AGENT SYNC HUB Private preview
Coding agents in the cloud: a task in, a PR out
Claude Code or Codex boots in a container with the right skills, MCP servers and secrets, does the work and opens a pull request. Spend per agent, host and model on one dashboard.
Your agent, your model and your repository. The hub runs, collects the result and counts the spend.
From a task to a pull request
Run closed — secrets revoked, spend recorded.
Three steps from a task to a result
For features, incidents and reviews you would rather not run on your own laptop.
Describe the task
A card with the goal, the repository and the spec. From the console, the API or Telegram.
The hub assembles the run
Picks the agent and model, attaches skills, MCP servers and secrets, starts a container in the cluster.
Collect the PR and the numbers
The agent opens a pull request. Logs, diff and token spend stay attached to the run.
Run and measure
Two halves of one product: orchestrating runs and knowing what they cost.
Runs per task
Claude Code and Codex in Kubernetes pods. Skills, MCP servers and environment are chosen per run; secrets live exactly as long as the work does.
The result is a pull request
The agent works on a branch and opens a PR. Diff, logs and run status live in the hub instead of scattered chats.
Access to its own project only
The agent is granted a role in the DADA Cloud project for the run and loses it when the run ends. No shared tokens in the pod.
Spend per agent
Every machine and every pod pushes daily aggregates: tokens and money by host, harness, model and project, week over week, with a month forecast.
Recommendations with a number
Deterministic rules spot silent hosts, spinning tasks and expensive sub-agents, estimate the 30-day saving and give a copy-paste fix.
Team skills and context
A versioned skill registry and the team knowledge base are available to agents over MCP. The skill hub is a section of the hub, not a separate product.
An example run through the hub
An illustration of a run log. This is a prepared scenario with placeholder numbers, not a live run.
# play demo →
Start with an everyday task
The hub runs the agent with your account’s permissions and your model keys. Projects and permissions are checked just as they are in the DADA Cloud console.
01«Implement the feature from this spec and open a PR»
02«Which agent burned the most tokens for nothing this week?»
03«Run the task on Claude and Codex and compare»
Before you start
How is this different from Dada Box?
Box gives your agent a cloud computer you drive yourself. The hub takes a task, starts the agent in a container, drives it to a pull request and counts the spend. Box is an environment; the hub is a run.
Which agents are supported?
Claude Code and Codex as workspace agents in the container. Spend analytics accepts data from any harness that pushes aggregates in the quickai format.
Whose model subscription?
Yours. The hub passes your model key or a configured gateway into the container and does not resell tokens.
Where does it run?
In the DADA Cloud cluster, one hub instance per team. Run data and metrics are not mixed with other customers.
What does it cost?
It is a private preview: instances are issued on request and pricing is discussed with the team. There are no public plans yet.
What works today, and what does not
The hub is in production and we use it ourselves. Parts of the panel are still catching up with the backend.
Works now
- Claude Code and Codex runs in a container from a task card, with a pull request as the result.
- Per-run skills, MCP and secrets; project access is revoked when the run ends.
- Spend dashboard by host, harness, model and project, with recommendations and estimated savings.
- Skill registry and team knowledge base over MCP.
Not yet
- Public sign-up: instances are issued on request.
- Choosing skills and MCP for a run in the UI — configuration only for now.
- Live run log streaming: logs are available after the run completes.
Need an instance for your team?
Write to us, or open the hub if you already have access.