You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Display additional AI provider plugins on Connectors page (alongside default Anthropic, Google, and OpenAI ones) #27
Enable developers, agencies, and hosts to pre-configure AI providers that will be automatically recognized by the plugin. This supports a “progressive provider selection” approach (where existing providers are utilized and only presents provider configuration needs if no applicable provider has been configured) and reduces the need for manual setup.
Goals
Define a configuration system (using constants, filters, or configuration files) to allow pre-configured providers.
Ensure the system is modular so each feature can independently detect and use available providers.
When a valid pre-configured provider exists, the UI for provider selection should be hidden.
Document the configuration process with clear examples for hosts and network administrators.
Align with the encapsulated feature architecture so each feature can extend or override provider handling if needed.
Acceptance Criteria
Plugin detects and auto-selects any pre-configured providers.
Provider selection UI is suppressed when configuration is detected.
Features automatically inherit available provider connections.
Developer documentation includes at least one working example.
Punting this to v0.2.0, but in chatting with @dkotter there's a question as to whether this should be in the AIE plugin or perhaps in the WP AI Client; @felixarntz@JasonTheAdams any thoughts/preference there?
So is the end goal that the AI Experiments plugin is a dependency plugin for others plugin introducing a Provider, as to only have a single settings screen for advanced users to select from?
@JasonTheAdams not sure my brain is able to parse your question, but I would say that the AIE plugin would ideally be able to present options (in an advanced settings mode perhaps) for adding in providers via their respective plugins. We would need a way for AIE to "know" the population of those provider plugins, but that otherwise a user could add those additional providers through AIE.
And this is purely for experimentation, right? If a host were to add a Provider is the expectation that they're also integrating with these settings and such?
Codewise, I would put const/filter handling as close to the code it's related to, so provider settings would live in wp-ai-client, settings-layer stuff would live in AI Experiments. Keeps things composable.
And just because the code lives somewhere shouldnt affect how implements consume it.
I think config files are overkill, if a host want's a config file they can make their own to wrap our const/filters.
I'd like to add an Amazon Bedrock connector for review.
AI Chatbot & Agents for Amazon Bedrock
WPORG: https://wordpress.org/plugins/ai-chat-for-amazon-bedrock/
WP minimum: 6.4 for the plugin; the AI Client provider registers on 7.0+, and on Settings > Connectors on 7.1+
Tested up to: 7.1
PHP minimum: 7.4
Source: https://github.com/noteflowai/ai-chat-for-amazon-bedrock
Capabilities: text generation (system instructions, candidates, top P and stop sequences where the model accepts them), vision (Claude, Nova Lite/Pro, Llama 4, Qwen3 VL), JSON output (schema-constrained on models that take one), image generation and editing (Stability AI Stable Image Core, Stable Diffusion 3.5 Large and Stable Image Ultra; editing with SD 3.5), and embeddings (Titan, Cohere) on 7.2. The plugin only declares capabilities the selected model actually has.
Setup and testing: on AWS (EC2/ECS/EKS) the connector shows as connected with nothing entered, signing with the IAM role. Elsewhere, paste an Amazon Bedrock API key under Settings > Connectors > Amazon Bedrock. For image generation, choose an image model under AI Chat Bedrock > Settings > Model. The provider is tested in CI against php-ai-client 1.4.0 and 1.5.0 and the copy bundled with WordPress.
Credentials: none needed with an IAM role. An API key is checked with a free ListFoundationModels call before it is kept, and stored encrypted (authenticated encryption). A key in wp-config.php takes precedence.
Accounts and costs: an AWS account with access to the chosen Bedrock models. Bedrock bills that account per use. The plugin is free and has no paid upgrade.
Known limitations: model availability depends on the AWS region and the account's model access. Embeddings need 7.2.
The plugin also offers a site chatbot, MCP and WooCommerce features, but the connector works on its own through the AI Client. Every core prompt follows the plugin's guardrail, token ceiling, daily limit and usage dashboard.
Update for the Amazon Bedrock connector above: 1.72.1 answers wpai_has_ai_credentials and wpai_has_image_generation_support. Sites that sign with an IAM role (EC2/ECS/EKS), access keys or a key in the plugin settings now count as connected in the AI plugin, not only sites with a key on Settings > Connectors. For image generation and editing, choose an image model under AI Chat Bedrock > Settings > Model; that choice is the site's consent to send prompts to the Stability models in US West (Oregon).
Tested the Amazon Bedrock connector the way you described, with the AI plugin's own image feature:
WordPress 7.1.3 + AI 1.4.0 + AI Chatbot & Agents for Amazon Bedrock 1.72.1, on a site that signs only with an EC2 instance role (no key stored on Settings > Connectors).
has_ai_credentials() and has_image_generation_support() both return true once an image model is chosen under AI Chat Bedrock > Settings > Model.
ai/image-generationgeneration: Stability Stable Image Core returned a 1536×1536 PNG in about 7 s.
ai/image-generationedit, with the generated image passed as reference (data URI): the provider routed it to Stable Diffusion 3.5 Large, the Bedrock model that accepts a reference image, and returned a 1024×1024 PNG in about 10 s, with the requested change applied.
Image requests go to US West (Oregon), where Bedrock offers the Stability models (configurable with a filter). Choosing an image model is the site's opt-in for that.
@dan2k3k4@noteflowai acknowledging both your comments here, but currently testing the initial list of connectors I provided above. I want to get through that testing first, then assess rendering additional connectors and getting that working, then can come back (potentially in a follow-up issue) to re-assess additional connectors to render on that screen (besides the 3 defaults from core).
WPORG: https://wordpress.org/plugins/ai-provider-for-lm-studio/
From the reading in their docs and trying to get image generation set up, it seems like an external plugin is needed to access models like FLUX.2 Klein and that other models like Gemma 3 only appear to have image understanding and not generation. So for now will skip over listing this provider.
AI Provider for Hugging Face
WPORG: https://wordpress.org/plugins/ai-provider-for-hugging-face/
Another case of lots of docs reading and testing but eventually continued to get errors like Bad request for model "black-forest-labs/FLUX.1-dev". The model could not process the request. Model not supported by provider fal-ai when trying to generate an image. So for now will skip over listing this provider.
AI Provider for Mistral
WPORG: https://wordpress.org/plugins/ai-provider-for-mistral/
Sigh, another round of reading docs and getting errors trying to use image gen. Currently blocked at "Unexpected Mistral API response: Missing the "file_id" key.". So for now will skip over listing this provider.
AI Provider for OpenRouter
WPORG: https://wordpress.org/plugins/ai-provider-for-openrouter/
Bah, lots of dead ends on testing today. Gave up after several tests ended with Image generation failed. Please ensure you have a connected provider that supports image generation.. So for now will skip over listing this provider.
AI Provider for OpenAI Compatible Servers
WPORG: https://wordpress.org/plugins/ai-provider-for-openai-compatible-servers/
This one felt like it was going to work with the option in its config to declare it supported images, but alas ended with This feature requires an AI Connector that supports image generation. Review your Connectors to ensure you have a valid AI Connector configured.. So for now will skip over listing this provider.
Vercel AI Gateway Provider
WPORG: https://wordpress.org/plugins/vercel-ai-gateway-provider/
Huzzah, and while not super user-friendly to set up (none of these really are that simple), I was able to successfully generate an image so will proceed with listing this connector.
AI Provider for Osaurus
WPORG: https://wordpress.org/plugins/ai-provider-for-osaurus/
Still finishing up local tests, but wanted to get the rest of this comment published before stepping away for the day. I'll report back shortly on the results for Osaurus.
Ok, hitting a wall with Osaurus as well with This feature requires an AI Connector that supports image generation. Review your Connectors to ensure you have a valid AI Connector configured. so for now will skip over listing this provider.
Activity
jeffpaul commentedon Nov 26, 2025
Punting this to v0.2.0, but in chatting with @dkotter there's a question as to whether this should be in the AIE plugin or perhaps in the WP AI Client; @felixarntz @JasonTheAdams any thoughts/preference there?
JasonTheAdams commentedon Nov 26, 2025
So is the end goal that the AI Experiments plugin is a dependency plugin for others plugin introducing a Provider, as to only have a single settings screen for advanced users to select from?
jeffpaul commentedon Nov 26, 2025
@JasonTheAdams not sure my brain is able to parse your question, but I would say that the AIE plugin would ideally be able to present options (in an advanced settings mode perhaps) for adding in providers via their respective plugins. We would need a way for AIE to "know" the population of those provider plugins, but that otherwise a user could add those additional providers through AIE.
JasonTheAdams commentedon Nov 26, 2025
And this is purely for experimentation, right? If a host were to add a Provider is the expectation that they're also integrating with these settings and such?
justlevine commentedon Dec 11, 2025
Codewise, I would put const/filter handling as close to the code it's related to, so provider settings would live in wp-ai-client, settings-layer stuff would live in AI Experiments. Keeps things composable.
And just because the code lives somewhere shouldnt affect how implements consume it.
I think config files are overkill, if a host want's a config file they can make their own to wrap our const/filters.
55 remaining items
noteflowai commentedon Oct 11, 2026
I'd like to add an Amazon Bedrock connector for review.
AI Chatbot & Agents for Amazon Bedrock
WPORG: https://wordpress.org/plugins/ai-chat-for-amazon-bedrock/
WP minimum: 6.4 for the plugin; the AI Client provider registers on 7.0+, and on Settings > Connectors on 7.1+
Tested up to: 7.1
PHP minimum: 7.4
Source: https://github.com/noteflowai/ai-chat-for-amazon-bedrock
Capabilities: text generation (system instructions, candidates, top P and stop sequences where the model accepts them), vision (Claude, Nova Lite/Pro, Llama 4, Qwen3 VL), JSON output (schema-constrained on models that take one), image generation and editing (Stability AI Stable Image Core, Stable Diffusion 3.5 Large and Stable Image Ultra; editing with SD 3.5), and embeddings (Titan, Cohere) on 7.2. The plugin only declares capabilities the selected model actually has.
ListFoundationModelscall before it is kept, and stored encrypted (authenticated encryption). A key inwp-config.phptakes precedence.The plugin also offers a site chatbot, MCP and WooCommerce features, but the connector works on its own through the AI Client. Every core prompt follows the plugin's guardrail, token ceiling, daily limit and usage dashboard.
noteflowai commentedon Oct 11, 2026
Update for the Amazon Bedrock connector above: 1.72.1 answers
wpai_has_ai_credentialsandwpai_has_image_generation_support. Sites that sign with an IAM role (EC2/ECS/EKS), access keys or a key in the plugin settings now count as connected in the AI plugin, not only sites with a key on Settings > Connectors. For image generation and editing, choose an image model under AI Chat Bedrock > Settings > Model; that choice is the site's consent to send prompts to the Stability models in US West (Oregon).noteflowai commentedon Oct 11, 2026
Tested the Amazon Bedrock connector the way you described, with the AI plugin's own image feature:
has_ai_credentials()andhas_image_generation_support()both return true once an image model is chosen under AI Chat Bedrock > Settings > Model.ai/image-generationgeneration: Stability Stable Image Core returned a 1536×1536 PNG in about 7 s.ai/image-generationedit, with the generated image passed asreference(data URI): the provider routed it to Stable Diffusion 3.5 Large, the Bedrock model that accepts a reference image, and returned a 1024×1024 PNG in about 10 s, with the requested change applied.Image requests go to US West (Oregon), where Bedrock offers the Stability models (configurable with a filter). Choosing an image model is the site's opt-in for that.
jeffpaul commentedon Oct 15, 2026
@dan2k3k4 @noteflowai acknowledging both your comments here, but currently testing the initial list of connectors I provided above. I want to get through that testing first, then assess rendering additional connectors and getting that working, then can come back (potentially in a follow-up issue) to re-assess additional connectors to render on that screen (besides the 3 defaults from core).
jeffpaul commentedon Oct 15, 2026
Research/reading on these providers plus some testing on installing each to see if image gen works seems to show the following.
AI Provider for Ollama
WPORG: https://wordpress.org/plugins/ai-provider-for-ollama/
Even though models like https://ollama.com/x/flux2-klein and https://ollama.com/x/z-image-turbo exist and appear to report as
imagecapabilities I continue to get anBad Request (400) - image generation models are not currently supportederror. So for now will skip over listing this provider.AI Provider for LM Studio
WPORG: https://wordpress.org/plugins/ai-provider-for-lm-studio/
From the reading in their docs and trying to get image generation set up, it seems like an external plugin is needed to access models like FLUX.2 Klein and that other models like Gemma 3 only appear to have image understanding and not generation. So for now will skip over listing this provider.
AI Provider for Hugging Face
WPORG: https://wordpress.org/plugins/ai-provider-for-hugging-face/
Another case of lots of docs reading and testing but eventually continued to get errors like
Bad request for model "black-forest-labs/FLUX.1-dev". The model could not process the request. Model not supported by provider fal-aiwhen trying to generate an image. So for now will skip over listing this provider.AI Provider for Mistral
WPORG: https://wordpress.org/plugins/ai-provider-for-mistral/
Sigh, another round of reading docs and getting errors trying to use image gen. Currently blocked at "Unexpected Mistral API response: Missing the "file_id" key.". So for now will skip over listing this provider.
AI Provider for OpenRouter
WPORG: https://wordpress.org/plugins/ai-provider-for-openrouter/
Bah, lots of dead ends on testing today. Gave up after several tests ended with
Image generation failed. Please ensure you have a connected provider that supports image generation.. So for now will skip over listing this provider.AI Provider for OpenAI Compatible Servers
WPORG: https://wordpress.org/plugins/ai-provider-for-openai-compatible-servers/
This one felt like it was going to work with the option in its config to declare it supported images, but alas ended with
This feature requires an AI Connector that supports image generation. Review your Connectors to ensure you have a valid AI Connector configured.. So for now will skip over listing this provider.Vercel AI Gateway Provider
WPORG: https://wordpress.org/plugins/vercel-ai-gateway-provider/
Huzzah, and while not super user-friendly to set up (none of these really are that simple), I was able to successfully generate an image so will proceed with listing this connector.
AI Provider for Osaurus
WPORG: https://wordpress.org/plugins/ai-provider-for-osaurus/
Still finishing up local tests, but wanted to get the rest of this comment published before stepping away for the day. I'll report back shortly on the results for Osaurus.
jeffpaul commentedon Oct 15, 2026
Ok, hitting a wall with Osaurus as well with
This feature requires an AI Connector that supports image generation. Review your Connectors to ensure you have a valid AI Connector configured.so for now will skip over listing this provider.saarnilauri commentedon Oct 16, 2026
Thanks @jeffpaul I will check on the Mistral Provider soonish.
saarnilauri commentedon Oct 20, 2026
@jeffpaul Mistral Provider image generation is now fixed in 1.2.1 release of the plugin.
Mistral API change broke the way the image generation used to work. This was bit tricky to discover as some of the changes were undocumented.
Feature reviewed Connector plugins on the Connectors screen