Skip to content

[Improvement] LLM indexability #372

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
g-despot merged 8 commits into from
Mar 20, 2026
Merged

[Improvement] LLM indexability #372

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
94e6e20
Initial LLM indexability testing
g-despot Mar 6, 2026
89a118e
Update tests
g-despot Mar 6, 2026
c804646
Merge branch 'main' into llm-indexability
g-despot Mar 20, 2026
203dde6
Trigger workflow
g-despot Mar 20, 2026
c906350
Update model
g-despot Mar 20, 2026
ced4b14
Update workflow
g-despot Mar 20, 2026
591604f
Add LLMs.txt note and tests
g-despot Mar 20, 2026
f7e6f54
Minor update
g-despot Mar 20, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
62 changes: 62 additions & 0 deletions .github/workflows/indexability_tests.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,62 @@
name: Docs Indexability Tests

permissions:
contents: read

on:
schedule:
# Run every Sunday at 22:00 UTC
- cron: "0 22 * * 0"
push:
branches:
- testing-ci
workflow_dispatch:

env:
PYTHON_VERSION: "3.10"

jobs:
test-indexability:
name: Test Indexability
runs-on: ubuntu-latest
timeout-minutes: 15

steps:
- name: Checkout code
uses: actions/checkout@v4

- name: Set up test environment
uses: ./.github/actions/setup-test-env
with:
python-version: ${{ env.PYTHON_VERSION }}

- name: Record test start time
run: echo "TEST_START_TIME=$(date +%s)" >> $GITHUB_ENV

- name: Run HTML structure tests
id: html-tests
continue-on-error: true
run: |
echo "Running HTML indexability tests..."
uv run pytest -m "indexability" -v --tb=short --junitxml=pytest-results.xml

- name: Run agent tests
id: agent-tests
continue-on-error: true
if: env.ANTHROPIC_API_KEY != '' && env.OPENAI_API_KEY != ''
Copy link

Copilot AI Mar 6, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The if condition references env.ANTHROPIC_API_KEY and env.OPENAI_API_KEY, but these environment variables are defined in this step's own env: block (lines 47-49). In GitHub Actions, the step-level env: is not available during evaluation of the same step's if: condition — the if: is evaluated before the step's environment is set up. Since these secrets are not defined in the workflow-level env: block either, both values will always be empty, and the agent tests will never run.

Use secrets directly in the condition instead: if: ${{ secrets.ANTHROPIC_API_KEY != '' && secrets.OPENAI_API_KEY != '' }}

Suggested change
if: env.ANTHROPIC_API_KEY != '' && env.OPENAI_API_KEY != ''
if: ${{ secrets.ANTHROPIC_API_KEY != '' && secrets.OPENAI_API_KEY != '' }}

Copilot uses AI. Check for mistakes.
env:
ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
run: |
echo "Running agent indexability tests..."
uv run pytest -m "indexability_agents" -v --tb=short --junitxml=pytest-results-agents.xml

- name: Handle test results
if: always()
uses: ./.github/actions/handle-test-results
with:
test-outcome: ${{ steps.html-tests.outcome }}
Copy link

Copilot AI Mar 6, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The test-outcome only considers the HTML tests outcome (steps.html-tests.outcome), ignoring the agent tests outcome entirely. If the HTML tests pass but agent tests fail, the workflow will report success and no failure notification will be sent. Consider combining both outcomes, e.g., reporting failure if either step failed: test-outcome: ${{ steps.html-tests.outcome == 'success' && (steps.agent-tests.outcome == 'success' || steps.agent-tests.outcome == 'skipped') && 'success' || 'failure' }}

Suggested change
test-outcome: ${{ steps.html-tests.outcome }}
test-outcome: ${{ steps.html-tests.outcome == 'success' && (steps.agent-tests.outcome == 'success' || steps.agent-tests.outcome == 'skipped') && 'success' || 'failure' }}

Copilot uses AI. Check for mistakes.
test-type: 'Indexability'
languages-tested: 'HTML + Claude + ChatGPT'
artifact-name: 'test-results-indexability'
slack-bot-token: ${{ secrets.TESTING_CI_SLACK_BOT }}
3 changes: 3 additions & 0 deletions pyproject.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,8 +5,11 @@ description = "Weaviate documentation test suite"
readme = "README.md"
requires-python = ">=3.10"
dependencies = [
"anthropic>=0.40.0",
"beautifulsoup4>=4.12.0",
"datasets>=4.0.0",
"ijson>=3.3.0",
"openai>=1.50.0",
"pandas>=2.2.3",
"pytest>=8.3.5",
"python-dotenv>=1.1.1",
Expand Down
2 changes: 2 additions & 0 deletions pytest.ini
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,3 +7,5 @@ markers =
ts: groups tests by language (typescript)
java_v6: groups tests by language (java v6 client)
csharp: groups tests by language (c#)
indexability: HTML structure tests for docs indexability (no API keys)
indexability_agents: AI agent tests for docs indexability (requires API keys)
115 changes: 115 additions & 0 deletions tests/README-INDEXABILITY.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,115 @@
# Docs Indexability Tests

This test suite validates that AI agents (Claude, ChatGPT) and web search crawlers can access all documentation content from the live docs site at [docs.weaviate.io](https://docs.weaviate.io).

## Why these tests exist

Documentation content that is hidden behind JavaScript interactions (tabs, collapsibles) or lazy-loaded may be invisible to AI agents and search crawlers. These tests verify that:

1. All HTML content is present in the server-rendered page (tabs, collapsibles, code snippets)
2. Claude can fetch and read page content via its `web_fetch` tool
3. ChatGPT can find and read page content via its `web_search_preview` tool
4. `llms.txt` and `sitemap.xml` are accessible and well-formed

## Test structure

| Part | Marker | API keys needed | Description |
|------|--------|-----------------|-------------|
| HTML structure | `indexability` | None | Fetches pages with `requests` + `beautifulsoup4` and checks HTML |
| Claude agent | `indexability_agents` | `ANTHROPIC_API_KEY` | Uses Anthropic API with `web_fetch` tool |
| ChatGPT agent | `indexability_agents` | `OPENAI_API_KEY` | Uses OpenAI Responses API with `web_search_preview` tool |

### HTML structure tests (Part 1)

These tests fetch ~10 representative pages from all doc sections and check:

| Test | What it checks |
|------|---------------|
| `test_page_returns_200` | All pages return HTTP 200 |
| `test_meta_tags` | `<title>`, `<meta description>`, `og:title`, `og:description` present |
| `test_heading_hierarchy` | Exactly 1 `<h1>`, content pages have `<h2>`s |
| `test_tabbed_code_blocks_all_present` | ALL tab panels have content in HTML (not just active tab) |
| `test_code_blocks_present` | Pages with code have non-empty `<pre><code>` blocks |
| `test_details_content_present` | `<details>` elements have body content (not lazy-loaded) |
| `test_images_have_alt_text` | Content images have alt text (excludes decorative SVGs/icons) |
| `test_llms_txt_accessible` | `/llms.txt` returns 200, has substantial content, mentions Weaviate |
| `test_sitemap_accessible` | `/sitemap.xml` returns 200, has 100+ URLs |

### Claude agent tests (Part 2)

Uses Claude Haiku with the `web_fetch` tool to verify Claude can:

- **Fetch quickstart code tabs** — fetches `/weaviate/quickstart` and must extract actual code lines for all 5 languages (Python, TypeScript, Go, Java, C#). Asserts on language-specific code tokens like `near_text` (Python), `nearText` (TS), `NearTextArgBuilder` (Go), etc.
- Fetch a page with collapsible sections and read the content
- Fetch `/llms.txt` and identify it as Weaviate documentation

### ChatGPT agent tests (Part 3)

Uses GPT-4.1 Mini with the `web_search_preview` tool to verify ChatGPT can:

- **Search quickstart code tabs** — searches for `/weaviate/quickstart` and must extract actual code lines for all 5 languages. Uses the same language-specific code token assertions as the Claude test.
Copy link

Copilot AI Mar 6, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The README documentation has multiple inaccuracies compared to the actual test code:

  1. It states the tests assert on "language-specific code tokens like near_text (Python), nearText (TS), NearTextArgBuilder (Go), etc." but the actual code asserts on vectorizer configuration lines like Configure.Vectors.text2vec_weaviate(), vectors.text2VecWeaviate(), etc. (see QUICKSTART_VECTORIZER_LINES in test_docs_indexability.py lines 297-303).

  2. Both lines 42 and 50 say the tests require "all 5 languages", but the ChatGPT test actually only requires 3 out of 5 (line 501 of test_docs_indexability.py).

These appear to be leftover from an earlier version of the tests and should be updated to match the current implementation.

Copilot uses AI. Check for mistakes.
- Search for and read Weaviate Cloud quickstart docs
- Search for and read Weaviate Agents docs

## Running the tests

```bash
# HTML structure tests only (no API keys needed)
uv run pytest -m indexability -v

# Agent tests only (requires ANTHROPIC_API_KEY and OPENAI_API_KEY)
uv run pytest -m indexability_agents -v

# All indexability tests
uv run pytest -m "indexability or indexability_agents" -v
```

## CI workflow

The tests run via `.github/workflows/indexability_tests.yml`:

- **Schedule**: Every Sunday at 22:00 UTC
- **Manual**: Via workflow dispatch
- **Branch**: Runs on push to `testing-ci`
- **Runtime**: ~15 minutes maximum

HTML structure tests always run. Agent tests only run if `ANTHROPIC_API_KEY` and `OPENAI_API_KEY` secrets are configured.

## Test pages

The suite tests ~10 representative URLs covering all doc sections:

| Page | Features tested |
|------|----------------|
| `/weaviate/manage-collections/collection-operations` | tabs, code, details |
| `/weaviate/search/similarity` | tabs, code |
| `/weaviate/search/hybrid` | tabs, code |
| `/weaviate/connections/connect-cloud` | tabs, code |
| `/weaviate/config-refs/collections` | details, table |
| `/weaviate/concepts/data-import` | images |
| `/cloud/quickstart` | code, images |
| `/cloud/manage-clusters/create` | images |
| `/agents/query/tutorial-ecommerce` | code |
| `/weaviate/search` | landing page |

## Dependencies

- `beautifulsoup4` — HTML parsing
- `requests` — HTTP fetching (already in project)
- `anthropic` — Claude API for agent tests
- `openai` — OpenAI API for agent tests

All are listed in the root `pyproject.toml`.

## Adding test pages

To test additional pages, add entries to the `TEST_PAGES` list in `tests/test_docs_indexability.py`:

```python
TEST_PAGES = [
("/path/to/page", {"tabs", "code", "details", "images", "table"}),
# ...
]
```

Available feature tags: `tabs`, `code`, `details`, `images`, `table`. Pages are parametrized — each feature tag enables the corresponding structural test for that page.
Loading
Loading