Suporta
smart_toy Six agents, one ticket

A team of agents that finish the ticket

Suporta's agents do not chat about your backlog. They open pull requests against it, run the tests, and stop at the merge button so a human always has the last word.

Start free trial arrow_forward Run them locally
key Bring your own model

Your key, your model, your machine

Suporta runs the process. It never sits between you and your inference bill. Connect your own provider, pick the model each agent uses, and decide where the work actually happens.

cloud

Cloud

Agents run on our side, on the provider and key you configured. You see exactly what each ticket cost, and a monthly cap stops it running away.

dns

On your own machine

Queued tickets are picked up by a developer's laptop and worked on by their own Claude Code subscription. Nothing is billed in tokens, and your code never leaves your own git remote.

Choose once for the whole organisation, then make an exception for a single project wherever it makes sense.

Settings, AI, development agents

Model per agent

Plannerthinkingclaude-opus-5
Coderthinkingclaude-opus-5
Reviewerbalancedclaude-sonnet-5
Doc reviewcheapclaude-haiku-4-5
Where agents runCloudLocal
Monthly budget400 EUR, 38% used
The line-up

What each agent is actually for

map01Read-only

Planner agent

Understands the codebase before anything is written

The planner reads its way through your repositories first and works out the approach. You get that approach on the ticket before a single line of code appears, so you can tell in thirty seconds whether it has understood the problem.

  • check_circleCan read your code, but is never able to change it
  • check_circleFollows the conventions your project already documents
  • check_circleThe plan lands on the ticket, so a bad approach is caught early and cheaply
  • check_circleIf the planner has nothing useful to say, the work continues without it
code02Multi-repo

Coding agent

One change across all of your repositories

A project is rarely one repository. It is an API, an admin app, maybe a mobile client. The coding agent treats them as a single problem, writes one coherent change across all of them, and opens a pull request on each repository it actually touched.

  • check_circleExplores the whole codebase, but only writes where you allow it to
  • check_circleRuns only the commands you have approved, and nothing else
  • check_circleSaves its work as it goes, so an interrupted run costs minutes instead of the whole ticket
  • check_circleA run that stops early keeps everything it did, and picks up where it left off
  • check_circleOpens pull requests. Never merges them.
verified03Runs your app

Review and test agent

Actually boots the change and tries it

The reviewer installs and starts your application in an isolated sandbox, runs your test suite, and where the change warrants it clicks through the result in a real browser. Green, and it approves the pull request. Red, and the failures go straight back to the coding agent.

  • check_circleCheap by default, thorough on the changes that deserve it
  • check_circleYour own tests run before the model does, so a broken build costs nothing to catch
  • check_circleA test that was already failing on your main branch is never blamed on the pull request
  • check_circleFixes and re-reviews by itself, up to a limit you set
  • check_circleApproves under its own identity, so nothing ever approves its own work
  • check_circleSays in the approval exactly which checks it ran, so you know what was verified
menu_book04Low cost

Doc-review agent

Catches the documentation everyone forgets

It compares the documentation in the pull request against the code that actually changed, and speaks up when the two have drifted apart. Small model, small bill, so you can leave it on permanently.

  • check_circleLeaves a comment, never blocks the merge
  • check_circleSkips itself entirely when no documentation is in play
  • check_circleCheap enough that nobody argues about switching it on
science05Opt-in

Test agent

Writes the tests alongside the work

A dedicated testing phase for teams who would rather have coverage written with the change than chase it afterwards. Off until you turn it on.

  • check_circleDisabled by default
  • check_circleIts own model choice, like every other agent
description06Opt-in

Documentation agent

Keeps the docs in step with the code

Where doc review points out that the documentation has fallen behind, this one closes the gap. Also off until you turn it on.

  • check_circleDisabled by default
  • check_circleRuns on a low cost model
Staying in control

Autonomy with a hand on the brake

Anything allowed to change your codebase needs governing. Every one of these is a setting you control, not a promise we make.

toggle_on

A switch per agent

Turn any of the six on or off for your whole organisation. Nothing sneaks past a switch that is off, whether the work was started by a person, by a queue, or by another agent.

tune

A model per agent

Put your best model on planning and a cheap one on doc review. The list of available models keeps itself up to date, and your choice carries over to agents running locally.

account_balance_wallet

A monthly budget

Set a ceiling in your own currency. You get a warning at 80 percent and a hard stop at the cap. Work running on a developer machine does not count, because it costs you nothing in tokens.

bolt

Auto-run on new tickets

Pick the projects where a new ticket should put an agent to work by itself. It only ever acts on genuinely new tickets, never on your existing backlog.

supervisor_account

Human in the loop

Redirect a run while it is working, stop it and keep what it has done, tell it to carry on, or close the ticket by hand. Anything that needs you ends up in one queue.

receipt_long

Cost you can invoice

What every agent spent is measured per ticket and totalled per client and project. Add your margin and the invoice line is written for you.

Everyday AI

The small things, handled

Not every ticket needs a coding agent. These are the quiet ones, and they are the reason the queue is still readable on a Monday morning.

auto_fix_high

Ticket elaboration

Turns a two line client complaint into a ticket a developer can pick up, with the original text kept right next to it.

magic_button

Generate with AI

Write a new ticket from a short prompt, in the language the ticket is in.

straighten

Story point suggestions

An estimate to argue with, so nobody has to stare at an empty field.

chat

Ticket chat

Ask about your queue in plain language, and create one ticket or several straight out of the answer.

mark_email_read

Email analysis

The Outlook add-in reads an incoming mail and proposes the client, the project and the ticket to file it under.

sentiment_dissatisfied

Attention reports

A daily sweep that surfaces the tickets going quiet, sitting too long, or turning sour.

shuffle

Sprint reshuffle

Suggests a fairer split of the sprint across the team. It proposes, you decide.

summarize

Summaries and search

Ticket summaries, search that understands what you meant, and the weekly progress mail your clients get.

Point the agents at your backlog

Fourteen days with every agent unlocked. Use your own API key, or your own Claude Code. Either way the bill is yours, not ours.

Start free trial arrow_forwardBook a demo