Change the event.Keep every consumer.
Shiftrail is the agent that turns a distributed change into coordinated execution: it finds the global impact across all your repositories and regions, plans the rollout, and opens the pull requests. You keep control of merge and deploy.
PaymentCreated v1, 4 consumers in production
Available today
Starting with Kafka and Protobuf. The agentic migration workflow is designed to extend across APIs, shared libraries, data contracts, and other cross-service changes.
- Apache Kafka
- GitHub
- GitHub Actions
- Java
- Kotlin
The problem
Your schema registry already knows the change is breaking. It rejects the pull request, and that is where it stops. What follows is weeks of archaeology: finding every consumer across dozens of repositories, writing dual-read code in each one, sequencing deploys, waiting out consumer lag, and hoping nobody replays an old topic.
Shiftrail does that part.
One contract change. Every service, in the right order.
The agent owns the work between deciding to make a change and safely completing it. Event contracts are the first supported workflow, executed in four moves.
Map
Find every reader, with receipts.
Static analysis across your repositories finds each producer and every field-level reader. Each edge carries a file, a line, a commit and a confidence score.
- Facts and inferences are kept apart
- Unproven consumers block automation until an owner confirms
Judge
Wire-compatible is not the same as safe.
Deterministic checks run first. Then every change is judged on four axes: wire, source, behavior, and whether old and new versions can coexist through a rolling deploy.
- Registry and Buf results feed the verdict
- The model never overrides a deterministic failure
Plan
A rollout, not a patch.
Shiftrail picks a strategy and orders it into steps. Each step has a gate that must hold before the next one starts, and irreversible steps are marked before anything runs.
- Expand, dual-read, dual-write, contract
- Replay and retention are part of the risk check
Execute
Draft pull requests in every repo. Gates between them.
Each consumer gets a tested pull request in its own repository. Your CI decides whether it passes. Shiftrail repairs failures, then waits for your deploys before it moves on.
- Never merges, never deploys
- Every action lands in the audit log
Autonomy, with the brakes built in.
Shiftrail may be slower than your best engineer. It is never allowed to be confidently unsafe.
A coding agent edits. A migration agent coordinates.
Registries detect incompatibility and coding agents work inside one repository. Shiftrail gives the entire change an owner: one agentic workflow with shared context, rollout state, and safety gates across every affected service.
| Capability | RegistrySchema registry | AgentCoding agent | Shiftrail |
|---|---|---|---|
| Detects a breaking change | Yes | No | uses yours |
| Finds every consumer across repositories | No | No | Yes |
| Chooses a migration strategy | No | No | Yes |
| Writes the code change | No | one repo | Yes |
| Orders the rollout across services | No | No | Yes |
| Waits for deploys and gates each step | No | No | Yes |
| Keeps migration-wide state and an audit trail | No | No | Yes |
Shiftrail