Anvil CoderLog in

FAQ

Answers before your first run

Find answers about Anvil Coder outputs, data access, project rules and costs. Each answer links to the details so you can check the requirements and limitations for your project.

Result and flow

Does Anvil Coder merge automatically?
In the hosted service, the result arrives as a draft pull request and a human merges it. Intermediate approvals are configured per project; without a configured supervision level, the run proceeds without intermediate approval.Details: Review and supervision
Which checks actually run?
Build and test execution is optional and enabled in the hosted service, but it requires a matching toolchain. Check the run report for executed, failed and unavailable checks. Generated test files and completed work steps alone do not prove that tests passed.Details: Build and verification
What does the organisation subscription include?
The Team subscription has one organisation base price covering up to ten seats at that price. Additional seats cost extra. 75% of the base is runtime credit for the current subscription period; extra seats do not increase it. Amounts remain placeholders. BYOK model costs are paid separately to the provider. Access uses a waitlist.Details: Plan prices and conditions
What do I get at the end?
A repository with history instead of a text block someone copies code out of: every completed work step leaves its own commit. In the hosted service the result arrives where review happens anyway — as a draft pull request against the main branch, in a private repository. There is nothing to import.In detail: Integrations: the result is a repository
Which languages and frameworks are supported?
Java with Spring Boot is the most established path. Other stacks can be described in an optional profile: language, framework, versions and check commands. Build and tests also need a matching toolchain. The published Laravel reference run generated a fix and regression test, but could not execute the test in the sandbox.Why no list: Frameworks: what a stack profile contains
What if the result is no good?
There are two places for that, one inside the run and one after it. Inside, a consistency pass over the whole stands at the end. If it measures a regression, it sends the affected work steps into capped repair rounds. If it cannot measure, it lets the run through like the VERIFY gate and records an error on the gate — nothing was checked then. The measurement itself is available in the hosted service and is switched on per project; an additional build-and-test pass over the full state is optional and switched on there. Afterwards the draft pull request sits in your review tool: a comment on it becomes the requirement of an incremental follow-up run — an optional channel, switched on in the hosted service.The flow as a whole: How it works: phases and slices · Integrations: the feedback channel on the pull request
Does this replace my review?
No. A system that checks every step against the same invariants produces consistent code — not necessarily correct code. Whether the requirement itself is right, whether the domain cut holds, whether the solution fits the problem remains a human decision. The handover point is built for exactly that: in the hosted service the result arrives as a draft pull request, and a human merges it to the main branch.What guardrails do and do not do: Guardrails: what they do not do

Trust, data and control

Who sees my code?
In the hosted service every access to the factory carries its organisation: one organisation’s data does not exist for members of another, and the result sits in a private repository. What goes into the prompt is seen by the model provider — that is a question of model choice, not of where the service runs. And where generated code is built and tested in the hosted service, it happens in an unprivileged, throwaway pod instead of the coordinator.The boundaries in detail: Security: the boundaries
Does it run in Switzerland?
Yes, the hosted service runs on the Infomaniak Public Cloud in Switzerland. Source code in model requests is sent to the configured model provider. Results can also leave Switzerland if your repository or tracker is hosted abroad. Swiss hosting does not mean all data stays in Switzerland.Where the data lives: Security: where the data lives
Can I use my own model keys?
Yes — on the Claude paths the service even requires it: without your organisation’s key it rejects the request with 401. For OpenAI-compatible providers, the organisation’s key overrides a configured default. This does not depend on the plan — each organisation sets its key itself, and whoever stores one settles directly with the provider.How it is stored and what it costs: Security: your organisation’s key · Pricing: what comes on top
How do I keep control?
How much supervision you keep is a per-project setting: from every action needs a release to only the consequential ones — but without a set level, the order runs through; the release levels are a per-project feature, not a default. What holds either way: in the hosted service the result arrives as a draft pull request, and a human merges it to the main branch.Why that is the real question: For product teams: hand over the execution, keep the decision

Getting started, connectivity and cost

How do I start?
In four steps, of which only the first is a form: create an account — registration currently goes through a waitlist; connect the organisation, i.e. store GitHub access; register a project, which binds the project key to a repository — only registered, active projects run, and no YouTrack is needed for that: a project can just as well be created with its own internal tracker; then give the order. There is no start button: the ticket triggers the run — an optional channel, switched on in the hosted service.What the system docks onto: Integrations: the order comes from the ticket · Pricing: organisation subscription
Do I need YouTrack?
No: a project can be created entirely without YouTrack, with its own internal tracker — epics, stories and their states then live in the service’s own data store, with no external system involved. Whoever wants YouTrack anyway needs no instance of their own either: a shared default instance can be configured that an organisation without one falls back to. And you do not have to work in the ticket in any case: the second path is a comment on the pull request — an optional channel, switched on in the hosted service — which becomes an incremental follow-up run.Both channels side by side: Integrations: the paths in · Integrations: the shared default instance
What does it cost?
The current model combines a Team organisation subscription, runtime credit and metered consumption. Up to ten seats cost the base price; additional seats carry a surcharge. 75% of the base covers Anvil consumption in the current subscription period; BYOK model costs remain separate. Current amounts are placeholders, not confirmed sales prices. Model choice and subscription are managed separately.The tariffs and their status: Pricing: organisation subscription · Pricing: status of this page
What does a mishap cost me?
As much as you cap. One ceiling for the organisation, one per project, one per run: each can be set, and of the three levels the tightest stops the run. Without a set amount, none applies — the default is unlimited. If one is set, the service adds up what organisation and project have already cost in the running month before starting; the run that breaks the ceiling itself is stopped on the way.How the three levels work: Operations: what can be capped

Your question is not here

Send us your question about your project, operating requirements or getting started.

Questions about operations, data sovereignty and billing resolve faster in a conversation than on a page that would have to anticipate every case. Questions about the flow are usually answered better by a real story from your backlog than by any text.