DreamVibe
Sign in Get notified

AI dev team · runs on your machine

Your own team of AI developers. You stay the boss.

Write tasks on a board. Each one runs through a workflow you control: AI steps write and review the code, command steps lint and test it, guardrails cap its budget and time, and nothing merges until you approve.

Join the launch list See a workflow

Free for 1 project · Pro free for 30 days · works with Claude Code and Codex

The DreamVibe board: tasks to do, one paused at its budget, two waiting for your approval, with spend per task.

1 task = 1 branchAgents never touch main.

$ cap per taskA run stops at its budget, not after it.

0 code uploadedYour repo stays on your disk.

Workflows

Every task runs a workflow. You decide the steps.

Like the workflow on your issue tracker, except the steps do the work. Start from the built-in pipeline or build your own per task type.

  • AI steps

    An agent writes the code in its own branch. Another reviews it. Pick the agent and model per step.

  • Command steps PRO

    Your lint, typecheck, test or migration commands, with a time limit. A non-zero exit sends the work back with the output.

  • Checks

    Commands and plain-language acceptance criteria every task must pass before it reaches you.

  • Routes and retries

    If a step fails, send the work back to Coding, retry the stage, or stop. Every loop has a limit.

  • Your approval

    A human gate before merge. Approve, or send it back with notes for the agent.

The flow editor: Preparing, Coding, Committing, a custom Lint and typecheck command step with a 10-minute limit, Verifying, AI review, human approval, Merging.

Epics

Design the feature. DreamVibe breaks it into tasks.

An epic is a big piece of work with a design canvas attached. Sketch it with the agent, split it into tasks, and let the tasks run.

An epic's design canvas, Data model tab: Customer, Plan, Subscription and Invoice entities with their fields and one-to-many relations.

  • Design canvas

    Data model, components and more, each on its own tab. Entities, fields, services, queues, and how they connect.

  • Design with the agent

    The agent reads and edits the same canvas you do. Ask it to sketch the model, or to react to what you drew.

  • Every task knows the design

    The epic's design goes into each task's prompt, so the agents build to the same plan.

  • Broken into tasks

    The epic holds its tasks. Each one runs its own workflow, with its own budget and checks.

  • Run all

    One click starts the tasks, as many side by side as your plan allows.

  • One branch for the epic

    Turn on integration and the tasks land on the epic's own branch. You review the whole feature once, then merge it.

An epic's page: the Components tab of its design, and its six tasks, from the subscriptions table to the plan picker, with Turn on integration and Run all buttons.

Guardrails

Agents work inside limits you set.

Every task has its own budget. Guardrails watch money, time, turns and tokens while the agent runs, and act at the line you drew.

Guardrails for a project: notify at 75% of a task's budget, pause at 100%; a duration limit per attempt that notifies at 80% and stops at 100%.

  • Budget per task

    A dollar cap for the task's whole life: every run, review and fix. Tokens are metered live.

  • Time per attempt

    Minutes an agent may take on one try. Plus optional caps on turns and tokens.

  • You choose what happens

    Notify, stop, hand it to a stronger model, or pause and ask you.

A task paused at its $2.00 budget: raise the budget and continue the same session, stop it, or review the work so far.
Paused at the budget: raise and continue the same session, or stop. The work so far is kept.

Review

Nothing reaches main until you say so.

Each task shows where it is in its workflow, what it spent, which checks passed and what the agent did, step by step.

A task waiting for approval: Coding and Checks done, budget $0.00 of $4.00, checks passed, the activity log, Approve and Send back buttons.

Dashboard

See what your team spent and shipped.

Spend, tokens and runs by day, model and project. Plus how much of your Claude Code and Codex plans is left, read from the agents' own local cache.

The dashboard: $148.69 spent over 30 days, 46.1M tokens, 2 runs waiting on you, plan usage for Claude Code and Codex, and a daily spend chart.

Also in the box

Built for one developer running many agents.

Isolated branches

Every task works in its own git worktree, so tasks run side by side without stepping on each other.

Your agent, your plan

Drives the Claude Code and Codex CLIs you already pay for. No token markup, no proxy in between.

Task types and epics

Tasks, bugs and epics. Each type runs its own workflow, and its own board shows that workflow's stages.

Conflicts handled

When two branches touch the same code, the agent resolves it and the task goes back through its checks.

Your code stays local

DreamVibe runs on your Mac. Your repo never leaves your disk through it.

Remote machines PRO

Run a project on a build box or a cloud VM over SSH, from the same board.

Pricing

Start free. Upgrade when one project isn't enough.

Pro is free for 30 days. Agent usage is billed by Anthropic or OpenAI on your own plan; DreamVibe adds nothing to it.

Free

$0

  • 1 project
  • 2 tasks running at once
  • Budgets, checks, review and merge
  • Claude Code and Codex
Get notified

Pro

$19.99 / month

  • Unlimited projects
  • AI review and merge
  • Unlimited concurrent runs
  • Semantic memory recall
  • Remote machines over SSH
  • Custom flows
Start the 30-day trial

Prices in USD. Billed and managed on your Verentic account. Cancel any time; see the refund policy.

Questions

Does my code leave my machine?

Not through DreamVibe. The agent CLI talks to its model provider exactly as it does when you run it yourself. DreamVibe is not in that path.

Do I need an API key?

No. A Claude subscription login works, as does an API key. Codex works with your ChatGPT plan or an OpenAI key.

How is this different from Claude Code alone?

Claude Code does one task in one terminal. DreamVibe runs a backlog in parallel, with a budget, checks and review on every task.

What happens when the trial ends?

You move to Free: 1 project, 2 tasks at once. Your work stays on your disk.

Which platforms?

macOS, on Apple silicon and Intel.

Who makes DreamVibe?

Verentic LLC, a US software company. Accounts, billing and support run on verentic.co; write to contact@verentic.co.

Hand off the next task tonight.

The macOS build is in final testing. One email when it ships, nothing else.

Join the launch list Compare plans