FAQ

The questions people ask first, grouped by topic. If yours is not here, ask your administrator, or write to us through the contact page.

#The basics

#What is Jarvis?

A delivery workspace for clients, projects and tickets, where AI agents can take a ticket, make the change on a branch of its own, and wait for a person to approve it. It also has a voice screen with the same name. See the Overview.

#How do I get an account?

Your workspace is set up and run by AiM Infosoft, and accounts are created by your administrator. There is no public sign-up.

#Does an agent ever deploy on its own?

No. Every piece of agent work stops at the approval gate, and nothing is deployed without an approval. See Runs and the approval gate.

#Which AI writes the code?

Claude Code or Codex, depending on the agent you choose. Each agent is set to a model, and the model decides which. See AI coding agents.

#Can an agent change our live code directly?

No. Agents work on a branch of their own for each ticket. Approved work arrives as a pull request your team can review like any other.

#The approach

#Do I have to write a goal, notes and outcomes before using agents?

No. A ticket with a good description and acceptance criteria can be worked on its own. But agents read the goal, the notes and the ticket's outcome every time, so the more of that is written, the less each one has to guess. See How we deliver.

#What is the difference between an outcome and a ticket?

An outcome is something the system can do that stays true after the work is finished: "a returning customer can pay with a saved card". A ticket is a piece of work that helps make it true: "add the saved cards table". See Outcomes and quality checks.

#Why can I not mark an outcome Ready?

All six quality checks have to be ticked first, and the same goes for In progress and Delivered. If a check does not honestly pass, rewrite the outcome until it does.

#Does Review with AI tick the checks for me?

No. It gives a pass or a concern for each check with a reason, and lists gaps and overlaps. Ticking the checks is a person's job.

#Why two lists of acceptance criteria?

They have different readers. The plain list is what the client signs off. The technical list is what the reviewer checks. See Tickets and acceptance criteria.

#Working with agents

#What does the agent actually read?

The ticket, your instruction, earlier attempts at the ticket, the acceptance criteria, the recent conversation, attachments, the project goal, the ticket's outcome, the notes and specs, its own knowledge and skills, and the repository's own instructions. The run screen shows all of it. See What an agent reads.

#Can one ticket change the app and its API together?

Yes. Link the API's project to the app's project on the project form. Agent work on an app ticket then includes both, and the approval screen shows the changes for each.

#Why did the agent write a plan instead of changing files?

Either you chose a plan or analysis mode, or the agent could not write code for that project. If you expected code, try again in Implement in branch mode, and if it still writes a plan, tell your administrator.

#Can an agent install packages or run the tests?

Only if Allow shell commands was ticked when it was assigned. Without it, the agent can read and change files only, and it will stop and ask if it needs more. You can grant that from the run screen and the same run carries on.

#What happens when I reject a run?

The attempt is thrown away, nothing is saved, and the ticket goes back to the board. Request changes does the same, records your comment, and moves the ticket to Changes requested so you can assign it again.

#How does feedback on the pull request get back to the agent?

Press Address review comments on the run screen. The new comments become the agent's next instruction, on the same branch, and that run stops at the gate like any other.

#Who can approve?

People whose group allows it, and the admins of that project. Nobody can approve on someone else's behalf, and Jarvis voice cannot approve at all.

#How much does agent work cost?

Each run records what it used. The usage report adds it up by project, agent, model and person, so whoever looks after the budget can see where the effort went.

#A run has been waiting to start for a long time, or says the agent cannot sign in.

Neither is something you can fix from your side. Tell your administrator. Once it is sorted, press Retry on the run.

#Using the workspace

#Can I use Jarvis with Jira, Trello or GitHub issues?

Yes, once your administrator has connected them. Tickets come in with their comments and attachments, and Jira sprints and story points come too. See Connectors.

#Does Jarvis listen all the time?

No. It only listens after you press the microphone. Listening and speaking happen in your browser, and only the words you said are sent, as text. See Jarvis voice.

#Can Jarvis voice approve or deploy?

No. Those are not on the list of things it can do. It will tell you to open the run instead.

#Why can a teammate not see a project?

People see the clients and projects they are members of, unless their group covers the whole workspace. Add them to the project, or ask your administrator to change their group.

#I lost my authenticator app.

Ask your administrator for a one-time sign-in link. Once you are in, set up two-factor again from your security page.

#Who can see my task logs?

You and administrators. Nobody else in the workspace can see them.

#About AiM Infosoft

#How do I get in touch?

Use the contact page. Your message reaches our team directly. You can also read about AiM Infosoft, our privacy policy and our terms.