# Getting metadata about intent/entity in response

**URL:** <https://forum.rasa.com/t/getting-metadata-about-intent-entity-in-response/25338>\
**Category:** Rasa Open Source\
**Created:** [February 26, 2020, 10:57pm UTC](https://forum.rasa.com/t/getting-metadata-about-intent-entity-in-response/25338 "2020-02-26T22:57:37Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![dcsan](https://dub1.discourse-cdn.com/flex013/user_avatar/forum.rasa.com/dcsan/32/1048_2.png) [@dcsan](https://forum.rasa.com/u/dcsan)\
**Post date:** [February 26, 2020, 10:57pm UTC](https://forum.rasa.com/t/getting-metadata-about-intent-entity-in-response/25338/1 "2020-02-26T22:57:37Z")

</div>

I’m working on debugging my chats with Rasa, and building a simple web client to do this.

When using the `rasa interactive` shell there’s a lot of metadata in the console and ways to correct the chat. Is there a way to get similar data when just “chatting” to the bot via an API? Then I could build a web debug client.

Even if it’s not fully interactive for on-the-fly training, it would be great if the normal `rasa server` would also send some metadata about which intent and entities were recognized.

I guess I could use the `http api` and the like [parseModelMessage](https://rasa.com/docs/rasa/api/http-api/#operation/parseModelMessage) but I’m not sure if that keeps things like the current conversation state, so the returned intents would not be in context.

Otherwise perhaps I can use a python action to somehow send back the state at the `lastMessage`?

It looks like there are some other http endpoints, but that would mean sending the query twice, once over sockets and once on http to “double predict” it (and that would change the state too)

```auto
/conversations/<conversation_id>/predict 

```

Any other thoughts on how to do this?

---

<div class="post-metadata">

**Author:** ![dcsan](https://dub1.discourse-cdn.com/flex013/user_avatar/forum.rasa.com/dcsan/32/1048_2.png) [@dcsan](https://forum.rasa.com/u/dcsan)\
**Post date:** [February 29, 2020, 12:49am UTC](https://forum.rasa.com/t/getting-metadata-about-intent-entity-in-response/25338/2 "2020-02-29T00:49:53Z")

</div>

Also discussing on BotFront repo in case people haven’t seen that neat tool for authoring rasa bots

> <https://github.com/botfront/botfront/issues/497>
>
> \*\*Is your feature request related to a problem? Please describe.\*\*
> When chattin…g to your bot, there is no info given on what intents are matching, and how to display that to the user.
> 
> \*\*Describe the solution you'd like\*\*
> It would be great to show more info, somewhat like rasa-X
> 
> \*\*Describe alternatives you've considered\*\*
> Currently the chat side-panel web version doesn't have much info
> When a chat isn't predictable I have to drop down to \`rasa interactive\` or \`rasa shell --debug\`
> 
> The normal rasa server doesn't send much debug info 
> I was thinking we could also send a \`rasa nlu\` type of request or just \`/predict\` to the web-servers http API to get more information on the handling of that input, but I'm not sure how that would track the conversation state.
> 
> https://forum.rasa.com/t/getting-metadata-about-intent-entity-in-response/25338
> 
> \*\*Additional context\*\*
> RASAs own visualizer is very helpful in debugging chats
> having a chat at the right while the visualizer is running 
> 
> https://legacy-docs.rasa.com/docs/core/interactive\_learning/
> 
> !\[interactive\_learning\_graph\](https://user-images.githubusercontent.com/514002/75597144-9b6d5400-5a48-11ea-9ab8-08b982d3c9e1.gif)
> 
> the vis runs on :5006 (not :5005 as written in the docs)
> http://localhost:5006/visualization.html
> 
> I'm not sure if rasa-x is using some other APIs to debug current bot state?
