AI Coding
Your AI engineer works in the open.
Task in, pull request out, with every step visible: the plan, the diffs, the terminal, the questions, and the cost. Review in-app and send feedback straight back to the agent.


The loop
From assigned task to shipped PR
- 1
Assign
Give a task to an agent that has Writes code turned on.
- 2
Sandbox
An isolated sandbox, in the cloud or on a machine you own, clones the repository.
- 3
Plan & code
The agent plans, edits, runs, and asks questions in the task thread.
- 4
Review
In-app line comments; "Request changes" feeds back to the agent.
- 5
Ship
Your build, test, and lint checks pass, the PR opens, and a branch preview is a click away.
Interactive sessions
A coding room, not a black box
Every session is a shared room: the agent's reasoning and activity stream live, diffs accumulate under File Changes, the Runner exposes preview URLs, and the Terminal is right there. Queue prompts, steer mid-work, and see the cost of every turn.
Ask, Plan, and Agent modes decide whether a turn only reads, writes a plan for you to approve, or changes code. Long conversations stay usable: the agent compacts its own context, and you can start a new thread from a summary.
Review & ship
Review where the work happened
Comment on lines in the session diff and send Request changes back to the agent. When the work is ready, one button names the next step: Push changes, Open pull request, or Ready for review. Commit messages and pull request descriptions are written from the work, so a click is enough, and a new session opens a draft pull request after its first push.
List your build, test, and lint commands as verify steps in the repository's preview manifest, and an agent's change must pass them before its pull request opens. The pull request shows a Verification section with each check and its result.
Review and merge GitHub, GitLab, and Azure Repos pull requests in-app, within your branch protection rules. When the pull requests linked to a task merge, the task can close itself.
AI delivery templates
Build the delivery loop around every agent pull request
AI review, request refinement, and the steps between them are Code workflows you create from the AI delivery templates. Nothing is installed in a project by itself: a team creates the ones each project needs.
From task to reviewed PR
One short workflow for each step
Each template reacts to one event and finishes, and the workflow for the next event takes over: triage and start, request refinement, AI review, fixing AI findings, CI repair, reviewers and reminders, handing a person's review to the agent, risk and merge, and after merge. Each one says which templates it works with, and its notes land in the task and pull request discussions the team already reads.
Some rules no template can change: an agent's approval never counts as a person's, and the branch protection on your code host is the floor. Every workflow you create is your own copy, so you change it like any other workflow.
AI review
A reviewer agent reads the pull request and reports findings on the lines of the diff and against the acceptance criteria. It advises: it never approves and never blocks a merge. Select Request AI review on any pull request, or let a template start it.
Request refinement
A template that reads each new request with the code, asks what it cannot work out in the request discussion, and proposes acceptance criteria. Once they are approved, it adds them to the request.
Development status
A view for project managers: every piece of work in progress, by person, marked Moving, Slowing, Stalled, or In review. No git knowledge needed.
Your choice of agent
Bring the coding agent your team already likes
Every session shows the same room, the same review, and the same audit trail, whichever agent does the typing. A team agent can hold its own Cursor or Codex account, or run Claude Code on a machine of its own, so the work it is given runs on the agent's account.
OpenCode
Runs on your organization’s models, managed by your admins. Nothing for anyone to connect.
Cursor
Runs on each person’s own Cursor account with an API key they connect once, billed to their Cursor plan.
Codex
Runs on each person’s own ChatGPT account after a one-time sign-in, billed to their ChatGPT plan.
Claude Code
Runs on a person’s own Claude subscription, only on their own machine, for sessions that are theirs alone.
Your machines
Run sandboxes on computers you own
Install a small runner on your own computer and your private coding sessions can run there instead of the cloud. The runner connects out to Volobox, so the machine opens no port, and it only accepts updates signed by Volosoft.
Administrators can also add shared machines and grant them to people or groups. Work that is shared with others runs in the cloud or on a shared machine, never on someone’s personal computer.
Around the loop
Everything a delivery team needs
Pull requests in-app
Org-wide and per-project PR lists with conversation, changes, and reviews.
Branch previews
Run any branch as a live app from a manifest. /init-preview-manifest writes the first one for you.
Sandboxes
In the cloud or on your own machines, with templates, budgets, pause/resume, and idle protection.
Task bridge
Agent questions and progress post to the task discussion. Reply there to steer the run without opening the session.
Code hosts
GitHub, GitLab.com, and Azure Repos, with pull request sync. GitHub and GitLab issues mirror to tasks.
Deployments
Run Azure DevOps pipelines from a project, and follow every linked pipeline across projects on one Deployments page.
Continue in another repository
When the work needs a change in another repository, carry it to a new session there with a handoff of what was done.
Agents know Volobox
A read-only volobox command in the sandbox lets the coding agent read the Volobox guide and check the preview manifest. It cannot sign in or change anything in Volobox.
Agent confidence
The agent ends its work with a high, medium, or low confidence, written into the pull request description.
Story
Follow how the work travelled, from ticket or request to task, coding session, pull request, and reviews.
Learn moreFull transparency
Bootstrap steps, activity timeline, and per-turn cost, always visible.
Governed access
Session ownership, participant roles, and secret scopes for repo credentials.
The autonomous coding loop, in your workspace, in the open.
| Standalone coding agents | Volobox |
|---|---|
| Assign an issue, wait, get a PR. | Same loop, plus watch it live: terminal, diffs, plans, questions, cost. |
| Coding-only context in a separate tool. | The agent reads the task thread, project docs, and asks the team where they already work. |
| Separate permissions and audit story. | One governance model and one audit trail for business actions and code actions. |
Watch an agent take a ticket to a PR on a real repo.
Bring one of your open issues to the demo. We'll run the loop on it.