@magpie-community/opencode-zen-free-auth 0.1.15 → 0.1.17

package/README.md

@@ -6,8 +6,15 @@ OpenCode Zen's free models in magpie and OpenCode. Provider id:
 Since 2026-10-13 Zen's free tier answers 403 FreeTierError ("can only be
 used from within OpenCode") to a request that isn't OpenCode's. From
 0.1.6 the plugin sends a free-model request with exactly the headers
-OpenCode 1.19.0 sends to Zen, and a session id of OpenCode's own form.
-Older versions (0.1.1 on npm before it) get the 403: update the plugin.
+OpenCode sends to Zen, and a session id of OpenCode's own form. Older
+versions (0.1.1 on npm before it) get the 403: update the plugin.
+
+Since 2026-10-26 Zen's free tier answers `429 FreeUsageLimitError`
+("Free models work in OpenCode") to the User-Agent OpenCode 1.19.0 sent,
+which 0.1.6 to 0.1.15 sent on every request. From 0.1.16 the plugin
+sends OpenCode 2.0.25's headers (`opencode/latest/2.0.25/cli`, its
+session id under `x-session-affinity` and `x-session-id` too): update
+the plugin.
 
 ## Install
 

package/models.mjs

@@ -53,6 +53,18 @@ function listedFree(id) {
   return /-free$/.test(id);
 }
 
+// "258K" is the model's working label, not an exact maximum. Zen's live
+// catalog lists deepseek-v4.1-flash-free at 200000, but on 2026-10-26 a
+// request whose upstream usage was in 576 + cache_read 258,048 = 258,624
+// tokens got a 200 from the free tier. That is the measured floor; the
+// later 283,545-token attempt was magpie's own estimate, not upstream
+// usage, so it is not a ceiling. Reporting more than the vendor accepts
+// routes requests that then fail, so the floor is used until a source
+// says more. Remove this when the catalog carries the real value.
+const CONTEXT_FLOOR = {
+  "deepseek-v4.1-flash-free": 258_624,
+};
+
 export function buildModels(ids, provider, protocols, baseURL) {
   const result = {};
   for (const id of ids) {
@@ -81,7 +93,13 @@ export function buildModels(ids, provider, protocols, baseURL) {
       providerID: PROVIDER,
       name: model.name ?? id,
       api: { id, url: baseURL, npm: SDK[protocol] },
-      limit: { context: 32768, output: 4096, ...model.limit },
+      limit: {
+        output: 4096,
+        ...model.limit,
+        context: CONTEXT_FLOOR[id]
+          ? Math.max(model.limit?.context ?? 0, CONTEXT_FLOOR[id])
+          : model.limit?.context ?? 32768,
+      },
       capabilities: {
         temperature: model.temperature ?? true,
         reasoning: model.reasoning ?? false,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@magpie-community/opencode-zen-free-auth",
-  "version": "0.1.15",
+  "version": "0.1.17",
   "description": "OpenCode Zen free models through the public credential, with native streaming protocols",
   "type": "module",
   "main": "./index.mjs",

package/provider.mjs

@@ -33,13 +33,17 @@ const CORE_PARAMETER = {
   grep: "pattern",
   read: "filePath",
 };
-const OPENCODE_VERSION = "1.19.0";
-// What OpenCode 1.19.0 sends as its User-Agent, ai-sdk and Bun included.
-// Each protocol's ai-sdk package bundles its own provider-utils, so the
-// version differs by protocol (captured from the release binary's requests).
-const PROVIDER_UTILS = { chat: "4.0.23", responses: "4.0.40", anthropic: "4.0.46" };
-export function userAgent(protocol) {
-  return `opencode/${OPENCODE_VERSION} ai-sdk/provider-utils/${PROVIDER_UTILS[protocol]} runtime/bun/1.3.14`;
+// What OpenCode 2.0.25 sends as its User-Agent on every protocol:
+// `opencode/${channel}/${version}/${name}` (its App service, read from the
+// release binary and captured from its requests on 2026-10-25). Since
+// 2026-10-26 Zen's free tier answers 429 FreeUsageLimitError ("Free models
+// work in OpenCode") to the exact User-Agent 0.1.6 to 0.1.15 sent, OpenCode
+// 1.19.0's `opencode/1.19.0 ai-sdk/provider-utils/4.0.23 runtime/bun/1.3.14`,
+// whatever the body says; the same request under any other string, OpenCode
+// 2.0.25's included, is served.
+const OPENCODE_VERSION = "2.0.25";
+export function userAgent() {
+  return `opencode/latest/${OPENCODE_VERSION}/cli`;
 }
 // How long a request for a model the last list didn't have waits before it
 // asks Zen's list again: a model Zen made free since is served at once,
@@ -375,24 +379,24 @@ async function server(_input, options = {}) {
             addEnvironment(body, protocol);
             if (protocol === "chat")
               body.stream_options = { ...body.stream_options, include_usage: true };
-            // Only what OpenCode 1.19.0 itself sends to its gateway, so a
-            // caller's own headers (an agent's User-Agent, x-stainless-*,
-            // x-session-affinity, ...) never mark the request as not
-            // OpenCode's (packages/opencode/src/session/llm/request.ts).
+            // Only what OpenCode 2.0.25 itself sends to its gateway, so a
+            // caller's own headers (an agent's User-Agent, x-stainless-*, ...)
+            // never mark the request as not OpenCode's.
             const given = request.headers;
             const session = sessionID(given, body);
             const project = _input?.project;
             const headers = new Headers({
               "content-type": "application/json",
               accept: "*/*",
-              "user-agent": userAgent(protocol),
+              "user-agent": userAgent(),
               "x-opencode-project":
                 given.get("x-opencode-project") ||
                 (project?.vcs === "git" && project.id ? project.id : "global"),
               "x-opencode-session": session,
-              "x-opencode-request": opencodeID("msg", false),
               "x-opencode-client": "cli",
               "x-opencode-session-id": session,
+              "x-session-affinity": session,
+              "x-session-id": session,
             });
             if (protocol === "anthropic") {
               headers.set("x-api-key", "public");