@bankr/cli 0.3.13 → 0.3.32

package/dist/commands/llm.js

@@ -5,7 +5,13 @@ import { spawn } from "child_process";
 import { CLI_USER_AGENT, getApiUrl, getLlmKey, getLlmUrl, requireApiKey, } from "../lib/config.js";
 import { getBalances } from "../lib/api.js";
 import * as output from "../lib/output.js";
+import { getErrorMessage } from "../lib/errors.js";
+import { isHelpRequest } from "../lib/passthrough.js";
 const ERR_LLM_NOT_ENABLED = "LLM Gateway not enabled on this API key. Enable at bankr.bot/api-keys";
+/** True when the model serves chat completions (image-only models don't). */
+function supportsChat(m) {
+    return (m.output ?? ["text"]).includes("text");
+}
 /**
  * Fallback model catalog — used when the gateway is unreachable or unauthenticated.
  * resolveModels() fetches live data from GET /v1/models; this list is the offline safety net.
@@ -14,6 +20,15 @@ const IMAGE_INPUT = ["text", "image"];
 const TEXT_INPUT = ["text"];
 const GATEWAY_MODELS = [
     // Claude
+    {
+        id: "claude-fable-5.1",
+        name: "Claude Fable 5.1",
+        owned_by: "anthropic",
+        contextWindow: 1000000,
+        maxTokens: 128000,
+        input: IMAGE_INPUT,
+        cost: { input: 10.0, output: 50.0, cacheRead: 0.25, cacheWrite: 12.5 },
+    },
     {
         id: "claude-fable-5",
         name: "Claude Fable 5",
@@ -23,6 +38,15 @@ const GATEWAY_MODELS = [
         input: IMAGE_INPUT,
         cost: { input: 10.0, output: 50.0, cacheRead: 1.0, cacheWrite: 12.5 },
     },
+    {
+        id: "claude-opus-5",
+        name: "Claude Opus 5",
+        owned_by: "anthropic",
+        contextWindow: 1000000,
+        maxTokens: 128000,
+        input: IMAGE_INPUT,
+        cost: { input: 5.0, output: 25.0, cacheRead: 0.5, cacheWrite: 6.25 },
+    },
     {
         id: "claude-opus-4.8",
         name: "Claude Opus 4.8",
@@ -59,6 +83,15 @@ const GATEWAY_MODELS = [
         input: IMAGE_INPUT,
         cost: { input: 5.0, output: 25.0, cacheRead: 0.5, cacheWrite: 6.25 },
     },
+    {
+        id: "claude-sonnet-5",
+        name: "Claude Sonnet 5",
+        owned_by: "anthropic",
+        contextWindow: 1000000,
+        maxTokens: 128000,
+        input: IMAGE_INPUT,
+        cost: { input: 2.0, output: 10.0, cacheRead: 0.2, cacheWrite: 2.5 },
+    },
     {
         id: "claude-sonnet-4.6",
         name: "Claude Sonnet 4.6",
@@ -88,22 +121,49 @@ const GATEWAY_MODELS = [
     },
     // Gemini
     {
-        id: "gemini-3.1-pro",
-        name: "Gemini 3.1 Pro",
+        id: "gemini-3.8-flash",
+        name: "Gemini 3.8 Flash",
         owned_by: "google",
         contextWindow: 1048576,
         maxTokens: 65536,
         input: IMAGE_INPUT,
-        cost: { input: 2.0, output: 12.0, cacheRead: 0.2, cacheWrite: 0.375 },
+        cost: { input: 0.75, output: 3.75, cacheRead: 0.075, cacheWrite: 0.04167 },
     },
     {
-        id: "gemini-3.1-flash-lite",
-        name: "Gemini 3.1 Flash Lite",
+        id: "gemini-3.7-flash",
+        name: "Gemini 3.7 Flash",
         owned_by: "google",
         contextWindow: 1048576,
         maxTokens: 65536,
         input: IMAGE_INPUT,
-        cost: { input: 0.25, output: 1.5, cacheRead: 0.025, cacheWrite: 0.08333 },
+        cost: { input: 0.75, output: 3.75, cacheRead: 0.075, cacheWrite: 0.04167 },
+    },
+    {
+        id: "gemini-3.6-flash",
+        name: "Gemini 3.6 Flash",
+        owned_by: "google",
+        contextWindow: 1048576,
+        maxTokens: 65536,
+        input: IMAGE_INPUT,
+        cost: { input: 0.75, output: 3.75, cacheRead: 0.075, cacheWrite: 0.04167 },
+    },
+    {
+        id: "gemini-3.5-flash",
+        name: "Gemini 3.5 Flash",
+        owned_by: "google",
+        contextWindow: 1048576,
+        maxTokens: 65536,
+        input: IMAGE_INPUT,
+        cost: { input: 1.5, output: 9.0, cacheRead: 0.15, cacheWrite: 0.08333 },
+    },
+    {
+        id: "gemini-3.1-pro",
+        name: "Gemini 3.1 Pro",
+        owned_by: "google",
+        contextWindow: 1048576,
+        maxTokens: 65536,
+        input: IMAGE_INPUT,
+        cost: { input: 2.0, output: 12.0, cacheRead: 0.2, cacheWrite: 0.375 },
     },
     {
         id: "gemini-3-flash",
@@ -132,7 +192,74 @@ const GATEWAY_MODELS = [
         input: IMAGE_INPUT,
         cost: { input: 0.3, output: 2.5, cacheRead: 0.03, cacheWrite: 0.0833 },
     },
+    // Gemma
+    {
+        id: "gemma-4-31b-it",
+        name: "Gemma 4 31B IT",
+        owned_by: "google",
+        contextWindow: 262144,
+        maxTokens: 131072,
+        input: IMAGE_INPUT,
+        cost: { input: 0.14, output: 0.4, cacheRead: 0, cacheWrite: 0 },
+    },
+    {
+        id: "gemma-4-26b-a4b-it",
+        name: "Gemma 4 26B A4B IT",
+        owned_by: "google",
+        contextWindow: 262144,
+        maxTokens: 262144,
+        input: IMAGE_INPUT,
+        cost: { input: 0.15, output: 0.6, cacheRead: 0, cacheWrite: 0 },
+    },
     // OpenAI
+    {
+        id: "gpt-6-astra",
+        name: "GPT 6 Astra",
+        owned_by: "openai",
+        contextWindow: 1050000,
+        maxTokens: 128000,
+        input: IMAGE_INPUT,
+        cost: { input: 10.0, output: 50.0, cacheRead: 1.0, cacheWrite: 12.5 },
+    },
+    {
+        id: "gpt-5.6-sol",
+        name: "GPT 5.6 Sol",
+        owned_by: "openai",
+        contextWindow: 1050000,
+        maxTokens: 128000,
+        input: IMAGE_INPUT,
+        cost: { input: 4.0, output: 20.0, cacheRead: 0.4, cacheWrite: 5.0 },
+    },
+    {
+        id: "gpt-5.6-terra",
+        name: "GPT 5.6 Terra",
+        owned_by: "openai",
+        contextWindow: 1050000,
+        maxTokens: 128000,
+        input: IMAGE_INPUT,
+        cost: { input: 2.0, output: 12.0, cacheRead: 0.2, cacheWrite: 2.5 },
+    },
+    {
+        id: "gpt-5.6-luna",
+        name: "GPT 5.6 Luna",
+        owned_by: "openai",
+        contextWindow: 1050000,
+        maxTokens: 128000,
+        input: IMAGE_INPUT,
+        cost: { input: 0.2, output: 1.2, cacheRead: 0.02, cacheWrite: 0.25 },
+    },
+    {
+        id: "gpt-image-2",
+        name: "GPT Image 2",
+        owned_by: "openai",
+        contextWindow: 32000,
+        maxTokens: 16384,
+        input: TEXT_INPUT,
+        output: ["image"],
+        // Image model — served via /v1/images/generations; output rate is per
+        // million image tokens.
+        cost: { input: 5.0, output: 30.0, cacheRead: 0, cacheWrite: 0 },
+    },
     {
         id: "gpt-5.5",
         name: "GPT 5.5",
@@ -206,6 +333,33 @@ const GATEWAY_MODELS = [
         cost: { input: 0.05, output: 0.4, cacheRead: 0.005, cacheWrite: 0 },
     },
     // Other providers
+    {
+        id: "grok-4.6",
+        name: "Grok 4.6",
+        owned_by: "x-ai",
+        contextWindow: 500000,
+        maxTokens: 500000,
+        input: IMAGE_INPUT,
+        cost: { input: 2.0, output: 6.0, cacheRead: 0.5, cacheWrite: 0 },
+    },
+    {
+        id: "grok-4.5",
+        name: "Grok 4.5",
+        owned_by: "x-ai",
+        contextWindow: 500000,
+        maxTokens: 500000,
+        input: IMAGE_INPUT,
+        cost: { input: 2.0, output: 6.0, cacheRead: 0.3, cacheWrite: 0 },
+    },
+    {
+        id: "grok-4.20",
+        name: "Grok 4.20",
+        owned_by: "x-ai",
+        contextWindow: 2000000,
+        maxTokens: 2000000,
+        input: IMAGE_INPUT,
+        cost: { input: 1.25, output: 2.5, cacheRead: 0.2, cacheWrite: 0 },
+    },
     {
         id: "grok-4.3",
         name: "Grok 4.3",
@@ -223,6 +377,39 @@ const GATEWAY_MODELS = [
         maxTokens: 30000,
         input: TEXT_INPUT,
         cost: { input: 0.2, output: 0.5, cacheRead: 0.05, cacheWrite: 0 },
+        // Hard-deprecated in the seed — the gateway 410s it. Without this the
+        // offline catalog would advertise it as active.
+        deprecation: {
+            replaced_by: "grok-4.3",
+            removed_at: "2026-06-17T00:00:00.000Z",
+        },
+    },
+    {
+        id: "deepseek-v4-pro-0813",
+        name: "DeepSeek V4 Pro 0813",
+        owned_by: "deepseek",
+        contextWindow: 1048576,
+        maxTokens: 384000,
+        input: TEXT_INPUT,
+        cost: { input: 1.32, output: 3.96, cacheRead: 0.044, cacheWrite: 0 },
+    },
+    {
+        id: "deepseek-v4-pro",
+        name: "DeepSeek V4 Pro",
+        owned_by: "deepseek",
+        contextWindow: 1048576,
+        maxTokens: 384000,
+        input: TEXT_INPUT,
+        cost: { input: 0.435, output: 0.87, cacheRead: 0.003625, cacheWrite: 0 },
+    },
+    {
+        id: "deepseek-v4-flash",
+        name: "DeepSeek V4 Flash",
+        owned_by: "deepseek",
+        contextWindow: 1048576,
+        maxTokens: 384000,
+        input: TEXT_INPUT,
+        cost: { input: 0.14, output: 0.28, cacheRead: 0.0028, cacheWrite: 0 },
     },
     {
         id: "deepseek-v3.2",
@@ -233,6 +420,42 @@ const GATEWAY_MODELS = [
         input: TEXT_INPUT,
         cost: { input: 0.26, output: 0.38, cacheRead: 0.13, cacheWrite: 0 },
     },
+    {
+        id: "qwen3.8-max",
+        name: "Qwen3.8 Max",
+        owned_by: "qwen",
+        contextWindow: 1000000,
+        maxTokens: 131072,
+        input: IMAGE_INPUT,
+        cost: { input: 2.0, output: 6.0, cacheRead: 0.25, cacheWrite: 2.5 },
+    },
+    {
+        id: "qwen3.8-flash",
+        name: "Qwen3.8 Flash",
+        owned_by: "qwen",
+        contextWindow: 1000000,
+        maxTokens: 131072,
+        input: IMAGE_INPUT,
+        cost: { input: 0.15, output: 0.47, cacheRead: 0.016, cacheWrite: 0.2 },
+    },
+    {
+        id: "qwen3.7-max",
+        name: "Qwen3.7 Max",
+        owned_by: "qwen",
+        contextWindow: 1000000,
+        maxTokens: 65536,
+        input: TEXT_INPUT,
+        cost: { input: 2.5, output: 7.5, cacheRead: 0.25, cacheWrite: 0 },
+    },
+    {
+        id: "qwen3.7-flash",
+        name: "Qwen3.7 Flash",
+        owned_by: "qwen",
+        contextWindow: 1000000,
+        maxTokens: 65536,
+        input: IMAGE_INPUT,
+        cost: { input: 0.1, output: 0.4, cacheRead: 0.02, cacheWrite: 0.125 },
+    },
     {
         id: "qwen3-coder",
         name: "Qwen3 Coder",
@@ -240,7 +463,7 @@ const GATEWAY_MODELS = [
         contextWindow: 262144,
         maxTokens: 65536,
         input: TEXT_INPUT,
-        cost: { input: 0.12, output: 0.75, cacheRead: 0.06, cacheWrite: 0 },
+        cost: { input: 0.3, output: 1.5, cacheRead: 0.06, cacheWrite: 0 },
     },
     {
         id: "qwen3.7-plus",
@@ -258,7 +481,7 @@ const GATEWAY_MODELS = [
         contextWindow: 1000000,
         maxTokens: 65536,
         input: TEXT_INPUT,
-        cost: { input: 0.25, output: 1.5, cacheRead: 0, cacheWrite: 0 },
+        cost: { input: 0.1875, output: 1.125, cacheRead: 0, cacheWrite: 0.234375 },
     },
     {
         id: "qwen3.5-plus",
@@ -276,7 +499,16 @@ const GATEWAY_MODELS = [
         contextWindow: 1000000,
         maxTokens: 65536,
         input: TEXT_INPUT,
-        cost: { input: 0.1, output: 0.4, cacheRead: 0, cacheWrite: 0 },
+        cost: { input: 0.065, output: 0.26, cacheRead: 0, cacheWrite: 0 },
+    },
+    {
+        id: "kimi-k3",
+        name: "Kimi K3",
+        owned_by: "moonshotai",
+        contextWindow: 1048576,
+        maxTokens: 131072,
+        input: IMAGE_INPUT,
+        cost: { input: 3.0, output: 15.0, cacheRead: 0.3, cacheWrite: 0 },
     },
     {
         id: "kimi-k2.7-code",
@@ -303,7 +535,16 @@ const GATEWAY_MODELS = [
         contextWindow: 262144,
         maxTokens: 65535,
         input: TEXT_INPUT,
-        cost: { input: 0.45, output: 2.2, cacheRead: 0.225, cacheWrite: 0 },
+        cost: { input: 0.6, output: 3.0, cacheRead: 0.1, cacheWrite: 0 },
+    },
+    {
+        id: "minimax-m3",
+        name: "MiniMax M3",
+        owned_by: "minimax",
+        contextWindow: 524288,
+        maxTokens: 524288,
+        input: IMAGE_INPUT,
+        cost: { input: 0.3, output: 1.2, cacheRead: 0.06, cacheWrite: 0 },
     },
     {
         id: "minimax-m2.5",
@@ -332,6 +573,42 @@ const GATEWAY_MODELS = [
         input: TEXT_INPUT,
         cost: { input: 0.6, output: 2.4, cacheRead: 0.06, cacheWrite: 0 },
     },
+    {
+        id: "glm-5.3-flash",
+        name: "GLM-5.3 Flash",
+        owned_by: "z-ai",
+        contextWindow: 1048576,
+        maxTokens: 131072,
+        input: IMAGE_INPUT,
+        cost: { input: 0.15, output: 0.5, cacheRead: 0.03, cacheWrite: 0 },
+    },
+    {
+        id: "glm-5.3",
+        name: "GLM-5.3",
+        owned_by: "z-ai",
+        contextWindow: 1048576,
+        maxTokens: 131072,
+        input: TEXT_INPUT,
+        cost: { input: 1.4, output: 4.4, cacheRead: 0.26, cacheWrite: 0 },
+    },
+    {
+        id: "glm-5.2",
+        name: "GLM-5.2",
+        owned_by: "z-ai",
+        contextWindow: 1048576,
+        maxTokens: 131072,
+        input: TEXT_INPUT,
+        cost: { input: 1.4, output: 4.4, cacheRead: 0.26, cacheWrite: 0 },
+    },
+    {
+        id: "glm-5.1",
+        name: "GLM-5.1",
+        owned_by: "z-ai",
+        contextWindow: 202800,
+        maxTokens: 202800,
+        input: TEXT_INPUT,
+        cost: { input: 1.4, output: 4.4, cacheRead: 0.26, cacheWrite: 0 },
+    },
     {
         id: "glm-5",
         name: "GLM-5",
@@ -339,7 +616,7 @@ const GATEWAY_MODELS = [
         contextWindow: 202752,
         maxTokens: 131072,
         input: TEXT_INPUT,
-        cost: { input: 0.72, output: 2.3, cacheRead: 0, cacheWrite: 0 },
+        cost: { input: 1.0, output: 3.2, cacheRead: 0.2, cacheWrite: 0 },
     },
     {
         id: "glm-5-turbo",
@@ -373,6 +650,9 @@ async function resolveModels() {
                 contextWindow: m.context_window ?? fallback?.contextWindow ?? 128000,
                 maxTokens: m.max_output_tokens ?? fallback?.maxTokens ?? 32768,
                 input: m.input_modalities ?? fallback?.input ?? TEXT_INPUT,
+                ...((m.output_modalities ?? fallback?.output) && {
+                    output: m.output_modalities ?? fallback?.output,
+                }),
                 cost: m.pricing
                     ? {
                         input: m.pricing.input,
@@ -388,6 +668,7 @@ async function resolveModels() {
                     }),
                 ...(m.deprecation && { deprecation: m.deprecation }),
                 ...(m.private && { private: true }),
+                ...(m.zdr && { zdr: true }),
             };
         });
         return { models, live: true };
@@ -426,53 +707,148 @@ function installConfigFile(configFile, toolName, merge) {
     output.success(`Bankr provider installed to ${configFile}`);
 }
 /* ───────────────────────────── bankr llm models ───────────────────────────── */
+/**
+ * The tier filters. Both flags read the gateway's own capability flags from
+ * /v1/models, so they track the live registry rather than a hardcoded list.
+ */
+const TIER_VIEWS = {
+    private: {
+        title: "Bankr LLM Gateway — Private Models",
+        match: (m) => m.private === true,
+        empty: "No models currently offer a private (TEE) compute environment.",
+        countSuffix: " support :private",
+    },
+    zdr: {
+        title: "Bankr LLM Gateway — Zero-Retention Models",
+        match: (m) => m.zdr === true,
+        empty: "No models currently guarantee zero data retention.",
+        countSuffix: " support :zdr",
+    },
+};
 export async function modelsCommand(opts = {}) {
     const spin = output.spinner("Fetching models…");
     const { models, live } = await resolveModels();
     spin.stop();
-    // `private` is the gateway's own "supports :private" flag from
-    // /v1/models (driven by each model's NEAR slot + NEARAI_API_KEY), so this
-    // tracks the live registry rather than a hardcoded CLI list.
-    const shown = opts.private ? models.filter((m) => m.private) : models;
-    output.brandBold(opts.private
-        ? "Bankr LLM Gateway — Private Models"
-        : "Bankr LLM Gateway — Available Models");
+    // `--private` wins when both are passed: private is the narrower set (every
+    // private model is also ZDR), so it's the more specific of the two requests.
+    const view = opts.private
+        ? TIER_VIEWS.private
+        : opts.zdr
+            ? TIER_VIEWS.zdr
+            : undefined;
+    const shown = view ? models.filter(view.match) : models;
+    output.brandBold(view?.title ?? "Bankr LLM Gateway — Available Models");
     console.log();
-    if (opts.private && shown.length === 0) {
+    if (view && shown.length === 0) {
         output.warn(live
-            ? "No models currently offer a private (TEE) compute environment."
-            : "Private status needs a live gateway — set BANKR_LLM_KEY (and BANKR_LLM_URL).");
+            ? view.empty
+            : "Tier status needs a live gateway — set BANKR_LLM_KEY (and BANKR_LLM_URL).");
         return;
     }
-    const COL = { id: 24, name: 24, provider: 12 };
-    const hasNotes = shown.some((m) => m.deprecation || m.private);
-    console.log(`  ${output.fmt.brandBold("Model ID".padEnd(COL.id))} ${output.fmt.brandBold("Name".padEnd(COL.name))} ${output.fmt.brandBold("Provider".padEnd(COL.provider))}${hasNotes ? ` ${output.fmt.brandBold("Notes")}` : ""}`);
-    console.log(output.fmt.dim(`  ${"─".repeat(COL.id + COL.name + COL.provider + (hasNotes ? 24 : 2))}`));
-    for (const m of shown) {
-        let notes = "";
+    // Widths are sized to the longest real value (`minimax-m2.7-highspeed` /
+    // `MiniMax M2.7 Highspeed` are both 22), keeping the table under ~90 columns
+    // once the tier list is included.
+    const COL = { id: 23, name: 23, provider: 11, privacy: 28 };
+    const isImageGen = (m) => m.output?.includes("image") === true;
+    /**
+     * Every tier this model can serve, not just the strongest.
+     *
+     * The tiers nest, so a private model is also reachable at zdr and standard —
+     * listing only the top one made the ladder look like three separate offers
+     * and hid that `glm-5.2` (no suffix) is a perfectly valid standard request.
+     * Read straight off the flags rather than inferring `zdr` from `private`, so
+     * the column reports what the gateway actually advertises.
+     */
+    const supportedTiers = (m) => {
+        // Each tier is painted individually. Colouring the joined string instead
+        // made "standard" grey on a standard-only row and green on a ZDR one — the
+        // same word changing colour purely because of what followed it.
+        const tiers = [
+            // The baseline, on every row, so it recedes rather than competing.
+            { label: "standard", paint: output.fmt.dim },
+        ];
+        if (m.zdr)
+            tiers.push({ label: "zdr", paint: output.fmt.success });
+        if (m.private) {
+            // Distinct from zdr: they are two rungs, not one capability.
+            tiers.push({ label: "private (tee)", paint: output.fmt.magic });
+        }
+        return tiers;
+    };
+    /**
+     * Pad from the PLAIN text — ANSI escapes have zero display width but count
+     * toward `.length`, so padding the painted string misaligns every coloured
+     * row against the header.
+     */
+    const tierCell = (m) => {
+        const tiers = supportedTiers(m);
+        const plain = tiers.map((t) => t.label).join(", ");
+        const painted = tiers
+            .map((t) => t.paint(t.label))
+            .join(output.fmt.dim(", "));
+        return painted + " ".repeat(Math.max(0, COL.privacy - plain.length));
+    };
+    /**
+     * Everything that is NOT a privacy tier. Kept in its own column because
+     * sharing one cell forced an `else if` chain where a deprecated ZDR model
+     * showed only "deprecated" and its tier silently disappeared.
+     */
+    const modelNotes = (m) => {
+        const parts = [];
         if (m.deprecation) {
-            notes = m.deprecation.replaced_by
+            parts.push(m.deprecation.replaced_by
                 ? `deprecated → ${m.deprecation.replaced_by}`
-                : "deprecated";
+                : "deprecated");
         }
-        else if (m.private) {
-            notes = "Private (TEE)";
-        }
-        console.log(`  ${output.fmt.brand(m.id.padEnd(COL.id))} ${m.name.padEnd(COL.name)} ${output.fmt.dim(m.owned_by.padEnd(COL.provider))} ${notes ? output.fmt.dim(notes) : ""}`);
+        if (isImageGen(m))
+            parts.push("Image gen (/v1/images/generations)");
+        return parts.join(" · ");
+    };
+    const showNotes = shown.some((m) => modelNotes(m) !== "");
+    const line = (cells) => console.log(`  ${cells.join(" ")}`.trimEnd());
+    line([
+        output.fmt.brandBold("Model ID".padEnd(COL.id)),
+        output.fmt.brandBold("Name".padEnd(COL.name)),
+        output.fmt.brandBold("Provider".padEnd(COL.provider)),
+        output.fmt.brandBold("Privacy Tier".padEnd(COL.privacy)),
+        ...(showNotes ? [output.fmt.brandBold("Notes")] : []),
+    ]);
+    // Measured, not guessed: a fixed fudge factor for the last column left the
+    // rule hanging well past the header once the tier list widened it.
+    const notesWidth = showNotes
+        ? Math.max(...shown.map((m) => modelNotes(m).length))
+        : 0;
+    const rule = COL.id +
+        COL.name +
+        COL.provider +
+        COL.privacy +
+        3 +
+        (showNotes ? notesWidth + 1 : 0);
+    console.log(output.fmt.dim(`  ${"─".repeat(rule)}`));
+    for (const m of shown) {
+        line([
+            output.fmt.brand(m.id.padEnd(COL.id)),
+            m.name.padEnd(COL.name),
+            output.fmt.dim(m.owned_by.padEnd(COL.provider)),
+            tierCell(m),
+            ...(showNotes ? [output.fmt.dim(modelNotes(m))] : []),
+        ]);
     }
     console.log();
     output.dim(`  Gateway: ${getLlmUrl()}`);
-    output.dim(`  ${shown.length} model${shown.length === 1 ? "" : "s"}${opts.private ? " support :private" : " available"}${live ? "" : " (cached)"}`);
-    if (!opts.private && shown.some((m) => m.private)) {
+    output.dim(`  ${shown.length} model${shown.length === 1 ? "" : "s"}${view?.countSuffix ?? " available"}${live ? "" : " (cached)"}`);
+    if (!view && shown.some((m) => m.private)) {
         output.dim("  Private (TEE) — append :private to the model id (e.g. glm-5.2:private). List: bankr llm models --private");
     }
+    if (!view && shown.some((m) => m.zdr)) {
+        output.dim("  Zero data retention — append :zdr to the model id (e.g. glm-5.2:zdr), or turn it on account-wide under Settings in the web terminal. List: bankr llm models --zdr");
+    }
 }
 /* ─────────────────────── bankr llm credits ─────────────────────────────────── */
 export async function creditsCommand() {
     const llmKey = getLlmKey();
     if (!llmKey) {
-        output.error("Not authenticated. Run `bankr login` first.");
-        process.exit(1);
+        output.fatal("Not authenticated. Run `bankr login` first.");
     }
     const llmUrl = getLlmUrl();
     const spin = output.spinner("Fetching credit balance…");
@@ -491,12 +867,10 @@ export async function creditsCommand() {
             return;
         }
         if (res.status === 401 || res.status === 403) {
-            output.error("Authentication failed. Check your API key or run: bankr login");
-            process.exit(1);
+            output.fatal("Authentication failed. Check your API key or run: bankr login");
         }
         if (!res.ok) {
-            output.error(`Failed to fetch credits (HTTP ${res.status})`);
-            process.exit(1);
+            output.fatal(`Failed to fetch credits (HTTP ${res.status})`);
         }
         const body = (await res.json());
         output.brandBold("Bankr LLM Gateway — Credits");
@@ -507,8 +881,7 @@ export async function creditsCommand() {
     }
     catch (err) {
         spin.stop();
-        output.error(`Failed to fetch credits: ${err.message}`);
-        process.exit(1);
+        output.fatal(`Failed to fetch credits: ${getErrorMessage(err)}`);
     }
 }
 /* ─────────────────────── bankr llm credits add ──────────────────────────── */
@@ -560,7 +933,7 @@ function resolveTokenAcrossChains(input, balances) {
                     symbol: nativeSym,
                     name: nativeSym,
                     decimals: 18,
-                    balance: parseFloat(chainBal.nativeBalance || "0"),
+                    balance: chainBal.nativeBalance || "0",
                     balanceUsd: nativeUsd,
                 });
             }
@@ -603,8 +976,7 @@ export async function creditsAddCommand(amount, opts) {
     // Validate amount
     const amountUsd = Number(amount);
     if (Number.isNaN(amountUsd) || amountUsd < 1 || amountUsd > 1000) {
-        output.error("Amount must be between $1 and $1,000");
-        process.exit(1);
+        output.fatal("Amount must be between $1 and $1,000");
     }
     // Fetch current balance for display (skip when --yes/--not-interactive since user won't see it)
     let currentBalance = "unknown";
@@ -627,7 +999,7 @@ export async function creditsAddCommand(amount, opts) {
         catch (err) {
             balSpin.stop();
             // Non-fatal — proceed without balance display
-            output.dim(`  (Could not fetch balance: ${err.message})`);
+            output.dim(`  (Could not fetch balance: ${getErrorMessage(err)})`);
         }
     }
     // When --token is passed, the chain holding the highest USD balance of
@@ -642,11 +1014,10 @@ export async function creditsAddCommand(amount, opts) {
         }
         catch (err) {
             resolveSpin.stop();
-            output.warn(`Could not fetch balances: ${err.message}`);
+            output.warn(`Could not fetch balances: ${getErrorMessage(err)}`);
         }
         if (!resolved) {
-            output.error(`Token "${opts.token}" not found in your wallet on any supported chain.`);
-            process.exit(1);
+            output.fatal(`Token "${opts.token}" not found in your wallet on any supported chain.`);
         }
     }
     const targetChain = resolved?.chain ?? "base";
@@ -693,22 +1064,18 @@ export async function creditsAddCommand(amount, opts) {
         spin.stop();
         if (res.status === 402) {
             const errBody = (await res.json().catch(() => ({})));
-            output.error(errBody.error ??
+            output.fatal(errBody.error ??
                 "Insufficient wallet balance. Add funds to your wallet first.");
-            process.exit(1);
         }
         if (res.status === 403) {
-            output.error(ERR_LLM_NOT_ENABLED);
-            process.exit(1);
+            output.fatal(ERR_LLM_NOT_ENABLED);
         }
         if (res.status === 502) {
-            output.error("Token swap failed. Try USDC or a different token.");
-            process.exit(1);
+            output.fatal("Token swap failed. Try USDC or a different token.");
         }
         if (!res.ok) {
             const errBody = (await res.json().catch(() => ({})));
-            output.error(errBody.error ?? `Failed to add credits (HTTP ${res.status})`);
-            process.exit(1);
+            output.fatal(errBody.error ?? `Failed to add credits (HTTP ${res.status})`);
         }
         const result = (await res.json());
         console.log();
@@ -720,8 +1087,7 @@ export async function creditsAddCommand(amount, opts) {
     }
     catch (err) {
         spin.stop();
-        output.error(`Failed to add credits: ${err.message}`);
-        process.exit(1);
+        output.fatal(`Failed to add credits: ${getErrorMessage(err)}`);
     }
 }
 function formatTimeAgo(dateStr) {
@@ -762,12 +1128,10 @@ export async function creditsAutoCommand(opts) {
             });
             spin.stop();
             if (res.status === 403) {
-                output.error(ERR_LLM_NOT_ENABLED);
-                process.exit(1);
+                output.fatal(ERR_LLM_NOT_ENABLED);
             }
             if (!res.ok) {
-                output.error(`Failed to fetch config (HTTP ${res.status})`);
-                process.exit(1);
+                output.fatal(`Failed to fetch config (HTTP ${res.status})`);
             }
             const { config } = (await res.json());
             output.brandBold("Bankr LLM Gateway — Auto Top-Up");
@@ -782,15 +1146,13 @@ export async function creditsAutoCommand(opts) {
         }
         catch (err) {
             spin.stop();
-            output.error(`Failed to fetch config: ${err.message}`);
-            process.exit(1);
+            output.fatal(`Failed to fetch config: ${getErrorMessage(err)}`);
         }
         return;
     }
     // POST update config
     if (opts.enable && opts.disable) {
-        output.error("Cannot use --enable and --disable together");
-        process.exit(1);
+        output.fatal("Cannot use --enable and --disable together");
     }
     const body = {};
     if (opts.disable) {
@@ -806,16 +1168,14 @@ export async function creditsAutoCommand(opts) {
     if (opts.amount) {
         const amt = Number(opts.amount);
         if (Number.isNaN(amt) || amt < 1 || amt > 1000) {
-            output.error("Amount must be between $1 and $1,000");
-            process.exit(1);
+            output.fatal("Amount must be between $1 and $1,000");
         }
         body.amountUsd = amt;
     }
     if (opts.threshold) {
         const thr = Number(opts.threshold);
         if (Number.isNaN(thr) || thr < 1 || thr > 500) {
-            output.error("Threshold must be between $1 and $500");
-            process.exit(1);
+            output.fatal("Threshold must be between $1 and $500");
         }
         body.thresholdUsd = thr;
     }
@@ -827,8 +1187,7 @@ export async function creditsAutoCommand(opts) {
             .map((s) => s.trim())
             .filter(Boolean);
         if (inputs.length === 0 || inputs.length > 3) {
-            output.error("Provide 1-3 comma-separated token symbols or addresses");
-            process.exit(1);
+            output.fatal("Provide 1-3 comma-separated token symbols or addresses");
         }
         const resolved = [];
         // Highest USD balance wins when the same symbol exists on multiple chains.
@@ -859,8 +1218,7 @@ export async function creditsAutoCommand(opts) {
             }
         }
         if (resolved.length === 0) {
-            output.error("No valid tokens resolved. Check your token symbols.");
-            process.exit(1);
+            output.fatal("No valid tokens resolved. Check your token symbols.");
         }
         body.tokens = resolved;
     }
@@ -883,13 +1241,11 @@ export async function creditsAutoCommand(opts) {
         });
         spin.stop();
         if (res.status === 403) {
-            output.error(ERR_LLM_NOT_ENABLED);
-            process.exit(1);
+            output.fatal(ERR_LLM_NOT_ENABLED);
         }
         if (!res.ok) {
             const errBody = (await res.json().catch(() => ({})));
-            output.error(errBody.error ?? `Failed to update config (HTTP ${res.status})`);
-            process.exit(1);
+            output.fatal(errBody.error ?? `Failed to update config (HTTP ${res.status})`);
         }
         output.success(opts.disable ? "Auto top-up disabled" : "Auto top-up settings updated");
         // Show config from POST response (avoids extra GET round-trip)
@@ -907,8 +1263,7 @@ export async function creditsAutoCommand(opts) {
     }
     catch (err) {
         spin.stop();
-        output.error(`Failed to update config: ${err.message}`);
-        process.exit(1);
+        output.fatal(`Failed to update config: ${getErrorMessage(err)}`);
     }
 }
 /* ──────────────────────── bankr llm setup openclaw ────────────────────────── */
@@ -916,13 +1271,31 @@ export async function setupOpenclawCommand(opts) {
     const llmKey = getLlmKey();
     const llmUrl = getLlmUrl();
     const { models } = await resolveModels();
-    const activeModels = models.filter((m) => !m.deprecation);
+    // Coding-tool config lists chat models only — image-only models can't chat.
+    const activeModels = models.filter((m) => !m.deprecation && supportsChat(m));
+    // --images routes OpenClaw's built-in image_generate tool through the gateway.
+    // That tool only uses OpenClaw's known image providers (not custom ones like
+    // `bankr`), so enabling it takes over the `openai` provider slot. OpenClaw's
+    // OpenAI provider expects a `/v1`-suffixed base; normalize so a llmUrl that
+    // already ends in /v1 (or a trailing slash) doesn't produce `/v1/v1`.
+    const imageBaseUrl = `${llmUrl.replace(/\/+$/, "").replace(/\/v1$/, "")}/v1`;
+    const imageProvider = opts.images
+        ? { baseUrl: imageBaseUrl, apiKey: llmKey ?? "${BANKR_API_KEY}" }
+        : null;
+    const imageGenerationModel = opts.images
+        ? { primary: "openai/gpt-image-2", timeoutMs: 180000 }
+        : null;
+    const warnImagesSlot = () => {
+        if (opts.images)
+            output.warn("--images points OpenClaw's `openai` provider at the gateway; image_generate (and any OpenAI-format request) now routes through Bankr. Skip --images if you use real OpenAI in OpenClaw.");
+    };
     const providerConfig = {
         baseUrl: llmUrl,
         apiKey: llmKey ?? "${BANKR_API_KEY}",
         api: "openai-completions",
-        // Private models get both entries: the plain id and a `:private`
-        // twin that routes through the TEE compute environment — so users can pick either.
+        // Models get a twin entry per privacy tier they support, so users can pick
+        // one from the model dropdown in tools that expose no other configuration.
+        // Base first, then weakest tier to strongest.
         models: activeModels.flatMap((m) => {
             const base = {
                 id: m.id,
@@ -933,12 +1306,15 @@ export async function setupOpenclawCommand(opts) {
                 maxTokens: m.maxTokens,
                 cost: m.cost,
             };
-            return m.private
-                ? [
-                    base,
-                    { ...base, id: `${m.id}:private`, name: `${m.name} (Private)` },
-                ]
-                : [base];
+            return [
+                base,
+                ...(m.zdr
+                    ? [{ ...base, id: `${m.id}:zdr`, name: `${m.name} (ZDR)` }]
+                    : []),
+                ...(m.private
+                    ? [{ ...base, id: `${m.id}:private`, name: `${m.name} (Private)` }]
+                    : []),
+            ];
         }),
     };
     if (opts.install) {
@@ -947,17 +1323,43 @@ export async function setupOpenclawCommand(opts) {
             const models = existing.models ?? {};
             const providers = models.providers ?? {};
             providers.bankr = providerConfig;
+            if (imageProvider) {
+                providers.openai = {
+                    ...(providers.openai ?? {}),
+                    ...imageProvider,
+                };
+            }
             models.providers = providers;
             existing.models = models;
+            if (imageGenerationModel) {
+                const agents = existing.agents ?? {};
+                const defaults = agents.defaults ?? {};
+                defaults.imageGenerationModel = imageGenerationModel;
+                agents.defaults = defaults;
+                existing.agents = agents;
+            }
         });
         warnIfNoKey(llmKey);
+        warnImagesSlot();
         return;
     }
-    console.log(JSON.stringify({ models: { providers: { bankr: providerConfig } } }, null, 2));
+    const printed = {
+        models: {
+            providers: {
+                bankr: providerConfig,
+                ...(imageProvider ? { openai: imageProvider } : {}),
+            },
+        },
+        ...(imageGenerationModel
+            ? { agents: { defaults: { imageGenerationModel } } }
+            : {}),
+    };
+    console.log(JSON.stringify(printed, null, 2));
     console.log();
     if (!llmKey) {
         output.warn("No API key set. Replace ${BANKR_API_KEY} with your key or run: bankr login");
     }
+    warnImagesSlot();
     output.dim("Add to ~/.openclaw/openclaw.json or run: bankr llm setup openclaw --install");
 }
 /* ──────────────────────── bankr llm setup opencode ─────────────────────────── */
@@ -965,11 +1367,15 @@ export async function setupOpenCodeCommand(opts) {
     const llmKey = getLlmKey();
     const llmUrl = getLlmUrl();
     const { models } = await resolveModels();
-    const activeModels = models.filter((m) => !m.deprecation);
+    // Coding-tool config lists chat models only — image-only models can't chat.
+    const activeModels = models.filter((m) => !m.deprecation && supportsChat(m));
     const modelsObj = {};
     for (const m of activeModels) {
         modelsObj[m.id] = { name: m.name };
-        // Private models also expose a `:private` variant (TEE compute environment).
+        // One variant per privacy tier the model supports, weakest to strongest.
+        if (m.zdr) {
+            modelsObj[`${m.id}:zdr`] = { name: `${m.name} (ZDR)` };
+        }
         if (m.private) {
             modelsObj[`${m.id}:private`] = { name: `${m.name} (Private)` };
         }
@@ -1015,7 +1421,7 @@ export async function setupCursorCommand() {
     const { models } = await resolveModels();
     const activeModels = models.filter((m) => !m.deprecation);
     // Pick one model per provider as recommended examples
-    const recommendedIds = ["claude-fable-5", "gemini-3.1-pro", "gpt-5.5"];
+    const recommendedIds = ["claude-fable-5.1", "gemini-3.1-pro", "gpt-5.5"];
     const recommended = recommendedIds.filter((id) => activeModels.some((m) => m.id === id));
     output.brandBold("Cursor — Bankr LLM Gateway");
     console.log();
@@ -1061,8 +1467,7 @@ export async function setupClaudeCodeCommand() {
 function requireAuth() {
     const llmKey = getLlmKey();
     if (!llmKey) {
-        output.error("Not authenticated. Run `bankr login` first.");
-        process.exit(1);
+        output.fatal("Not authenticated. Run `bankr login` first.");
     }
     return llmKey;
 }
@@ -1102,8 +1507,8 @@ function fileContains(filePath, search) {
  * `--model` flag appears honored but the real request is a different
  * model. Any suffix (e.g. `[1m]` context tier) is preserved.
  */
-function toAnthropicModelId(model) {
-    return model.replace(/^(claude-(?:opus|sonnet|haiku)-)(\d+)\.(\d+)(.*)$/, "$1$2-$3$4");
+export function toAnthropicModelId(model) {
+    return model.replace(/^(claude-(?:opus|sonnet|haiku|fable)-)(\d+)\.(\d+)(.*)$/, "$1$2-$3$4");
 }
 function translateClaudeModelArgs(args) {
     const out = [...args];
@@ -1122,6 +1527,9 @@ function translateClaudeModelArgs(args) {
     return out;
 }
 export async function claudeCommand(args) {
+    if (isHelpRequest(args)) {
+        return launchTool("claude", args, {}, "https://docs.anthropic.com/en/docs/claude-code");
+    }
     const llmKey = requireAuth();
     const llmUrl = getLlmUrl();
     const translated = translateClaudeModelArgs(args);
@@ -1130,6 +1538,9 @@ export async function claudeCommand(args) {
 }
 /* ────────────────────────── bankr llm opencode ───────────────────────────── */
 export async function opencodeCommand(args) {
+    if (isHelpRequest(args)) {
+        return launchTool("opencode", args, {}, "https://opencode.ai");
+    }
     const llmKey = requireAuth();
     const llmUrl = getLlmUrl();
     const configFile = join(homedir(), ".config", "opencode", "opencode.json");