# Epic AA-231 - Agents in OpenProject: Triage Bot

Project: Stream Agents & Assistants
Status: new
URL: https://community.openproject.org/work_packages/AA-231

## Detailed specification

### Motivation and background information

### Acceptance criteria

### Out of scope

| Attribute | Value |
| --- | --- |
| Mockups |  |

## Planning

| Attribute | Value |
| --- | --- |
| Roadmap | 2027 Q2 |
| Date |  |
| List |  |
| Target versions | 18.5.0 |
| Relevance |  |

## Subitems

_No work packages_

## Bugs and other work packages

_No work packages_

## Details

| Attribute | Value |
| --- | --- |
| Module | Artificial Intelligence (AI) |
| Enterprise plan | Enterprise / Professional (PRO) |
| Labels |  |

## People

| Attribute | Value |
| --- | --- |
| Assignee |  |
| Accountable |  |
| Product manager | Dominic Bräunlein |
| Feature Lead |  |
| Designer |  |
| Developers | OP Dev AI team |
| Interest by |  |

## Estimates and progress

| Attribute | Value |
| --- | --- |
| Work |  |

## Description

# User Problem

## User

*   Product owners and project managers who triage incoming work packages in a project
    

## Problem

*   Every new work package needs a type, priority, assignee, potentially a specific custom fields set and a duplicate check before anyone can work on it.
    
*   The knowledge behind these decisions (who handles which area, which earlier report covers the same issue) sits with a few people.
    
*   OpenProject has no way to run an AI agent inside a project that helps with this
    

## Pain

*   The triager opens each new work package, searches for similar ones by hand and sets the fields one by one.
    
*   When the triager is absent, new work packages stay unassigned in status New.
    

# Business Case

## Solution

A generic agent runtime with Triage Bot as the first built-in preset. The first version reuses existing OpenProject concepts.

*   Agent as a user: each agent is a special user type and a project member with a role. The role limits what it can change. Every change shows in the activity journal under the agent&#x27;s name.
    
*   Owner: every agent has a human owner. The agent never gets more permissions than its owner. The owner is the default reviewer and gets notified on failures.
    
*   Suggestion card: A new component on the work package that lists proposed changes, each with a short reason. The user ticks the changes and clicks &quot;Apply selected&quot;. Applying runs a normal work package update.
    
*   Filter &quot;Has open suggestions&quot;, so saved views such as a Triage view (status New, no assignee) show what waits for review.
    
*   Manual trigger: In the work package detail view there is a trigger to start the Bot. In the work package list view there is a bulk action for all or selected rows in lists.
    
*   There is a settings page per project: enabled, role, owner, fields it may suggest, guidance text and recent runs.
    
*   Run log per agent with input, reasoning and the resulting changes.
    
*   Everything is a suggestion. Nothing is applied without a person.
    

## Out of Scope for the MVC

1.  Auto-apply the suggested changes
    
2.  @mention and assignment of work packages to agents
    
3.  Further presets: e.g. work package refresher
    
4.  Custom action or automations that runs an agent
    

## Next iteration

*   Automatic trigger: run Triage Bot when a work package is created automatically
