@tokagent/tokagentos 2.0.20 → 2.0.22

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tokagent/tokagentos",
-  "version": "2.0.20",
+  "version": "2.0.22",
   "description": "tokagentOS CLI - Create and upgrade tokagentOS project templates",
   "type": "module",
   "bin": {

package/scaffold-patches/packages/agent/src/runtime/core-plugins.ts

@@ -115,8 +115,24 @@ function configureBillingChatMirror(): void {
     .replace(/\/v1$/, "");
   process.env.OPENAI_API_KEY = process.env.BILLING_CHAT_KEY!.trim();
   process.env.OPENAI_BASE_URL = `${base}/v1`;
+
+  // Default model: glm-4.7 is on the Tokamak LiteLLM allowlist. Without
+  // this, plugin-openai falls back to its OpenAI default (gpt-4o-mini /
+  // gpt-4o), which the billing gateway rejects with
+  // "unsupported_model" — and the agent's retry loop turns a fast 400
+  // into a multi-minute "Something went wrong" UX. User can override
+  // with OPENAI_SMALL_MODEL / OPENAI_LARGE_MODEL in .env.
+  const TOKAMAK_DEFAULT_MODEL = "glm-4.7";
+  if (!process.env.OPENAI_SMALL_MODEL?.trim()) {
+    process.env.OPENAI_SMALL_MODEL = TOKAMAK_DEFAULT_MODEL;
+  }
+  if (!process.env.OPENAI_LARGE_MODEL?.trim()) {
+    process.env.OPENAI_LARGE_MODEL = TOKAMAK_DEFAULT_MODEL;
+  }
   console.info(
-    "[tokagent] BILLING_CHAT_KEY + TOKAGENT_GATEWAY_URL detected — wired as OpenAI-compatible provider for chat.",
+    "[tokagent] BILLING_CHAT_KEY + TOKAGENT_GATEWAY_URL detected — wired as OpenAI-compatible provider for chat (model=" +
+      process.env.OPENAI_SMALL_MODEL +
+      ").",
   );
 }
 configureBillingChatMirror();

package/templates/fullstack-app/plugins/plugin-tokagent-billing/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tokagent/plugin-tokagent-billing",
-  "version": "2.0.12",
+  "version": "2.0.13",
   "description": "elizaOS plugin: Web3 credit-billing routes and middleware for the tokagentos LLM gateway.",
   "type": "module",
   "publishConfig": { "access": "public" },

package/templates/fullstack-app/plugins/plugin-tokagent-billing/src/routes/messages-proxy-routes.ts

@@ -261,6 +261,63 @@ async function handleChatCompletions(
   return proxyToLiteLLM(req, res, "/v1/chat/completions");
 }
 
+/**
+ * OpenAI-compatible model catalog. plugin-openai (and many OpenAI SDKs)
+ * call GET /v1/models on startup to validate the API key — if this returns
+ * 401/404, the plugin marks the provider unhealthy and the agent's chat
+ * composer never gets an active backend.
+ *
+ * We return a static list of the models the gateway actually supports
+ * (currently glm-4.7 on Tokamak's LiteLLM). Two reasons static beats
+ * proxying upstream:
+ *   1. Tokamak's LiteLLM /v1/models requires the operator's key, not the
+ *      user's sk-ai-* — proxying would either expose the operator key or
+ *      require a separate auth path. Static avoids the leak.
+ *   2. The billing layer's allowlist is the source of truth for "what
+ *      models a billing client can use"; the upstream catalog is the
+ *      operator's concern. Decoupling them lets us add/remove allowlisted
+ *      models without redeploying the upstream.
+ *
+ * Auth: still gated by applyBillingGate so only authenticated clients see
+ * the list. Returns the same 401 envelope as the chat routes on bad auth.
+ */
+async function handleModels(
+  req: RouteRequest,
+  res: RouteResponse,
+  _runtime: IAgentRuntime,
+): Promise<void> {
+  if (!isBillingStateInitialized()) return billingUnavailable(res);
+  const state = getBillingState();
+  if (!state.config.enabled) return billingUnavailable(res);
+
+  // Auth check — applyBillingGate is overkill here (no model/body to gate
+  // on) but using it keeps the auth-error envelope consistent across routes.
+  const incoming = toIncomingMessage(req);
+  const { resolveBillingIdentity } = await import(
+    "../middleware/api-key-resolve.js"
+  );
+  const identity = await resolveBillingIdentity(incoming);
+  if (!identity) {
+    res.status(401).json({
+      error: { type: "invalid_auth", message: "Authentication required." },
+    });
+    return;
+  }
+
+  const now = Math.floor(Date.now() / 1000);
+  res.status(200).json({
+    object: "list",
+    data: [
+      {
+        id: "glm-4.7",
+        object: "model",
+        created: now,
+        owned_by: "tokamak",
+      },
+    ],
+  });
+}
+
 export const messagesProxyRoutes: Route[] = [
   {
     type: "POST",
@@ -278,6 +335,14 @@ export const messagesProxyRoutes: Route[] = [
     name: "billing-chat-completions-proxy",
     handler: handleChatCompletions,
   },
+  {
+    type: "GET",
+    path: "/v1/models",
+    rawPath: true,
+    public: true,
+    name: "billing-models-catalog",
+    handler: handleModels,
+  },
 ];
 
 export function getMessagesProxyRoutes(mode: "server" | "client"): Route[] {

package/templates-manifest.json

@@ -1,6 +1,6 @@
 {
   "version": "1.0.0",
-  "generatedAt": "2026-05-19T20:06:30.663Z",
+  "generatedAt": "2026-05-19T20:51:16.832Z",
   "repoUrl": "https://github.com/elizaos/eliza",
   "templates": [
     {