Log inGet started

Agentic Software Engineering

Create repeatable agentic software engineering processes

Create agents in DataPrompt with the models, tools, and context they need. Combine them into workflows that plan, implement, test, and review software, for your own projects or across a team.

Get startedExplore the workflow
process / feature-deliveryRUNNING
PROCESSTake a scoped change from plan to reviewed pull request.
  1. 01
    Plan and scopeadvanced model · plan required before step 02
    done
  2. 02
    Implementefficient model · repository workspace
    done
  3. 03
    Run your checksyour suite, your commands · halts on failure
    42s
  4. 04
    Reviewadvanced model · diff, risks and cost attached
    next
Process budget used34%
4 stages · 2 models · 61% cachedApproval required before handoff

How it works

Create your agents and put them to work together

Build and configure agents directly in DataPrompt. Give each one a role, instructions, a model, and access to the tools and project knowledge it needs. Test and refine its behavior as you develop it.

Use agents individually or combine them with Python steps in a reusable workflow. Connect your repository, define the handoffs, and add checks and approvals where needed. Review the results and costs to improve the next run.

Workflow StepsSteps run sequentially; next links are set by order.
No steps defined yet.
  1. 1
    Plan and scopeAgent step
    Agent
  2. 2
    ImplementAgent step
    Agent
  3. 3
    Run checksAgent step
    Agent
  4. 4
    ReviewAgent step
    Agent

Try the workflow editor: add, reorder, or remove a step.

  1. 01
    Build your agents

    Create agents for planning, coding, testing, or review. Configure their instructions, models, tools, and context, then test each agent on a task.

  2. 02
    Compose a workflow

    Arrange agents and code steps into a repeatable process. Define what each stage receives, what it produces, and which checks must pass before work moves on.

  3. 03
    Run and refine

    Start from chat, a webhook, or a schedule. Review each stage and adjust the workflow as you learn from its results.

Workflow structure

A consistent process, from planning to review

Define the stages of your workflow once, with checks and approval points where you need them. Reuse that structure across tasks and adapt it as your working practices evolve.

  1. 01Planplan the change first gate
  2. 02Implementwrite the code gate
  3. 03Testrun the tests gate
  4. 04Reviewreview before merge gate

Workflow optimization

Optimize agent behavior and control costs

Use run history to identify repeated tool calls and unnecessary spending, then refine the models, tools, and context each stage uses. Compare runs to assess the effect of your changes.

In this example, repeated searches increased the implementation cost. Explore the finding below and review a suggested change to the tools available at that stage.

1 finding in Run 15
Run 14two days ago
  1. 01
    Plan and scopeadvanced model
    $0.022
  2. 02
    Implementefficient model
    $0.068
  3. 03
    Run checkscommand step
    $0.019
  4. 04
    Reviewadvanced model
    $0.027
Whole run$0.135
Run 15just now
  1. 01
    Plan and scopeadvanced model
    $0.022—
  2. 02
    Implementefficient model
    $0.149 2.2×

    Recommended change

    Implement called repo_search14 times in this run and twice in the last one, fetching files it had already been given.

    • Run 14Run 15tokens
    • read_file668.4k
    • apply_patch223.2k
    • repo_search21437.8k

    Disable repo_search for this stage so the agent stops reaching for it.

  3. 03
    Run checkscommand step
    $0.019—
  4. 04
    Reviewadvanced model
    $0.027—
Whole run$0.216
01
Validation steps

Run checks before each handoff and review failures in the context of the stage that produced them.

02
Cost visibility

Review model usage and spending by stage to make informed choices about where to use each model.

03
Targeted retries

Review a failed stage, make an adjustment, and resume from that point in the workflow.

04
Run history

Review the models, actions, checks, and approvals associated with each run.

05
Isolated execution

Agents work in a sandbox, with repository access handled by the platform and credentials kept outside the agent environment.

Secure agent execution

Put agents to work without exposing repository credentials

Set the scope of repository work before a run begins. Agents work in an isolated workspace while DataPrompt manages credentials and access to remote actions.

The platform checks each requested remote action against the permissions assigned to the run and records it for review. This keeps access decisions separate from the agent's work.

Managed by DataPrompt

  • GitHub credential
  • Capability grants
  • Run scope and budget

Credentials and access grants remain with the platform. The agent requests repository actions through the tools assigned to its run.

agent sandbox

Tools available in this example

  • github_issue_get
  • github_issue_comments
  • source_control_push
  • source_control_create_change_request
  • prepare_change
  • publish_change

no credentialno networkno other tool

What it can do

  • Read and edit files in its own workspace
  • Run the commands and tests you allowed
  • Ask for one of the six tools it was given
  • Open a pull request for you to review

What it cannot do

  • Hold a credential or reach the network
  • Reach for a tool that is not on that list
  • Merge, approve, deploy or release anything
  • Act outside the scope the run started with

Plans

Start on your own, or run a pilot with your team

Self-serve plans for individual projects and small teams. Teams usually start with a four-week pilot on their own repository.

Professional

For one developer or a single project.

$300 / month
  • Up to 5 agents running at once
  • 50 sandbox hours a month in isolated containers we build and host
  • Additional sandbox hours at $1 each, metered on your usage page
  • Bring your own provider keys or Codex subscription
  • Approvals, validation steps, and run evidence
  • Email support
Get started
For teams
Business

For a team sharing workflows across projects.

$700 / month
  • Everything in Professional
  • Up to 15 agents running at once
  • 200 sandbox hours a month, then $1 an hour
  • Larger sandbox containers for bigger builds and test suites
  • One shared workspace for workflows and runs
  • Priority email support
Choose Business
Pilot and Enterprise

For organizations with specific requirements.

Custom
  • Four-week pilot on your repository
  • Hosted by DataPrompt or in your own cloud
  • Dedicated sandbox capacity and agent limits sized to your team
  • Guided setup with your engineers
  • Invoicing and commercial terms
  • Security and architecture review
Contact us

Sandbox hours are the time your runs spend in the isolated containers DataPrompt builds and hosts for each run. Every plan includes a monthly allowance; hours beyond it are billed at $1 each and shown on your usage page. Model usage is billed by your provider, not by DataPrompt.

Getting started

See how To Build Coding Agents With DataPrompt

Build agents. Create workflows. Automate software engineering.

Get startedContact us