Getting started
VectorPress adds AI-powered search and chat to a WordPress site. It turns your published posts and media files into vector embeddings, stores them in a vector database, and uses that index to power two Gutenberg blocks: a semantic search box and a retrieval-augmented chat agent that answers questions from your own content.
You bring your own provider accounts. VectorPress calls a model provider for embeddings, chat completions, and image analysis, and a vector database for storage. Both are billed to your accounts, not to VectorPress.
Supported providers and media types
VectorPress is built so that the model provider, the vector database, and the file types it can read are each pluggable. Today one of each is implemented; the rest are on the roadmap. Anything marked TBD below is planned but not yet available, and the admin screens only offer what is marked Supported.
Providers
| Provider | Role | Status | Notes |
|---|---|---|---|
| OpenAI | Model provider | Supported | Embeddings (text-embedding-3-small, text-embedding-3-large, text-embedding-ada-002), chat completions, and vision for image analysis. Bulk indexing uses the OpenAI Batch API. |
| Anthropic (Claude) | Model provider | TBD | Planned. |
| Google (Gemini) | Model provider | TBD | Planned. |
| Pinecone | Vector database | Supported | Serverless indexes are recommended; some maintenance tools (Audit Index) only work on serverless indexes. |
Media types
Posts, pages, and custom post types are always supported. Attachments in the Media Library are indexed by file type:
| Media type | MIME type | Status | Notes |
|---|---|---|---|
| PDF documents | application/pdf | Supported | Text extraction per page, with optional image analysis for tables, charts, and scanned pages. |
| Word documents | .docx | TBD | Planned. |
| Plain text and Markdown | .txt, .md | TBD | Planned. |
| Spreadsheets | .xlsx, .csv | TBD | Planned. |
| Presentations | .pptx | TBD | Planned. |
| Images | .png, .jpg | TBD | Planned. |
The rest of this documentation describes the supported providers and types by name wherever a detail is specific to them (for example OpenAI's 24-hour batch window, or how PDF pages are chunked). Everything else, such as the index settings, the Batch page, and the per-item management panels, is provider- and type-agnostic and will apply to new providers and file types as they are added.
How the pieces fit together
Everything in VectorPress hangs off a configured index:
- A configured index is one vector database index plus the settings that describe what goes into it: which post types and media types, which embedding model, whether only titles or full content are embedded, and whether new content syncs automatically.
- A search engine or a chat agent points at exactly one configured index and adds its own behaviour on top (ranking, result count, prompt, model, and so on).
- A block on a page or post references a search engine or chat agent by ID. Blocks are rendered server-side, so changing an engine or agent later updates every page that uses it without re-saving those pages.
You can have several indexes (for example one per site section, or one for posts and one for documents), and several engines and agents per index.
Before you install
| Requirement | Notes |
|---|---|
| WordPress 6.0 or newer | Single site and multisite are both supported. |
| PHP 8.2 or newer | With the OpenSSL extension (used to encrypt your API keys at rest). cURL is recommended. |
| A model provider API key | Currently OpenAI (see the table above), from an account with billing enabled. Embeddings run through the OpenAI Batch API; chat uses the Chat Completions API. |
| A vector database account and index | Currently Pinecone. VectorPress connects to an index you have already created. It does not create indexes for you. |
| Working WP-Cron | Batch jobs are finished in the background by WP-Cron. If your host disables it, set up a system cron that calls wp-cron.php. See Installation. |
| Outbound HTTPS | To your model provider's API (api.openai.com) and your vector database host (*.pinecone.io). |
The index's dimension must match the embedding model you choose. The setup wizard only offers models compatible with the index you pick, so decide on a model before creating the index at your vector database. See Embedding models and dimensions.
Five-minute setup
- Install and activate the plugin. On a fresh install you are redirected into the setup wizard automatically.
- Providers. Paste your model provider and vector database API keys. Both are verified against the provider before you can continue.
- Destination. Pick the vector database index to write into and a compatible embedding model.
- Content. Choose the post types (and optionally the media types) to index, whether to embed titles only or full content, and whether to keep the index in sync automatically.
- Sync. Review the configuration and click Process Data to start the first batch run, or Save and skip sync to finish setup without indexing yet.
- Add the VectorPress Search or VectorPress AI Chat block to a page, pick the engine or agent you create on the Search or Chat admin page, and publish.
:::note What happens after "Process Data" The first run packages your posts into a bulk embedding job at your model provider (with OpenAI, a Batch API job). Keep the browser tab open while the launch modal is on screen (it is uploading the batch). Once the modal reports the batch was handed off you can leave; WP-Cron polls for completion and writes the vectors into the index. OpenAI batches usually finish in minutes but can take up to 24 hours. Progress is shown on the Batch page. :::
Where things live in wp-admin
After setup, a VectorPress top-level menu appears in wp-admin:
| Menu item | URL | What it is for |
|---|---|---|
| Data | admin.php?page=vectorpress | Create, edit, and delete configured indexes. |
| Batch | admin.php?page=vectorpress-batch | Launch and monitor post and media batch runs; re-ingest previous batches. |
| Chat | admin.php?page=vectorpress-chat | Create and test chat agents; org-wide instructions; chat settings. |
| Search | admin.php?page=vectorpress-search | Create and test search engines. |
| Settings | admin.php?page=vectorpress-settings | Provider keys, encryption key health and backup, hard reset. |
Two more surfaces live outside the menu:
- A VectorPress Management box on the edit screen of every indexed post type, for syncing or removing a single post. See Indexing posts.
- A VectorPress Management panel on each supported attachment in the Media Library, for syncing or removing a single file. See Indexing media files.
Until the wizard is completed, every VectorPress page redirects back into it and the submenu is hidden. If you ever need to restart from scratch, Settings → Danger Zone → Hard Reset wipes the plugin's data and re-arms the wizard.
Next steps
- Installation: packaging, activation, upgrades, and what deactivation and uninstall remove.
- Configuration: every setting on an index, provider keys, and the encryption key.
- Indexing: how posts and media files get into the index and how they are removed.
- Search and Chat agents: the two front-end features.
- API: public endpoints, rate limits, and the filters and constants a site owner can set.
- Troubleshooting: symptom → cause → fix.