Bludit Wizard

AI-assisted writing, editing, SEO analysis, and content management for Bludit CMS — twelve guided tasks with human-reviewed changes.

Bludit Wizard is a native Bludit plugin. It connects the admin area to a configurable OpenAI-compatible AI provider and guides you through focused writing, editing, SEO and analysis workflows. Generated content is always shown for review: nothing is written to a page until you explicitly approve it, and analysis tasks never modify content.

The plugin is designed for a human publisher. It is not an autonomous site manager and does not publish anything unattended.

Screenshots

Wizard workspace	Review before applying

Features

Write & Create

1. Create a New Page — Generate a complete new page or article draft from instructions you supply. Review the generated content before saving or publishing. Primary action: Generate Draft.
2. Create an Outline — Generate a structured outline for an article, blog post, or other content. Primary action: Generate Outline.
3. Generate Article Ideas — Brainstorm content topics, article concepts, and potential writing directions. Primary action: Generate Ideas.

Edit & Improve

All four tasks require you to select an existing Bludit page.

4. Improve an Existing Page — Analyze existing content and propose improvements to its quality, organization, and clarity. Primary action: Review Improvements.
5. Rewrite Tone or Style — Rewrite an existing page in the requested voice, tone, or writing style. Primary action: Rewrite Content.
6. Expand or Shorten Content — Adjust the length of an existing page while retaining its intended meaning. Primary action: Adjust Length.
7. Proofread Content — Identify and correct grammar, spelling, punctuation, wording, and clarity problems. Primary action: Proofread Page.

SEO & Quality

All four tasks require you to select an existing Bludit page.

8. SEO + Clarity Audit — Analyze a page for search optimization, readability, clarity, and content organization. Primary action: Analyze Page.
9. Generate SEO Metadata — Generate suggested SEO titles, descriptions, and supported metadata for an existing page. Primary action: Generate Metadata.
10. Suggest Internal Links — Identify useful internal linking opportunities based on other relevant pages on the same Bludit site. Suggested targets are checked against the real page index. Primary action: Find Link Opportunities.
11. Audit Content Freshness — Review content for outdated information, information that may need verification, and opportunities for updating. Primary action: Check Content Freshness.

Ask Wizard

12. Ask Wizard — Ask questions about writing, the website, or supported Bludit functionality. Optionally select an existing page to provide relevant context. This task does not modify site content. Primary action: Ask Wizard.

Additional capabilities

• Task-specific instructions, guidance, placeholders and controls.
• Existing-page selection for every task that needs it.
• Configurable OpenAI-compatible providers and models.
• Searchable model discovery with optional free-model and tool-support filters.
• Connection and inference testing for the configured model.
• Context-aware assistance with strict isolation between tasks and pages.
• Human review before changes, with original/proposed comparison.
• Draft and metadata proposal workflows.
• Read-only SEO, internal-link and freshness analysis.
• Markdown result copying.
• Security-focused API communication.

Requirements

• Bludit 3.22.x — tested with Bludit Community 3.22.0 and Bludit Pro 3.22.0.
• PHP 8.0 or newer — tested with PHP 8.2.
• PHP extensions: curl, mbstring, json (bundled with PHP). HTTPS provider endpoints also need working CA certificates.
• Administrator access to the Bludit admin area.
• An API key and base URL for an OpenAI-compatible provider, or a reachable local/self-hosted endpoint.

The plugin has no Composer dependencies and no external runtime service.

Installation

1. Download bludit-wizard-1.1.0.zip from the latest release.
2. Extract the archive. It contains a single wizard/ directory.
3. Upload the wizard/ directory to your Bludit installation's bl-plugins/ folder so the path is bl-plugins/wizard/plugin.php.
4. In Bludit Admin, navigate to Settings → Plugins and activate Bludit Wizard.
5. Open Bludit Wizard's settings to configure your AI provider.
6. Select a Provider preset to automatically populate the Base URL. For a custom OpenAI-compatible provider, enter the Base URL manually.
7. Enter your provider's LLM API key.
8. Click Refresh Models, then select your preferred model from the available results. Use Free models only or Tool support only to filter the catalog when supported.
9. Click Test Model to verify the selected model, then click Save to save your configuration.
10. Open Wizard from the Bludit admin sidebar, choose a task, select an existing page when required, and enter your instructions.

Note: You don't need to enter a Model ID manually when model discovery is working. Manual entry remains available if discovery is unavailable or your preferred model isn't listed.

Upgrading

Replace the entire bl-plugins/wizard/ directory with the new release. Do not copy only plugin.php: the JavaScript, CSS, lib/netpolicy.php and assets.json must all come from the same release. Settings and keys live in bl-content/databases/plugins/wizard/db.php and are preserved.

After upgrading, purge any site/CDN cache and hard-refresh once. Asset URLs carry the release version (for example wizard.js?v=1.1.0) so a cached old script cannot silently break the interface.

Supported AI providers

Bludit Wizard supports configurable OpenAI-compatible AI providers, including OpenAI, OpenRouter, LLM Gateway, and compatible Ollama deployments. OpenRouter and LLM Gateway model discovery, filtering, and connection testing have been verified in live use. Other providers and models may work depending on API compatibility, authentication, capabilities, and network accessibility.

Tested in live use

• OpenRouter — provider configuration, model discovery, model selection, model testing, free-model filtering, and task execution.
• LLM Gateway — provider configuration, model discovery, model selection, model testing, free-model filtering, and task execution.

OpenAI

OpenAI is supported through its OpenAI-compatible API. Use the API base URL provided by OpenAI (for example https://api.openai.com/v1), a model your account can access, and an API key. The maintainer's live acceptance testing covered OpenRouter and LLM Gateway; OpenAI compatibility is by API design.

Ollama and self-hosted endpoints

Ollama exposes an OpenAI-compatible API (commonly http://127.0.0.1:11434/v1). The Ollama service must be reachable from the server that runs Bludit.

• A remote hosting installation cannot ordinarily reach an Ollama instance on your own computer through localhost; that address refers to the hosting server itself.
• Private-network destinations are blocked by the plugin's SSRF protections unless an administrator explicitly enables private-network providers.
• Do not expose an unauthenticated Ollama instance directly to the public internet. Use a secure, explicitly supported network configuration.

The SSRF and private-network protections are not weakened to accommodate a local provider.

Other providers

Services that implement sufficiently compatible /chat/completions and /models endpoints may work. This is compatibility by API design, not individually verified support. A model returned by a provider is not a guarantee that it supports every task or every optional parameter.

Model discovery and testing

• Refresh Models calls the provider's /models endpoint and caches the catalog. The saved model is not changed.
• The catalog is searchable. Select a model or keep typing a model id manually; manual entry always works as a fallback.
• Free models only and Tool support only filters appear when the provider supplies the metadata needed for them. Some providers omit that metadata, so the filters may be hidden or incomplete.
• Test Model verifies URL validity, network reachability, authentication and a minimal inference request. A successful test confirms that the endpoint and credentials work for that request; it does not prove that every task or every advanced parameter works with that model.

How it works

1. Select a task — writing, editing, SEO/quality analysis, or Ask Wizard.
2. Provide instructions — and select an existing page when the task needs one.
3. Generate a proposal or analysis.
4. Review the result. Editable tasks show the original page beside the proposed content.
5. Apply only when authorized — press Apply this change to save an editable proposal. Read-only audits have no Apply action.

Switching task or page clears the previous selection, result and approval state, so content from one page cannot be applied to another.

Human approval and safety

• Explicit approval: no generated content is written until you press Apply this change.
• Read-only analysis: SEO + clarity, internal links and freshness audits never modify pages.
• Stale-edit protection: a proposal is pinned to the page revision it was generated from. If the page changes in another tab, the write is refused.
• Revision check: applying to an existing page requires the matching revision; a missing or stale revision is rejected.
• API credential protection: saved LLM and GPT Action keys are never echoed back into the admin page. Blank submissions keep the stored keys.
• SSRF safeguards: provider URLs are validated; private and restricted destinations are blocked unless explicitly allowed; cloud metadata endpoints are always refused.
• Request safety: CSRF validation, administrator authorization for provider actions, escaped output, and safe Markdown rendering are enforced.
• No page deletion: Bludit Wizard has no deletion action. Use Bludit's Content screen to delete pages.

Troubleshooting

Symptom	Likely cause / fix
No models returned by Refresh Models	Check the Base URL, provider key and network access. Some providers require a different /models path or do not expose a catalog; enter the model id manually.
A model fails tasks	The model may not support the required request shape, context length or parameters. Try a model the provider documents for chat completions.
Provider authentication failure	Re-enter the LLM API key. Saved keys are never displayed; leave the field blank to keep the current key only when it is already correct.
Incorrect Base URL	Use the provider's OpenAI-compatible base URL ending in /v1, not a dashboard or chat page URL.
“The Wizard interface did not finish loading”	The browser is running a missing, blocked or stale script. Purge the site/CDN cache, confirm wizard.js?v=… returns HTTP 200, and reload.
Page selector does not appear	Confirm the complete release was uploaded (especially js/wizard.js and assets.json) and that the asset URL includes the current version.
Old interface after upgrading	The old plugin directory or a cache is still being served. Replace the whole wizard/ directory, purge caches and hard-refresh.
“Wizard unavailable”	The workspace configuration did not load. Re-upload the complete wizard/ directory and reload.

For deployment verification, fetch https://YOUR-SITE/bl-plugins/wizard/assets.json and compare its files["js/wizard.js"] SHA-256 with the served file.

Changelog

v1.1.0

• Twelve guided content workflows across writing, editing, SEO/quality and contextual assistance.
• Expanded writing and editing tools: tone/style rewriting, length adjustment, proofreading, SEO metadata, internal-link suggestions and content-freshness auditing.
• Task-specific page selection, instructions and primary action labels.
• Full-width workspace with original/proposed review and explicit approval.
• Configurable OpenAI-compatible providers, searchable model discovery and model testing.
• Stability corrections following the 1.1.0 release candidate: reliable form dispatch, server-rendered page selection, safe degraded-JavaScript handling and versioned assets.

See wizard/CHANGELOG.md for the complete history.

Development

The repository includes an isolated Docker test harness:

testenv/fetch-upstream.sh
docker compose -f testenv/compose.yaml up -d --build
tests/run-all.sh local

Requirements: Docker, Node.js 20+, and PHP 8.2 for the network-policy unit tests. The AI provider used by the tests is a deterministic local mock; no real API keys are required. Browser tests use Playwright.

License

Bludit Wizard is licensed under the GNU General Public License, version 3 only (SPDX: GPL-3.0-only). See LICENSE.

Copyright © 2026 Evan Zellner (@gitdamnit).

See docs/LICENSING.md for the third-party audit and the source of the official license text. Bludit itself is a separate project with its own license; this plugin does not alter or redistribute Bludit.

Contributing

Issues and pull requests are welcome at gitdamnit/bludit-wizard. Please:

• describe the Bludit and PHP versions you tested;
• include the provider and model only when relevant (never include API keys);
• run the test suite and include the result when changing runtime code;
• keep changes focused and preserve the human-approval safety model.

Do not include secrets, private site content or production credentials in issues, screenshots or pull requests.