Discourse AI - AI triage using Persona

[Saif (Saif Murtaza)]

AI triage uses the power of AI Personas to respond to topics intelligently.

Please note that this feature requires both the discourse-automation and discourse-ai plugins to function.

Use cases

1. Automatic support: Generate intelligent responses to people’s support questions. For example, in a Support category, the persona can use community content for context. A bonus is to have the answers generated through whispers so a responder can first use the suggested answers before replying

2. Category expert: Have a designated persona expert for each category in your community. Respond to the first post of a topic, with human follow-up if needed

3. Subtle moderation: Nudging users to post community-appropriate content in the right places. For example, if a topic seems like a mismatch with the category, instead of making that change automatically, replying in the topic suggesting other categories to alert the poster

Due to the Personas being fully customizable, a community can leverage this feature in many other ways.

How it works

AI triage using Persona can unconditionally responds to topics and posts by using personas to understand context and drive intelligent responses.

The following is an example automation rule and an associated persona. You must have a persona created before using this feature.

Configuration

To configure this feature, you need to specify certain parameters:

• Persona: AI Persona to use for triage. Must have default LLM and user set

• Reply as Whisper: Whether the persona’s response should appear as whisper

• Silent Mode: If selected the persona will receive the content but will not post anything in the forum. An admin must create Custom tools before using this option. This is used in conjunction with the Tools field in the Persona settings. This can be useful for…

  • @sam

    • Make an HTTP request to a company internal API informing that something just happened.
    • Perform side effects on Discourse such as create a chat message, or flag a post

Caveats

• LLM calls can be expensive. Be careful to monitor costs and always consider running on small subsets
• Be mindful that LLMs can hallucinate, so make sure to check any output carefully

Last edited by @Saif 2025-03-20T16:54:44Z

Check document

Perform check on document:

[thoka (Thomas Kalka)]

Awesome.

Could this also be used to write summaries for incoming mails, responding with a new topic in a different category?

[sam (Sam Saffron)]

@thoka

Could this also be used to write summaries for incoming mails, responding with a new topic in a different category?

Not this feature, but we have a tool based automation that could do that with custom js , it is not quite ready yet

[sam (Sam Saffron)]

Also worth noting, llm triage can also respond using a persona now, this means you can support custom replies based on triage

This automation unconditionally respond. Triage will conditionally respond

[sam (Sam Saffron)]

Note @Saif we need to update the OP to explain “silent” option.

If you select “silent” no post is created, and no information is left behind.

This can sound extremely odd, but if you have a forced tool, it can perform side effects like

• Make an HTTP request to a company internal API informing that something just happened.
• Perform side effects on Discourse such as create a chat message, or flag a post

Using silent requires you also create a custom tool to perform the special action you want to perform.

Can you incorporate this into the OP (needs new screenshots etc)? Feel free to ask for clarifications.

[Saif (Saif Murtaza)]

@sam

Using silent requires you also create a custom tool to perform the special action you want to perform.

Can you incorporate this into the OP (needs new screenshots etc)? Feel free to ask for clarifications.

A few questions

• Do you have to exclusively use silent mode? Can I have reply as whisper selected? Im assuming it doesn’t matter as there is no reply (from a UI perspective we should probably disable certain options if it doesn’t depend on it)

• Does it need to be only Forced tools? What about just regular tool calls from the persona? What if the persona has no Forced tools?

• I think we should put a disclaimer for this option given the dependency on Custom tools and having forced tools selected via the Persona

• Going back to the use cases I think this changes the name of this feature from AI auto responder back to AI triage via Persona since this option doesn’t have “auto-responses” that are visible or invisible (whispers)

[sam (Sam Saffron)]

@Saif

Do you have to exclusively use silent mode? Can I have reply as whisper selected? Im assuming it doesn’t matter as there is no reply (from a UI perspective we should probably disable certain options if it doesn’t depend on it)

Yes it is exclusive at the moment. Which is a good point, I will see if I can make this a choice selector so it is less confusing.

@Saif

Does it need to be only Forced tools? What about just regular tool calls from the persona? What if the persona has no Forced tools?

Forced tools give you guarantees that tool will be called, but it can be used fine without forced tools.

@Saif

I think we should put a disclaimer for this option given the dependency on Custom tools and having forced tools selected via the Persona

Yeah it is a very difficult to explain and very powerful setting.

@Saif

Going back to the use cases I think this changes the name of this feature from AI auto responder back to AI triage via Persona since this option doesn’t have “auto-responses” that are visible or invisible (whispers)

That is a good point.