCost: optionalNumber(rate?.usd_cost) ?? 0,
+	}));
+}
+
+function mediaParameters(value) {
+	return (Array.isArray(value) ? value : []).map((parameter) => {
+		const defaultValue = parameter?.default_value;
+		return {
+			key: textOf(parameter?.key),
+			label: textOf(parameter?.label),
+			kind: textOf(parameter?.kind),
+			description: textOf(parameter?.description),
+			required: Boolean(parameter?.required),
+			defaultValue:
+				typeof defaultValue === "string" ||
+				typeof defaultValue === "number" ||
+				typeof defaultValue === "boolean"
+					? defaultValue
+					: null,
+			placeholder: textOf(parameter?.placeholder),
+			min: optionalNumber(parameter?.min) ?? null,
+			max: optionalNumber(parameter?.max) ?? null,
+			step: optionalNumber(parameter?.step) ?? null,
+			options: (Array.isArray(parameter?.options) ? parameter.options : []).map((option) => ({
+				label: textOf(option?.label),
+				value: textOf(option?.value),
+			})),
+		};
+	});
+}
+
+function mediaRows(catalog) {
+	const rows = [];
+	for (const item of Array.isArray(catalog?.data) ? catalog.data : []) {
+		const kind = textOf(item?.capability_type).toLowerCase();
+		const modelCode = textOf(item?.model_code);
+		if (!MEDIA_KINDS.has(kind) || !modelCode) continue;
+		rows.push({
+			kind,
+			capabilityCode: textOf(item?.capability_code),
+			capabilityName: textOf(item?.capability_name),
+			modelId: textOf(item?.model_id),
+			modelCode,
+			serviceCode: textOf(item?.service_code),
+			displayName: textOf(item?.display_name) || modelCode,
+			description: textOf(item?.description),
+			submitPath: textOf(item?.submit_path),
+			fetchPath: textOf(item?.fetch_path),
+			resultKind: textOf(item?.result_kind),
+			promptLabel: textOf(item?.prompt_label),
+			promptPlaceholder: textOf(item?.prompt_placeholder),
+			promptRequired: Boolean(item?.prompt_required),
+			consumptionUnit: textOf(item?.consumption_unit) || "per_request",
+			durationSeconds: optionalNumber(item?.duration_seconds) ?? null,
+			tierRates: mediaTierRates(item?.tier_rates),
+			parameters: mediaParameters(item?.parameters),
+		});
+	}
+	return rows;
+}
+
+function mediaModel(row) {
+	return {
+		kind: row.kind,
+		capabilityCode: row.capabilityCode,
+		capabilityName: row.capabilityName,
+		modelId: row.modelId,
+		modelCode: row.modelCode,
+		serviceCode: row.serviceCode || null,
+		displayName: row.displayName,
+		description: row.description,
+		submitPath: row.submitPath,
+		fetchPath: row.fetchPath || null,
+		resultKind: row.resultKind,
+		promptLabel: row.promptLabel,
+		promptPlaceholder: row.promptPlaceholder,
+		promptRequired: row.promptRequired,
+		consumptionUnit: row.consumptionUnit,
+		durationSeconds: row.durationSeconds,
+		tierRates: row.tierRates,
+		parameters: row.parameters,
+	};
+}
+
+function mediaLimit(input = {}) {
+	return Math.max(
+		1,
+		Math.min(100, Math.floor(optionalNumber(input.limit) ?? MEDIA_CATALOG_LIMIT)),
+	);
+}
+
+function mediaMatches(rows, input = {}) {
+	const kind = textOf(input.kind).toLowerCase();
+	if (kind && !MEDIA_KINDS.has(kind)) {
+		throw new Error("media_catalog kind must be image, video or audio");
+	}
+	const capability = textOf(input.capabilityCode).toLowerCase();
+	const model = textOf(input.model).toLowerCase();
+	const service = textOf(input.serviceCode).toLowerCase();
+	const terms = textOf(input.query).toLowerCase().split(/\s+/u).filter(Boolean);
+	return rows
+		.filter((row) => !kind || row.kind === kind)
+		.filter((row) => !capability || row.capabilityCode.toLowerCase() === capability)
+		.filter((row) => !model || row.modelCode.toLowerCase() === model)
+		.filter((row) => !service || row.serviceCode.toLowerCase() === service)
+		.filter((row) => {
+			if (!terms.length) return true;
+			const haystack = [
+				row.modelCode,
+				row.serviceCode,
+				row.capabilityCode,
+				row.capabilityName,
+				row.displayName,
+				row.description,
+			]
+				.join(" ")
+				.toLowerCase();
+			return terms.every((term) => haystack.includes(term));
+		});
+}
+
+function filterMedia(rows, input = {}) {
+	return mediaMatches(rows, input).slice(0, mediaLimit(input));
+}
+
+async function mediaCatalog(base, input, signal) {
+	const catalog = await fetchGatewayModels(base, signal);
+	const matches = mediaMatches(mediaRows(catalog), input);
+	const models = matches.slice(0, mediaLimit(input)).map(mediaModel);
+	return { models, count: models.length, matched: matches.length };
+}
+
 // ---- plugin --------------------------------------------------------------------
 
 const DaiPlugin = async (_input, options = {}) => {
@@ -364,7 +573,7 @@ const DaiPlugin = async (_input, options = {}) => {
 				return {
 					baseURL: apiBase(base),
 					apiKey: key,
-					headers: { "x-dai-client": "magpie/0.1.0" },
+					headers: { "x-dai-client": CLIENT },
 				};
 			},
 
@@ -412,11 +621,62 @@ const DaiPlugin = async (_input, options = {}) => {
 				try {
 					const models = await catalogFor(key, undefined);
 					return models ?? provider.models;
-				} catch {
+				} catch (error) {
+					if (error?.signIn === "expired") throw error;
 					return provider.models;
 				}
 			},
 		},
+
+		"chat.headers": async (input, output) => {
+			if (input?.model?.providerID !== PROVIDER) return;
+			const session = textOf(input?.sessionID);
+			if (session) output.headers["x-dai-session"] = session;
+		},
+
+		tool: {
+			media_catalog: {
+				description:
+					"List D.AI image, video and audio models published by the public generation gateway. Read-only and free. Returns submit and fetch paths, parameters and tier prices; generation requires a separately authenticated tool.",
+				args: {
+					type: "object",
+					properties: {
+						kind: {
+							type: "string",
+							enum: ["image", "video", "audio"],
+							description: "Optional capability type.",
+						},
+						capabilityCode: {
+							type: "string",
+							description: "Exact capability code, such as openai_image or gemini_video.",
+						},
+						model: {
+							type: "string",
+							description: "Exact model code, such as gpt-image-2.5-flare.",
+						},
+						serviceCode: {
+							type: "string",
+							description: "Exact operation code when one model has several, such as midjourney_imagine.",
+						},
+						query: {
+							type: "string",
+							description: "Case-insensitive search in model, service, capability and description text.",
+						},
+						limit: {
+							type: "integer",
+							minimum: 1,
+							maximum: 100,
+							default: 20,
+							description: "Maximum number of models to return.",
+						},
+					},
+					additionalProperties: false,
+				},
+				async execute(input, ctx) {
+					return mediaCatalog(base, input, ctx?.abort);
+				},
+			},
+		},
 	};
 };
 
@@ -425,4 +685,17 @@ const DaiPlugin = async (_input, options = {}) => {
 export default { id: PROVIDER, server: DaiPlugin };
 
 /** For tests and manual probing: the pure pieces, never called as a plugin. */
-export const _internal = { buildModel, fallbackCatalog, fetchCatalog, keyLabel, normalizeBase };
+export const _internal = {
+	balanceCard,
+	buildModel,
+	fallbackCatalog,
+	fetchCatalog,
+	fetchGatewayModels,
+	filterMedia,
+	keyLabel,
+	mediaCatalog,
+	mediaMatches,
+	mediaModel,
+	mediaRows,
+	normalizeBase,
+};

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@d-ai/magpie",
-  "version": "0.1.1",
-  "description": "D.AI (dai.itbug.shop) models in magpie and OpenCode: one ddsk_ API key, chat models from the public catalog, per-key usage",
+  "version": "0.1.2",
+  "description": "D.AI (dai.itbug.shop) chat models in magpie and OpenCode, plus a read-only image/video/audio model catalog",
   "type": "module",
   "main": "./index.mjs",
   "exports": {
@@ -26,6 +26,7 @@
     "url": "git+https://github.com/mdddj/d-ai-magpie.git"
   },
   "magpie": {
+    "maxConcurrency": 4,
     "icon": "data:image/svg+xml;base64,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"
   },
   "publishConfig": {

package/LICENSE

@@ -1,21 +0,0 @@
-MIT License
-
-Copyright (c) 2026 mddd
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.