knowhycodata · Oct 29, 2026
diff --git a/README.md b/README.md
@@ -120,6 +120,7 @@ To use another provider or model:
 
 - **Another OpenAI model.** Change `default_model` in `openbot/examples/openduet/model.yaml`, or in your copy of it (see [Make it your own](#make-it-your-own)). `BOT_MODEL` does not change an OpenAI model on its own. It is used only when `OPENAI_BASE_URL` is also set to an OpenAI-compatible endpoint, and then it names the model that endpoint serves.
 - **Anthropic.** Set `BOT_PROVIDER=anthropic` and `ANTHROPIC_API_KEY=...`. The model is `BOT_MODEL` if you set one, such as `BOT_MODEL=claude-sonnet-5-5`, and `claude-sonnet-4-5` if you do not.
+- **LLMTR.** Set `OPENAI_BASE_URL=https://llmtr.com/v1`, `OPENAI_API_KEY` to an LLMTR key (it starts with `llmtr-`), and `BOT_MODEL` to a catalogue id such as `qwen/qwen3.8-27b-free`. Leave `BOT_PROVIDER` as `openai`. The id is sent as published, slash included. Keys are created from [LLMTR's API keys page](https://llmtr.com/docs/api-keys/); `GET https://llmtr.com/v1/models` lists models and needs no key. `llmtr/qwen3-8-27b` is the Turkey-hosted alternative. Leave `BOT_REASONING_EFFORT` unset.
 
 Set these before you start it. **Settings → Setup** in the app says whether the model key works.
 

diff --git a/openbot/.env.example b/openbot/.env.example
@@ -251,6 +251,13 @@ OPENAI_API_KEY=
 #   OPENAI_API_KEY=...
 #   BOT_MODEL=openai/gpt-5.5
 #
+# LLMTR (https://llmtr.com/docs/api-keys/) is one such gateway. Leave BOT_PROVIDER as openai.
+# The model id is a catalogue id, sent verbatim. Leave BOT_REASONING_EFFORT unset.
+#
+#   OPENAI_BASE_URL=https://llmtr.com/v1
+#   OPENAI_API_KEY=llmtr-...
+#   BOT_MODEL=qwen/qwen3.8-27b-free
+#
 # OPENAI_BASE_URL=
 
 # Optional dictation, independent of every agent/chat provider. Set provider, base URL,

diff --git a/openbot/CHANGELOG.md b/openbot/CHANGELOG.md
@@ -35,6 +35,13 @@ Newest first. `Unreleased` is what is on `main` and not yet tagged.
   request for an administrator to switch on a work app, grant it to a Bot, or connect a Shared app's
   team account is kept, with who answered it and when. It starts empty and nothing is backfilled.
 
+### Desktop setup offers LLMTR
+
+Desktop setup offers LLMTR beside Google Gemini and xAI. The row uses `https://llmtr.com/v1` and an
+API key, and the model name starts as `qwen/qwen3.8-27b-free`. A deployment configured by hand sets
+`OPENAI_BASE_URL` to that address, `OPENAI_API_KEY` to an LLMTR key, and `BOT_MODEL` to a catalogue
+id. `BOT_PROVIDER` stays `openai`.
+
 ### Plainer words, and one word for each thing
 
 OpenBot now says **Bot** everywhere a person reads it, where screens used to mix Bot, agent, coworker

diff --git a/openbot/desktop/src-tauri/src/provider.rs b/openbot/desktop/src-tauri/src/provider.rs
@@ -1,7 +1,8 @@
 //! The model provider screen's list, as data.
 //!
-//! OpenAI and Anthropic offer plan sign-in alongside API keys. Google and xAI use named endpoint
-//! presets with API keys; their official addresses are supplied by the frontend.
+//! OpenAI and Anthropic offer plan sign-in alongside API keys. Google, xAI and LLMTR use named
+//! endpoint presets with API keys; their official addresses are supplied by the frontend. Google
+//! and xAI also offer sign-in.
 //!
 //! The row that asks for a URL is the last one, and it is the only one that asks. Keeping it there
 //! is what keeps a base URL off the main path, which the audience rule at the top of the build doc
@@ -104,6 +105,14 @@ pub fn catalogue() -> Vec<Provider> {
             mark: None,
             caution: None,
         },
+        Provider {
+            id: "llmtr".into(),
+            name: "LLMTR".into(),
+            summary: "Use an LLMTR API key.".into(),
+            logins: vec![Login::Endpoint],
+            mark: None,
+            caution: None,
+        },
         Provider {
             id: "openai-compatible".into(),
             name: "Any OpenAI-compatible endpoint".into(),
@@ -129,7 +138,7 @@ mod tests {
             .filter(|p| p.logins.contains(&Login::Endpoint))
             .map(|p| p.id)
             .collect();
-        assert_eq!(asking, vec!["google", "xai", "openai-compatible"]);
+        assert_eq!(asking, vec!["google", "xai", "llmtr", "openai-compatible"]);
         for provider in catalogue()
             .into_iter()
             .filter(|provider| provider.logins.contains(&Login::Endpoint))
@@ -156,12 +165,19 @@ mod tests {
 
     /// Named providers stay ahead of the custom endpoint escape hatch.
     #[test]
-    fn four_named_providers_and_one_custom_endpoint() {
+    fn five_named_providers_and_one_custom_endpoint() {
         let rows = catalogue();
         let ids: Vec<&str> = rows.iter().map(|provider| provider.id.as_str()).collect();
         assert_eq!(
             ids,
-            vec!["openai", "anthropic", "google", "xai", "openai-compatible"]
+            vec![
+                "openai",
+                "anthropic",
+                "google",
+                "xai",
+                "llmtr",
+                "openai-compatible"
+            ]
         );
     }
 

diff --git a/openbot/desktop/src/ProviderPicker.test.tsx b/openbot/desktop/src/ProviderPicker.test.tsx
@@ -53,6 +53,13 @@ const cloudProviders = [
     model: "grok-4.7",
     keyUrl: "https://console.x.ai/",
   },
+  {
+    id: "llmtr",
+    name: "LLMTR",
+    baseUrl: "https://llmtr.com/v1",
+    model: "qwen/qwen3.8-27b-free",
+    keyUrl: "https://llmtr.com/docs/api-keys/",
+  },
 ];
 
 const allProviders: Provider[] = [
@@ -849,56 +856,60 @@ test("OAuth tab switch cancels a late authorization before opening the browser",
 });
 
 for (const provider of cloudProviders) {
-  test(`${provider.name} OAuth completes through native storage without exposing tokens`, async () => {
-    invokeHandler = async (command) => {
-      if (command === "providers")
-        return allProviders.map((row) =>
-          row.id === provider.id
-            ? { ...row, logins: ["endpoint", "oauth"] }
-            : row,
-        );
-      if (command === "begin_model_oauth")
-        return {
-          attemptId: "synthetic-attempt",
-          url: "https://authorization.example.test/approve",
-          userCode: "SYNTHETIC",
-        };
-      if (
-        command === "plugin:opener|open_url" ||
-        command === "finish_model_oauth"
-      )
-        return null;
-      throw new Error(`unexpected command ${command}`);
-    };
-    const choices: unknown[] = [];
-    const view = await renderPicker((choice) => choices.push(choice));
-    const user = userEvent.setup({ document: view.container.ownerDocument });
-    await user.click(
-      await view.findByRole("radio", { name: new RegExp(provider.name) }),
-    );
-    await user.click(view.getByRole("tab", { name: "Sign in" }));
-    await user.click(
-      view.getByRole("button", { name: `Sign in with ${provider.name}` }),
-    );
-    await view.findByText(`Signed in to ${provider.name}.`);
-    await user.click(view.getByRole("button", { name: "Continue" }));
-    expect(choices).toEqual([
-      {
-        provider: provider.id,
-        login: "oauth",
-        model: provider.model,
-        saved: true,
-      },
-    ]);
-    expect(invokeCalls).toContainEqual({
-      command: "begin_model_oauth",
-      args: { root: "/tmp/openbot-provider-root", provider: provider.id },
-    });
-    expect(invokeCalls).toContainEqual({
-      command: "finish_model_oauth",
-      args: { attemptId: "synthetic-attempt" },
+  // LLMTR's row is an API key only. This case injects a sign-in login, so it runs for the
+  // providers that offer one.
+  if (provider.id !== "llmtr") {
+    test(`${provider.name} OAuth completes through native storage without exposing tokens`, async () => {
+      invokeHandler = async (command) => {
+        if (command === "providers")
+          return allProviders.map((row) =>
+            row.id === provider.id
+              ? { ...row, logins: ["endpoint", "oauth"] }
+              : row,
+          );
+        if (command === "begin_model_oauth")
+          return {
+            attemptId: "synthetic-attempt",
+            url: "https://authorization.example.test/approve",
+            userCode: "SYNTHETIC",
+          };
+        if (
+          command === "plugin:opener|open_url" ||
+          command === "finish_model_oauth"
+        )
+          return null;
+        throw new Error(`unexpected command ${command}`);
+      };
+      const choices: unknown[] = [];
+      const view = await renderPicker((choice) => choices.push(choice));
+      const user = userEvent.setup({ document: view.container.ownerDocument });
+      await user.click(
+        await view.findByRole("radio", { name: new RegExp(provider.name) }),
+      );
+      await user.click(view.getByRole("tab", { name: "Sign in" }));
+      await user.click(
+        view.getByRole("button", { name: `Sign in with ${provider.name}` }),
+      );
+      await view.findByText(`Signed in to ${provider.name}.`);
+      await user.click(view.getByRole("button", { name: "Continue" }));
+      expect(choices).toEqual([
+        {
+          provider: provider.id,
+          login: "oauth",
+          model: provider.model,
+          saved: true,
+        },
+      ]);
+      expect(invokeCalls).toContainEqual({
+        command: "begin_model_oauth",
+        args: { root: "/tmp/openbot-provider-root", provider: provider.id },
+      });
+      expect(invokeCalls).toContainEqual({
+        command: "finish_model_oauth",
+        args: { attemptId: "synthetic-attempt" },
+      });
     });
-  });
+  }
 
   test(`${provider.name} opens its API key page externally without sending entered credentials`, async () => {
     invokeHandler = async (command) => {

diff --git a/openbot/desktop/src/ProviderPicker.tsx b/openbot/desktop/src/ProviderPicker.tsx
@@ -78,6 +78,12 @@ const endpointPresets: Record<
     model: "grok-4.7",
     keyUrl: "https://console.x.ai/",
   },
+  // https://llmtr.com/docs/quickstart/
+  llmtr: {
+    baseUrl: "https://llmtr.com/v1",
+    model: "qwen/qwen3.8-27b-free",
+    keyUrl: "https://llmtr.com/docs/api-keys/",
+  },
 };
 
 function endpointIdentity(baseUrl: string | undefined): string {
@@ -133,8 +139,8 @@ export function recordedModel(held: HeldConfiguration): ModelChoice | null {
 /**
  * Connect a model.
  *
- * Plans remain the default wherever supported. Google and xAI use the same endpoint credential
- * route as custom models, while their named rows supply the address for the user.
+ * Plans remain the default wherever supported. Google, xAI and LLMTR use the same endpoint
+ * credential route as custom models, while their named rows supply the address for the user.
  */
 export function ProviderPicker({
   chosen,

diff --git a/openbot/docs/configuration.md b/openbot/docs/configuration.md
@@ -235,6 +235,16 @@ model:
   default_model: openai/gpt-5.6-terra
 ```
 
+[LLMTR](https://llmtr.com) is one such gateway. Its address is `https://llmtr.com/v1`, the key goes in `OPENAI_API_KEY` (an LLMTR key starts with `llmtr-`), and `BOT_PROVIDER` stays `openai`:
+
+```sh
+OPENAI_BASE_URL=https://llmtr.com/v1
+OPENAI_API_KEY=llmtr-...
+BOT_MODEL=qwen/qwen3.8-27b-free
+```
+
+`BOT_MODEL` is the catalogue id, sent verbatim, so an id that already contains a slash is written in full. `qwen/qwen3.8-27b-free` accepts chat completions and tools. `llmtr/qwen3-8-27b` is the same shape and is hosted in Turkey. `GET https://llmtr.com/v1/models` lists the catalogue and needs no key. Keys are created from the [API keys page](https://llmtr.com/docs/api-keys/). Leave `BOT_REASONING_EFFORT` unset: that setting is sent as an object, and LLMTR accepts `reasoning` only as a boolean.
+
 Most gateways publish a model list, which is the way to check a name before configuring it.
 
 Two things are worth knowing before pointing a deployment at any gateway. Not every catalogue entry accepts tools, and a Bot without tool calling cannot drive its computer; the model list says which do. And `BOT_RESPONSES_API=true` needs an endpoint that implements the Responses API, not only chat completions.