Builds By Luke
All Atlassian apps

Jira · Confluence · v1.0.0 · Planned

ModelDesk

bring-your-own-model assistant

Summarise and triage issues with your own model provider and your own API key — including self-hosted.

Lowest-priority build. Native platform AI could close this gap at any time — the reason to build it is bring-your-own-key and self-hosting, which the platform does not offer.

  • Your key, your provider
  • Self-hosted option
  • Per-action, not per-seat
  • Nothing retained
ModelDesk icon

The gap it closes

The platform’s own assistant is capable, and it is licensed per seat and bound to one model. Teams who already pay for a model provider — or who are required to keep inference inside their own boundary — are paying twice, or cannot use it at all.

ModelDesk does the opposite: a short list of specific jobs, with the provider and key chosen by you.

Being honest about the risk: this is the one app in this portfolio the platform vendor might make redundant, which is exactly why it is scoped small and priced low. Its lasting advantage is that your key and your self-hosted endpoint remain options.

How it works

Three steps, then it runs itself.

  1. 01

    Bring your provider and key

    Configure the endpoint and API key for the provider you already use — or the address of a model you host yourself.

  2. 02

    Run a specific action

    Summarise a long comment thread, suggest a category or priority, or draft a first reply. Each is a deliberate action, never an automatic background process.

  3. 03

    You keep the result or discard it

    Output is shown for review before anything is written to the issue. Nothing is applied silently.

How it behaves

The decisions that matter.

  • Nothing is sent without an explicit action

    There is no background analysis. Content only reaches a model when someone presses the button for a specific job.

  • The destination is yours and it is declared

    Requests go to the endpoint you configure, with your key. The permitted hostnames are in the app manifest and shown to your admin at install time.

  • Output is proposed, never auto-applied

    Generated text is presented for review. An assistant that edits your tracker unattended is a liability, not a feature.

  • No prompts or responses are retained

    Content is sent, the answer is displayed, and neither is stored by the app.

Where it runs, and what it reaches

Every outbound connection, named.

The app runs on Atlassian’s own infrastructure. Where it must reach outside, these are the only hostnames it is permitted to contact — the same list Atlassian shows your admin before you install, and the same list a Marketplace reviewer checks against the manifest.

  • your-model-endpoint The provider or self-hosted endpoint you configure, contacted with your key, only when you run an action. No other host is reachable.

Runs on

Atlassian Forge (runs on Atlassian’s own infrastructure)

Your credentials live in

Forge encrypted secret storage, inside your own installation

Surfaces it adds

Jira admin page, Issue panel, Confluence content action

Pricing

Priced per user, billed by Atlassian.

Every paid plan starts with a free 30-day evaluation, arranged by Atlassian. Atlassian collects the payment, handles invoicing and tax, and is the merchant of record — so this app appears on the Atlassian bill you already pay, not a separate one from me. Sites of ten users or fewer pay one flat monthly rate.

Evaluation

Free

30 days, full functionality

Up to 10 users

$12/month

Flat rate for the whole site

11–100 users

$1.20/user/mo

Billed by Atlassian

101+ users

$0.85/user/mo

Volume rate

Indicative pricing for an app that is not listed yet. The rate on the Marketplace listing is the one that applies.

Documentation & data handling

Everything it touches, stated plainly.

Installing & getting started

  1. Install ModelDesk from the Atlassian Marketplace into your Jira site and review the permissions listed here.
  2. Open Settings → Apps → ModelDesk and work through the three steps above.
  3. Your evaluation runs for 30 days with everything switched on, so it can prove itself before you pay for it.
  4. Stuck? Email [email protected] — one business day response.

Permissions, and why each one

read:jira-work
Read the issue or thread you asked to summarise.
write:jira-work
Apply a result — only after you approve it.
storage:app
Hold the endpoint configuration and the encrypted API key inside your own installation.

What is stored, and where

  • Your API key is stored in Forge encrypted secret storage inside your own installation and is never returned to the browser.
  • Issue or page content is transmitted to your configured endpoint only when you explicitly run an action, and only the content needed for that action.
  • Prompts and responses are not stored by the app. What the provider you chose does with them is governed by your agreement with them — which is a reason to choose one you already trust.
  • Uninstalling removes the stored configuration and key.

Full policy: Privacy · Terms

Technical detail

Version
1.0.0
Platform
Atlassian Forge
Runtime
Node.js 22 — Forge runtime
Products
Jira, Confluence
Modules
Jira admin page, Issue panel, Confluence content action
External connections
1 declared host
Billing
Paid through the Atlassian Marketplace
Support SLA
One business day
Publisher
Builds By Luke, Canada

Questions people ask.

Whose model is used?

Yours. You configure the provider and key, including a self-hosted endpoint if you run your own.

Are my issues sent anywhere automatically?

No. Content is only sent when someone runs a specific action, and the app never analyses in the background.

Does it store prompts or responses?

No. The result is displayed and, if you approve it, written to the issue. Nothing is retained by the app.

ModelDesk is on its way.

It is on the build list for the Atlassian Marketplace. If you want it sooner — or want it to do one specific thing — that is genuinely useful to hear.