01 · Task
Start with a sentence
Say what to change, in plain words. Infrar works out which repositories it touches.
The agentic loop for software in production.
Every signal, from an engineer, a user or a metric, becomes a verified change running in production.
Request access- Three kinds of signal enter the same loop: an engineer asks for a feature, a user reports a bug, a metric goes out of bounds.
- 01 Task: intent and checks.
- 02 Change: code and infrastructure, on a branch.
- 03 Preview: the full app, running.
- 04 Release: approved by a person or by team policy.
- 05 Run: watched in production.
- Production generates the next signal, and the loop starts again.
A change in Infrar, from plan to merge
Task
Add rate limiting to the public endpoints in @billing-api. 100 requests/min per API key, 429 with a Retry-After header when exceeded.
Preview
Release
A plan before any code. Infrar reads the repositories, proposes a plan and a branch, and waits for your go.
The loop
One loop, from signal to production.
Three kinds of signal feed the same loop. Agents do the work. Your reviewers and your rules decide what ships.
- 01
Task
A request becomes a goal and a plan: the repositories it touches, the branch to work on, how to check the result.
- 02
Change
Agents edit application and infrastructure code on a branch, across every repository involved. When a change touches how a component is built, Infrar builds it before pushing.
- 03
Preview
Infrar runs the whole application from the branches you choose, with its real dependencies. Your team reviews a running app, not only a diff.
- 04
Release
Today a change closes into a pull request on your git provider, merged under your branch rules. Releasing to production from Infrar will come with approval, policy and rollback.
- 05
Run
Infrar watches what it released and turns what it sees into the next signal.
Every step leaves a trail: who asked, what changed, what ran.
Use cases
Three signals, one loop.
Examples of what can start a change.
Build a feature
“Add single sign-on to the admin console.”
The team reviews the feature running in a preview before it merges.
Fix a bug
“Invoice export fails when I change the filter.”
Bring the report into a change: Infrar writes a test that catches the bug, fixes it, and you compare the fix running in a preview with your current version.
Fix the infrastructure
“Checkout latency is above target.”
Infrar finds the cause, changes the infrastructure, verifies the result and rolls back if it fails.
The product
The loop, inside Infrar.
Screens from the new Infrar interface. Names and data are illustrative.
02 · Change
A change you can review
Every change is a diff your team can read.
03 · Preview
The whole app, running
Each preview runs at a link your team can open.
04 · Release
Merge under your rules
Review the diff, the commits and the checks, then merge. Permissions and branch protection stay with your git provider.
Control
Agents do the work. You keep the keys.
This is how Infrar works today.
In the cloud, agents work in a sandbox
The AI’s tools run with no credentials and no direct network access, and only Infrar pushes, to the branch you chose. With Infrar Desktop, your own coding agent works on your machine, outside that sandbox.
Infrar proposes, you approve
The branch to use, a change to how a repository is structured, a stronger model, a merge conflict: each comes to you as a question with a suggested answer. Nothing structural happens silently.
Your name, your permissions
Commits carry your name and email and use your own key. Permissions stay with your git provider, and Infrar keeps no permission model of its own.
A trail you can read
Every step shows what the agent read, edited and ran. Every commit explains what changed and why, in a file inside your repository.
Secrets never reach a model
Infrar points to your secrets and reads them only at the moment of use, from its own encrypted store or a secret manager you connect. It never shows a value back and never gives one to a model.
Separate by design
Nothing of a namespace, the space that holds a team’s repositories and work, is visible from outside it.
Your stack
Fits the tools you already use.
Keep your git provider, connect your AI and bring your team's tools.
- GitHubCloud or self-hosted
- GitLabCloud or self-hosted
- BitbucketCloud or Data Center
- Azure DevOpsCloud or Server
- GiteaBuilt-in or self-hosted
Your git, hosted or self-hosted, or the git server built into Infrar. Servers behind your VPN are reached from your own machine through Infrar Desktop.
Your AI, on your terms.
Use your organization’s API keys from Anthropic, OpenAI or Google, or a gateway such as OpenRouter, Amazon Bedrock or Vertex AI. Or run Claude Code or Codex on your own subscription, through Infrar Desktop.
Your repositories stay the source of truth
Branches, commits and pull requests, as you know them. Infrar’s notes are plain files in a .infrar folder that travels with your branches.
Your conventions and tools
Rules, skills and tool servers (MCP) for your issue tracker, your docs or your internal APIs, set for the organization, a team or yourself. Files such as AGENTS.md and CLAUDE.md in your repositories can be honored too.
Cloud or your own machine
Work in the browser, or in Infrar Desktop for Windows, macOS and Linux, where changes and previews can run on your computer.
FAQ
Questions
How Infrar works, what you control and how to get started.
- What does Infrar do today?
- It plans a request, makes the change on a branch across your repositories, runs the whole application in a preview, and closes the work into a pull request you review and merge. Releasing to production and watching it run come next.
- Who is it for?
- Companies that want AI in their software work and have to answer for every change, starting with European companies that answer to regulators. Any team that wants the loop is welcome to talk to us.
- Which git providers does it work with?
- GitHub, GitLab, Bitbucket, Azure DevOps and Gitea, hosted or self-hosted, plus a git server built into Infrar. Your provider stays the authority for permissions and pull requests.
- Which AI does it use?
- Yours: your organization’s API keys, or your own coding-agent subscription through Infrar Desktop. Infrar does not supply models of its own today.
- Where does our data live?
- Previews and the agents’ working copies run on European infrastructure, or on your own computer if you choose. Your code stays with your git provider, and the AI provider you choose processes what you send it under its own terms.
- Can an agent change production without a person?
- Not through Infrar today. In Infrar’s cloud, agents cannot push, and merges always follow your git provider’s rules; releasing to production from Infrar will need a person’s approval or a policy your team sets.
- Do we have to change how we work?
- No. Git stays the source of truth, and Infrar’s notes are plain files in your repositories, reviewable in pull requests.
- How do we get access?
- Write to hello@infrar.io about your team, your git provider and the first thing you would like the loop to do. We onboard teams one at a time.
Built in Europe, for companies that answer to regulators.
Bring us a real change.
Pick a feature, a bug or a slow endpoint from your own backlog. We will run it through the loop with you, on your repositories.
Request accessor write to hello@infrar.io
Looks good. Do internal routes stay unlimited?
Yes. The limiter only wraps /v1, and commit 9a5f3d7 adds a test for it. Checks passed again.