---
title: Limitations And Roadmap
description: "## Current Provider Scope Text feeds use Anthropic or OpenAI. On Premium and higher they can also use any service that speaks the OpenAI Chat Completions format, through an OpenAI-compatible provider profile. Image feeds use OpenAI only. A custom model…"
url: "https://brightleafdigital.io/docs/limitations-and-roadmap/"
updated: "2026-11-18"
category: GravityOps AI Feed / Changelog Roadmap
---

# Limitations And Roadmap

## Current Provider Scope

Text feeds use Anthropic or OpenAI. On Premium and higher they can also use any service that speaks the OpenAI Chat Completions format, through an OpenAI-compatible provider profile. Image feeds use OpenAI only. A custom model ID selects a model within the chosen provider's request path; it does not change provider or API protocol.

OpenAI-compatible services vary. The plugin requests only the capabilities enabled on the profile, offers no effort, thinking or speed controls for them, and has no reviewed model list, lifecycle notices or output limits for their models. Their prices are not known to the plugin.

A fallback model is tried once, only when the first provider is unavailable. The plugin does not retry the same provider.

The bundled model catalog is intentionally static for a release. Provider model availability can change before the plugin catalog is updated. Unknown/custom models receive less compatibility and output-limit validation.

## Frontend Generation

Frontend text, streaming, and images require Premium or Agency. A frontend feed is frontend-only; it does not also execute automatically after submission. Streamed text uses Server-Sent Events and can be buffered by hosting or proxy infrastructure.

Frontend form serialization excludes file inputs, passwords, nonces, actions, and plugin-internal controls. Frontend image uploads use a separate verified temporary-session path.

## Structured Output

Native structured output is used only when the catalog says the selected text model supports it. Other compatible models receive a prompt-only JSON instruction. The contract is derived from the active destination adapters, so it can include numbers, constrained choices, arrays, dates, and compound objects rather than representing every property as a string.

Text, Paragraph, and Hidden destinations require Pro. Premium adds the documented Email, Website, Number, Date, Time, supported Phone, choice, Name, Address, List, and Entry Tags configurations. Consent, credential, payment/pricing, upload/media, calculated, relationship, identifier, workflow, and structural fields remain intentionally unsupported. Third-party field support requires an explicit adapter and is never inferred from a similar-looking input.

Each active property is requested, but a missing or invalid property does not discard other valid writes. Such a run is reported as partial. Fill-empty execution can remove populated properties from the active contract or skip the provider entirely when nothing remains.

## Images

Rough previews while an image is generated are optional and off by default; they are not available for Animated GIF or SVG output. Custom arbitrary dimensions are restricted to the models reviewed as accepting them: GPT Image 2.5 Flare, GPT Image 2.5 Sunburst, and GPT Image 2.

GPT Image 2.6 Flare and Sunburst expose planning Complexity and Animated GIF output. GIFs use the same preview and verified attachment paths as still images. Frame count and duration are validated locally; OpenAI enforces its combined animation-size budget and GIF preview access requirements. SVG output is offered only on GPT Image 2.6 Flare and Sunburst and needs SVG access on the OpenAI project. Generated SVG is accepted only when it is static drawing content; anything else is rejected rather than repaired. The model that writes the SVG can be named on the feed, but the plugin cannot validate it because OpenAI does not publish the accepted IDs.

Image editing runs only for saved entries: after submission, in a Gravity Flow step, or as an administrator rerun. It is not available from a frontend button. One source image is sent, with an optional markup overlay; masks and multiple reference images are not supported. The form has no drawing tool, so an overlay is uploaded as a file. Image feeds use the default OpenAI key or a selected OpenAI profile's key, but not a profile's custom base URL.

## Usage And Budgets

Plugin monthly token caps use locally recorded input/output usage. Requests to an OpenAI-compatible provider that does not report usage are not counted. They are not provider invoices, do not cap image billing, and do not replace provider project budgets or alerts.

## Planned Areas

- **Image editing from a frontend button:** let a visitor edit an image before submitting, which needs the source file to be uploaded with the button request.
- **A drawing tool for markup overlays:** an in-form editor for circling or marking the area to change. This is a separate UX decision.

Potential future work includes additional provider and field adapters, refreshed model catalogs, richer usage reporting, provider-supported image progress, and more granular operational diagnostics. Roadmap items are not commitments and should not be used as a dependency for a live workflow until released.

## Stuck? Ask in GravityOps Help

Search past answers or post your question, tagged "ai-feed".

- [GravityOps AI Feed: a feed did not run on submission](https://brightleafdigital.io/community/gravityops-help/4147/)GravityOps Help - 0 replies - 2 weeks ago
- [GravityOps AI Feed is live: Claude and OpenAI as Gravity Forms feeds](https://brightleafdigital.io/community/announcements/3931/)Announcements - 0 replies - 4 weeks ago

[Ask in GravityOps Help](https://brightleafdigital.io/community/gravityops-help/?gofrm_tag=ai-feed)[Open a support request](https://brightleafdigital.io/support/)

GravityOps Weekly

[This iframe contains the logic required to handle Ajax powered Gravity Forms.](about:blank)
