Feature request: Ability to modify assistant responses on the client side #159
Description
We are using ChatKit with the OpenAI-hosted backend (Agent Builder workflows):
https://platform.openai.com/docs/guides/chatkit
In a project we are delivering for a client, we are building an AI assistant that provides links to relevant content and products. In practice, the model occasionally omits links despite system instructions, or produces incorrect/hallucinated ones. We can reliably fix this on the client side by detecting patterns in the response text (e.g., product codes, names, or placeholders) and inserting or correcting links using our own trusted mapping (e.g., regex + catalog lookup).
We would welcome a supported client-side mechanism that allows developers to intercept and adjust the assistant’s output text either during streaming or after the response is complete while still using the managed, OpenAI-hosted ChatKit backend.
Having this capability would make ChatKit more flexible for real-world production use cases that require deterministic post-processing of model output without using a self-hosted backend, where this would likely be possible, but at the cost of losing the benefits of the managed, OpenAI-hosted solution.