release: prepare 0.1.0b1

.github/workflows/publish.yml

@@ -1,11 +1,11 @@
-name: Publish Python alpha
+name: Publish Python pre-release
 on:
   workflow_dispatch:
     inputs:
       version:
         description: Exact version in pyproject.toml
         required: true
-        default: '0.1.0a1'
+        default: '0.1.0b1'
 permissions:
   contents: read
 jobs:
@@ -25,11 +25,13 @@ jobs:
           RELEASE_VERSION: ${{ inputs.version }}
         run: |
           python - <<'PY'
-          import os, tomllib
+          import os, re, tomllib
           with open('pyproject.toml', 'rb') as f:
               version = tomllib.load(f)['project']['version']
           assert version == os.environ['RELEASE_VERSION'], 'release version mismatch'
-          assert 'a' in version, 'this workflow publishes alpha versions only'
+          assert re.fullmatch(r'\d+\.\d+\.\d+(a|b)\d+', version), (
+              'this workflow publishes alpha (aN) or beta (bN) pre-releases only'
+          )
           PY
       - run: python -m pip install build twine
       - run: python -m build

CHANGELOG.md

@@ -0,0 +1,25 @@
+# Changelog
+
+## 0.1.0b1
+
+First beta of the Python AI Protection SDK. The `webdecoy_ai_protection` package
+API is unchanged from `0.1.0a1`; this release moves the package from alpha to beta.
+
+- Add a runnable Ollama example (`examples/ollama_budget.py`) that wraps one
+  non-streaming `ollama.AsyncClient.generate` call in a per-attempt model budget,
+  with pinned example requirements (`examples/requirements-ollama.txt`).
+- Add tests for the Ollama example against the pinned `ollama==0.6.3` Python SDK
+  with an in-memory HTTP transport: serialization, usage normalization, denial
+  before the provider, privacy, cancellation and accounting failures. `ollama` is
+  added to the `dev` extra only; it is not a runtime dependency.
+- Document the Ollama example in the README.
+- The publish workflow now accepts beta (`bN`) as well as alpha (`aN`)
+  pre-release versions.
+- Update maturity labels and install instructions to `0.1.0b1`.
+
+```sh
+python -m pip install 'webdecoy-ai-protection[fastapi]==0.1.0b1'
+```
+
+Earlier releases are described in the
+[GitHub releases](https://github.com/WebDecoy/ai-protection-python/releases).

README.md

@@ -3,7 +3,7 @@
 Async request admission for Python AI endpoints. Local application rules run in
 your process; bot detection runs in WebDecoy. Apache-2.0 licensed.
 
-**Alpha preview.** This first implementation covers
+**Beta.** This release covers
 admission, shared account/session quotas, reporting and an explicit FastAPI/Starlette
 route wrapper, including optional concurrency leases across streamed responses.
 Per-attempt model budgets and usage reporting are also available. Browser evidence
@@ -13,10 +13,10 @@ and MCP adapters are not yet implemented. It is not a prompt-injection filter or
 
 Requires Python 3.11+ and asyncio. HTTPX is the only core runtime dependency.
 FastAPI/Starlette support is optional; synchronous applications and Trio are not
-supported in this preview.
+supported in this beta.
 
 ```sh
-python -m pip install 'webdecoy-ai-protection[fastapi]==0.1.0a1'
+python -m pip install 'webdecoy-ai-protection[fastapi]==0.1.0b1'
 ```
 
 For core-only applications, omit `[fastapi]`. For development from this checkout,
@@ -274,7 +274,7 @@ Use `BudgetCompletion(value, usage)` to preserve provider output separately from
 accounting. A return value without that wrapper is preserved with unknown usage.
 For streaming, consume the stream **inside** the callback and await terminal usage;
 returning a `StreamingResponse` or iterator is not completion. Arrange budget
-denial before committing HTTP headers. This preview has no automatic budgeted
+denial before committing HTTP headers. This beta has no automatic budgeted
 HTTP-streaming bridge. When composing concurrency, wrap the complete budgeted
 provider attempt inside the concurrency-owned work; FastAPI's concurrency wrapper
 already owns its handler and full response lifecycle.
@@ -363,7 +363,7 @@ not detection effectiveness, customer savings, or production capacity.
 
 [`examples/ollama_budget.py`](examples/ollama_budget.py) wraps a real
 [`ollama.AsyncClient.generate`](https://github.com/ollama/ollama-python) call
-with the existing published WebDecoy SDK. Install the example dependencies:
+with the published WebDecoy SDK. Install the example dependencies:
 
 ```sh
 python -m pip install -r examples/requirements-ollama.txt

examples/requirements-ollama.txt

@@ -1,2 +1,2 @@
-webdecoy-ai-protection==0.1.0a1
+webdecoy-ai-protection==0.1.0b1
 ollama==0.6.3

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "webdecoy-ai-protection"
-version = "0.1.0a1"
+version = "0.1.0b1"
 description = "Async request admission for Python AI endpoints using WebDecoy"
 readme = "README.md"
 requires-python = ">=3.11"