Turn your triage criteria into Jira actions.
Choose the Jira fields to read, describe your possible outcomes, and assign actions to each one. Jev classifies the issue; your configuration determines what happens next.
Currently in private testing. Not yet available on the Atlassian Marketplace.
Interactive example — simulated results, no Jira connection
First-line support triage
Issue
Prompt
{{issue_description}}
Outcomes & actions
Add label first-line · Set title from a support template
Add label specialist
Issue
Prompt
{{issue_description}}
Outcomes & actions
Add label billing
Add label checkout
Add label identity
Add label platform
First-line support triage — Yes
Added label first-line. Set the title from the configured support template.
Product area routing — Billing
Added label billing. Billing was the single selected outcome.
Backend title — Yes
The title expanded to 260 characters; the limit is 255. Classification retained. No actions applied.
Specialist routing — Yes
The Jira write could not be confirmed. Review the issue before taking further action; the write was not automatically repeated.
Four illustrative entries, not customer data. In the app, activity is visible for 14 days.
No API key to supply
We cover the Jev processing costs. Your site does not need its own AI account or API key.
Your outcomes. Your actions.
Jev selects from your configured outcomes. The app validates the response and applies only the actions attached to those outcomes. Test your criteria before enabling a rule.
No second tool to log into
Manage project automations and review activity inside Jira. Configuration and logs are stored in Forge; AI classification uses OpenRouter and TypeSafe.
Make your triage criteria explicit.
Does this request fit first-line support? Does this bug need backend work? Which product area does it concern? Turn those decisions into named outcomes and configured Jira actions.
The inconsistency
Put classification criteria in a shared rule that project administrators can review and refine.
The queue
Run enabled rules when issues are created or relevant fields change, using the triggers you choose.
The criteria
Describe what an outcome means in plain language, with separate criteria for each possible choice.
The record
Review classifications, model scores and action results together in Activity log, without storing a full copy of the issue body.
Describe the decision. Assign the actions. Enable the rule.
Pick a project and your inputs
Choose a project you administer. Insert Description, Summary, Labels, Priority or supported custom-field variables into the prompt. The app shows which Jira values the classification will use.
Write the classification prompt
A yes/no decision, a single choice among named outcomes, or multiple independent choices. Describe each possible result so the model understands what it means.
Assign actions per outcome
Add a label, set the title, replace the description, update a supported custom field, or move to a supported destination project and issue type. An outcome can also have no actions.
Test, then enable
Test with a real issue without editing it. Tests still send the configured input to the model. Then enable the rule for issue creation, relevant updates, or both.
Use Jira values in prompts and action templates.
Insert supported Jira variables from a picker. They work in prompts, outcome descriptions and text action values, including ticket titles. The app supplies the Jira bindings and rejects unknown or unavailable variables when you save.
{{issue_description}}{{issue_summary}}{{issue_labels}}{{issue_priority}}{{issue_key}}{{project_key}}{{customfield_12345}}Five configurable action types
Action text comes from your templates. Custom field updates support text, numbers and Jira option selections. Field availability, value types and permissions are checked before execution.
Your site controls AI processing.
Explicit authorization
A site administrator must authorize external AI processing before classifications or tests can run. Disabling it stops new model requests and new actions; it cannot recall data already sent or cancel an accepted move.
App edits, administrator permissions
Background reads use the administrator who last saved the rule. New edits use the app account. Before each edit, the app rechecks that administrator's issue access and permissions; Jira also enforces the app's access.
Know what leaves Jira
The prompt, outcome criteria and referenced Jira values are sent to OpenRouter and TypeSafe. Values used only in action templates are expanded locally. Activity access also requires permission to view the issue.
14 days of activity
Activity is hidden after 14 days and configured to expire in Forge storage. Execution receipts expire 14 days after the original run was created; updates do not extend retention. Physical cleanup can lag; provider-side retention follows separate policies.
Uncertain writes require review. The app uses event receipts and self-trigger guards to reduce duplicate execution. When a Jira write cannot be confirmed, it marks the run Needs review instead of automatically repeating that write.
AI processing takes place outside Atlassian. This app does not qualify for the Runs on Atlassian badge. It does not enforce provider zero retention, no training, or a fixed processing region.
See the classification and the action result.
The log shows the issue, rule, time, outcomes, model probabilities and action results or errors. It also records failed dry tests. The scores come from the model; they do not guarantee that a classification is correct.
What a run keeps
Issue identifiers, rule and outcome labels, model scores, action statuses and sanitized errors. Full issue bodies, expanded action values and provider credentials are excluded.
What a failure keeps
A valid classification is retained when an action fails. Planning errors, such as an overlong title, identify failed and skipped actions. Later Jira errors can leave a run partially applied.
Who can read it
Project administrators for their projects, and Jira site administrators. Entries are shown only when the viewer can access the referenced issue.
How long
Activity is visible for 14 days. Rules remain until deleted; the activity retention period does not apply to all configuration or external providers.
Built for project administrators.
The questions admins ask first.
Is Jev for Jira available on the Marketplace?+
Not yet. The app is implemented and in private testing on a Jira Cloud site. App Terms and the Privacy Policy are available on this website. Public distribution still requires a production release, a Marketplace listing, data processing terms and Atlassian approval.
Which model does it use, and do I need an API key?+
It uses TypeSafe's typesafe/jev-1.13 through OpenRouter. We pay for Jev usage and manage the backend credential. Customers do not need their own OpenRouter account or API key.
Can the model create new actions or write the replacement text?+
The model selects configured outcomes; it does not author the action plan or replacement prose. Titles and field values come from your templates. A classification can still be wrong, so test representative issues before enabling actions.
Who can create automations?+
Project administrators can manage rules for their projects. Jira site administrators can manage projects and authorize external AI processing. Activity entries also require access to the referenced issue.
What happens after installation?+
A site administrator must authorize external processing before model requests are allowed. Rules start disabled. An administrator can test a rule, then enable it for issue creation, relevant updates, or both.
Which Jira fields can I use?+
Insert Description, Summary, Labels, Priority and supported custom-field variables into your prompt or outcome descriptions. These variables determine the classification input. Custom-field updates depend on Jira field context and allowed values.
Can I use variables in ticket titles?+
Yes. The title action has a variable picker and supports templates such as [Support] {{issue_summary}}. The 255-character limit is checked after expansion. An invalid title stops the action plan and is recorded in Activity log.
What happens if the prompt or issue is too large?+
Prompt templates allow 8,000 characters; outcome descriptions allow 2,000 each. Jev has a 32,000-token context. The app separately caps the expanded request at 30,000 UTF-8 bytes. Oversized input fails without Jira actions or truncation.
Who appears as the author in Jira history?+
New edits use the app account. Reads and permission checks use the administrator who last saved the rule. The app also needs Jira access. Existing history and already-submitted older moves keep their original author.
Can it move issues between projects?+
Supported moves use a configured destination and issue type, subject to Jira permissions and restrictions. Accepted moves are tracked asynchronously. Unsupported cases or failures are reported; acceptance is not reported as completion.
Does issue data stay inside Atlassian?+
No. The prompt, outcome criteria and referenced Jira values are sent to OpenRouter and TypeSafe after site-admin consent. The app does not enforce provider zero retention, no training, or a fixed processing region.
What is stored, and for how long?+
Activity and execution receipts expire 14 days after the run was created. Progress updates do not extend this period; physical cleanup in Forge can lag. Rules remain until deleted. Full issue bodies, expanded action values and credentials are excluded from activity.
Does it search for duplicates or generate answers?+
No. The current app classifies the configured issue input. It does not search other Jira issues for duplicates, generate free-form replies, or independently choose additional tools and actions.
Is the app available in other languages?+
The app interface and documentation are currently English only.
Jev for Jira is in private testing.
Classification automations and an Activity log are implemented and running on our test site. Public Marketplace distribution is still being prepared.
The public listing, data processing terms, production release and Atlassian approval are still required. Read the App Terms and Privacy Policy. This page is a product preview, not an installation link.