Zagru PlatformAI software engineering
From an idea to a reviewed pull request.
Bring a product idea or a written specification. Platform shapes it into a product definition and an architecture, sets up the repositories, plans the work and implements it with AI — then runs the tests itself, has the change reviewed and opens the pull request.
Early access — we are onboarding a small number of design partners.
Illustrative interface
Nine stages. A pull request at the end.
Every stage leaves a record you can open: what was decided, what was built, and the evidence that it works.
Illustration: a specification moves through discovery, product definition, architecture, repositories, backlog, implementation, validation and review, and ends as a pull request with passing checks and an approved review.
What it does
Software work you can hand over, and check.
From idea to definition
Discovery examines the idea and the market around it; an adaptive question flow turns your answers into a product definition, an architecture and decision records.
Real repositories, a real backlog
Platform sets up the repositories on GitHub, GitLab or your own Git server and breaks the work into tasks with dependencies, each passing a readiness check before any code is written.
Implementation in a sandbox
Each task runs in its own isolated workspace. AI workers change code only there, with the tools policy allows and a guard on the scope of every change.
Done means proven
Platform runs the build and the tests itself, a different model reviews the change, and anything that fails goes back through a fix loop before it is committed.
You choose the autonomy
SAFE asks for your approval before each commit, STANDARD leaves the merge to you, AUTONOMOUS merges approved work on its own — always within the limits and budget you set.
How it works
Describe. Plan. Build. Prove.
The control plane runs the plan and keeps the record; the code is written and tested on compute nodes you choose.
Describe
Submit an idea or a Markdown specification, with the project, the repository and the budget it should use.
Define
Discovery, product definition and architecture are written down as documents and decisions you can review.
Plan
The work becomes a backlog of bounded tasks with dependencies. New findings join the backlog instead of quietly widening a task.
Build
Workers implement each task in an isolated workspace on a compute node, under policy.
Prove and deliver
Tests and validation run again, the change is reviewed, then committed, pushed and opened as a pull request.
Illustrative interface
One run, with all its evidence.
Requirements and their proof, the autonomy mode, and the audit trail behind the pull request.
In depth
What stands behind each pull request.
From idea to plan
- Discovery with an evidence-graded market report
- Adaptive questions for the product definition
- Architecture, components and decision records
- A versioned project knowledge base
- Backlog with a dependency graph and a readiness check
Execution
- An isolated workspace for every task
- Tools and commands under a policy engine
- A guard on the size and scope of every change
- Review and fix loop with a second model
- Safe recovery after an interruption
Control
- Autonomy modes SAFE, STANDARD and AUTONOMOUS
- Questions and approvals in one inbox
- Usage and cost per project, with budgets
- Retries, concurrency limits and a circuit breaker
Where it runs
- SaaS control plane for projects, planning and audit
- Compute nodes on your own hardware or in your cloud
- Outbound-only connection from node to control plane
- Local models where they fit; cloud models when you allow them
Security and data
AI output is treated as untrusted input.
Platform is built like a CI system, not a chatbot: workers are not trusted, so everything they do is limited, recorded and checked.
Sandboxed workers
Workers change code only inside their own workspace; the paths and commands they can reach are confined.
Policy before every action
A policy engine decides which tools and commands a worker may use, task by task.
Independently verified
Platform runs validation itself. A model saying the work is done is never enough.
Every action audited
An append-only audit trail records each step, approval and change.
Nodes connect out
Compute nodes join after a secure enrolment and connect outbound only; there are no inbound ports to open.
Isolation and secrets
Each organisation's data is isolated at the database level; secrets are encrypted and kept out of what models see.
In the suite
Strong on its own. Stronger together.
Platform works on your own codebases. Requirements, bugs and technical-debt items become reviewed pull requests, and in the suite the issues Pulse or Shield bring to light can be handed to it as work.
The rest of Zagru Suite
Questions
What customers ask us first.
Does it replace our developers?
No. Your team sets the direction, answers the questions Platform raises and approves what is merged. Platform takes on the planning and implementation in between, and shows its evidence.
Which repositories does it work with?
GitHub, GitLab and plain Git repositories. It can start a new project from scratch or bring in an existing one.
Where is our code processed?
Code is checked out, changed and tested on compute nodes you choose, such as a machine in your office or a server in your cloud. The control plane coordinates the work and keeps the plan, the state and the audit trail.
How do we stay in control?
Choose an autonomy mode: SAFE, STANDARD or AUTONOMOUS. Budgets cap usage, and whatever Platform cannot decide on its own comes to you as a question with options.
Is it available today?
Yes, through our design-partner programme. We roll Zagru Platform out with a small number of companies, shaped around their needs; request a demo to join.
Bring a specification. Leave with a pull request.
Start with one feature on one repository; we'll go through the evidence together.
- 30-minute intro with the team that builds the product
- Demo on your own products and flows
- No commitment, no sales pressure
Get a demo with your own data
Leave your details and we'll get back to you shortly.