docs: use uvx for the PyPI quickstart

docs/guides/python-installation.md

@@ -46,11 +46,14 @@ Use `--vllm-port` and the API-key environment-variable options on the launcher i
 
 ### Run with uvx
 
-For an isolated base-package invocation:
+With uv installed, run the packaged Rust CLI in an isolated environment without a global installation.
+Install Codex or Claude Code separately and connect to an existing inference server:
 
 ```bash
-uvx --from agentic-api==0.7.0 agentic-api doctor
-uvx --from agentic-api==0.7.0 agentic-api serve --vllm-base-url http://existing-vllm:8000
+uvx --from agentic-api==0.7.0 agentic --version
+uvx --from agentic-api==0.7.0 agentic run codex --upstream http://existing-vllm:8000
+uvx --from agentic-api==0.7.0 agentic run claude --upstream http://existing-vllm:8000
+uvx --from agentic-api==0.7.0 agentic serve --upstream http://existing-vllm:8000
 ```
 
 ### Install a workflow artifact

website/README.md

@@ -80,7 +80,7 @@ Integration copy and commands were checked against the project's [README](https:
 
 The quickstart also links to the [Codex Desktop guide](../docs/guides/codex-desktop.md), with the verified Linux setup and freeform `apply_patch` limitation. The guide appears in `/docs/latest` and `/llms.txt`; tagged documentation snapshots omit it until their source contains the guide.
 
-The quickstart defaults to the published `agentic-server` crate and offers PyPI and build-from-source tabs. Both registry install commands pin version 0.7.0. Package installs launch `agentic`; source builds launch `./target/debug/agentic`. The optional WebMCP tool accepts an `installation` value (`crates`, `pypi`, or `source`) and defaults to crates.io. `PUBLISHED_VERSION` in `lib/quickstart.ts` supplies the install commands and visible release labels; update it and `public/llms.txt` after publishing a new release. The base Python wheel bundles the Rust executables and does not install vLLM.
+The quickstart defaults to the published `agentic-server` crate and offers PyPI and build-from-source tabs. Both registry options pin version 0.7.0. PyPI uses `uvx --from agentic-api==0.7.0 agentic` for version checks, serving, and harness launches without a global installation. Cargo installs launch `agentic`; source builds launch `./target/debug/agentic`. The optional WebMCP tool accepts an `installation` value (`crates`, `pypi`, or `source`) and defaults to crates.io. `PUBLISHED_VERSION` in `lib/quickstart.ts` supplies the install commands and visible release labels; update it and `public/llms.txt` after publishing a new release. The base Python wheel bundles the Rust executables and does not install vLLM.
 
 The community information architecture is inspired by the supplied vLLM Semantic Router pages. The layout, visual system, and wording were created for this site. The website does not run an agent or contact a model: the launch instructions are examples for a user's own environment.
 

website/components/site/quickstart.tsx

@@ -136,10 +136,10 @@ export function Quickstart() {
             <li>
               <span>02</span>
               <div>
-                <strong>Install Agentic API</strong>
+                <strong>Get Agentic API</strong>
                 <p>
-                  Install v{PUBLISHED_VERSION} from PyPI or crates.io, with
-                  source builds available for development.
+                  Run v{PUBLISHED_VERSION} from PyPI with uvx or install from
+                  crates.io, with source builds available for development.
                 </p>
               </div>
             </li>
@@ -193,7 +193,7 @@ export function Quickstart() {
                     method === 'source'
                       ? 'Build from source'
                       : method === 'pypi'
-                        ? `Install from PyPI (v${PUBLISHED_VERSION})`
+                        ? `Check the PyPI CLI (v${PUBLISHED_VERSION})`
                         : `Install from crates.io (v${PUBLISHED_VERSION})`
                   }
                 />

website/lib/quickstart.ts

@@ -1,6 +1,7 @@
 import { REPO } from './site';
 
 export const PUBLISHED_VERSION = '0.7.0';
+const PYPI_EXECUTABLE = `uvx --from agentic-api==${PUBLISHED_VERSION} agentic`;
 
 export const INSTALL_METHODS = {
   crates: {
@@ -10,8 +11,8 @@ export const INSTALL_METHODS = {
   },
   pypi: {
     label: 'PyPI',
-    command: `python -m pip install agentic-api==${PUBLISHED_VERSION}`,
-    note: 'Requires Python 3.10+. The wheel includes the gateway and agentic CLI; vLLM is installed separately.',
+    command: `${PYPI_EXECUTABLE} --version`,
+    note: 'Requires uv. Runs the released agentic CLI in an isolated environment; no global install needed. vLLM is served separately.',
   },
   source: {
     label: 'Build from source',
@@ -33,15 +34,19 @@ export function isInstallMethod(value: unknown): value is InstallMethod {
 
 export function getServeCommand(installation: InstallMethod) {
   if (installation === 'pypi')
-    return 'python -m agentic_api serve --vllm-base-url http://127.0.0.1:5050';
+    return PYPI_EXECUTABLE + ' serve --upstream http://127.0.0.1:5050';
   const executable =
     installation === 'source' ? './target/debug/agentic' : 'agentic';
   return executable + ' serve --upstream http://127.0.0.1:5050';
 }
 
 export function getLaunchCommands(installation: InstallMethod) {
   const executable =
-    installation === 'source' ? './target/debug/agentic' : 'agentic';
+    installation === 'source'
+      ? './target/debug/agentic'
+      : installation === 'pypi'
+        ? PYPI_EXECUTABLE
+        : 'agentic';
   const options =
     ' \\\n  --upstream http://127.0.0.1:5050 \\\n  --model Qwen/Qwen3-30B-A3B-FP8';
   return {

website/public/llms.txt

@@ -4,7 +4,7 @@
 
 vLLM serves the model; Agentic API coordinates the tool loop and state. The project provides Responses and Messages API surfaces, conversation storage, streaming, and WebSocket transport for Responses. Built-in web search and MCP tools can execute on the gateway; client-executed function tools remain in the harness.
 
-Version 0.7.0 is available on PyPI and crates.io. Install with `python -m pip install agentic-api==0.7.0` (Python 3.10+) or `cargo install agentic-server --version 0.7.0 --locked` (Rust and Cargo). Both install the Rust gateway and agentic CLI; the Python wheel does not install vLLM. Start the standalone server with `python -m agentic_api serve --vllm-base-url http://127.0.0.1:5050`, or use `agentic run codex` or `agentic run claude` with your upstream and model.
+Version 0.7.0 is available on PyPI and crates.io. With uv installed, run `uvx --from agentic-api==0.7.0 agentic run codex --upstream http://127.0.0.1:5050` (or `run claude`) without a global install. To start only the gateway, use `uvx --from agentic-api==0.7.0 agentic serve --upstream http://127.0.0.1:5050`. Alternatively, install with `cargo install agentic-server --version 0.7.0 --locked` (Rust and Cargo), then run `agentic`. The Python wheel bundles the Rust executables; serve vLLM separately.
 
 Use a tool-capable model and the appropriate tool-calling configuration. SGLang and NVIDIA Dynamo also have documented, tested upstream configurations. Compatibility depends on the model, serving configuration, and API surface. The separate agentic-llm-d component provides response hydration and persistence for llm-d.