Anvil CoderLog in

AI coding agents · Built in Switzerland

AI coding agents for your software workflow

AI coding agents carry out software tasks inside a repository. Anvil Coder coordinates their work around requirements, project rules and configurable checks. You define the task and review the changes; the hosted service hands over a draft pull request.

Delegate a software task to agents

  1. Define the outcome

    Describe the change, acceptance criteria and boundaries. Your repository and project rules provide the working context.

  2. Coordinate the work

    Anvil plans small work units and their dependencies. Agents implement those units and leave traceable commits.

  3. Assess the result

    Review the changes, executed checks and unresolved findings together. A person decides whether to merge in the hosted workflow.

Choose a bounded first task

Illustrative task · not a benchmark

Reject an invalid registration

A specific failure case is a useful way to evaluate AI software development on your own project. Describe observable behaviour in the task:

During review, inspect the diff, test changes and evidence that the tests ran. A generated test file alone does not establish that the bug is fixed.

How autonomous should your coding agents be?

Autonomous coding agents need a clear assignment and appropriate approval points. Anvil lets you configure oversight per project. Without a configured approval level, a started task runs through; a person retains the merge decision in the hosted service.

Guardrails carry your architecture and quality rules into the work. Executable rules can report or block violations. An instruction in a prompt alone does not establish compliance.

Explore project rules and their verification limits

Is your project a good fit?

An existing repository, clear requirements and repeatable check commands are a useful starting point. First establish that the execution environment has the tools your project needs. Java and Spring Boot are the most established path; other stacks depend on their toolchain.

Anvil is developed by Halvic Labs in Switzerland. Your model provider’s data processing and credentials are also part of choosing how the project runs.

Check the requirements for your stack

Questions before your first run

Does an AI coding agent replace code review?
Your team still judges the result. Check whether the change solves the problem, fits the architecture and has sufficient coverage from the checks that actually ran.
Will build and tests run?
Build and tests are configurable and enabled in the hosted service. Execution requires a matching toolchain. The run report shows which checks ran and where verification is missing.
How can we start with our own ticket?
Access currently opens through a waitlist. Bring a real backlog item to a demo to discuss its scope, stack, verification steps and model usage.