Ship at the speed of AI.
Keep production safe.

The open-source release platform that scores every change, verifies every rollout, and rolls back when a regression hits.

Start building Star on GitHub

Apache-2.0 licensed Self-hosted by design Your existing CI

Interactive demo

DEPLOYMENT OVERVIEW

Every release. Under control.

Verified

main·registry/checkout:1.42#4821

Error rate0.197%6.2% vs baseline

p95 latency176ms1.7% vs baseline

Canary traffic25%Progressive rollout

Error rate

CanaryBaseline

Stage startVerification windowStage end

Your infrastructure. A little more confidence.

From pull request to production. Every decision, explained.

Works with the stack you already run

Kubernetes
Argo Rollouts
Amazon ECSbeta
Prometheus
Datadog
GitHub Actions
Slack
Go + TS SDKs

The missing step in your pipeline

More velocity.
Less crossed fingers.

Your tests check the code. Alror checks the release.
One continuous safety net, from merge to production.

Understand the change.

A risk score with a reason behind every point. Know what changed and what deserves a closer look.

Explore risk scoring

Release at the right pace.

Small changes move fast. Risky changes take smaller steps, with live metrics checking every stage.

See progressive delivery

Recover automatically.

When the canary regresses, traffic returns to the baseline. You get the evidence behind the decision.

Watch a rollback

Change intelligence

Every change has a story.Know it before you ship.

Turn a diff into a clear, auditable risk score. Critical paths, missing tests, recent rollbacks: see what matters and how it shapes your release. Try it below.

Pick a pull request

#4817 Bump analyzer to v3search-indexer · m.okafor18#4821 Batch retries for payment capturecheckout-api · agentAI62#4806 Rotate session key derivationapi-gateway, auth-gateway · j.lindqvist84

What if

1 files+98 −22

services/checkout/payment/retry.go+98 −22

Alror / change-risk · check run on #4821

checkout-api

0/100Risk score 62 out of 100, medium risk.

medium risk

Risk factors
Pts	Factor	Detail
+12	Critical servicecheckout-api is marked critical	checkout-api is marked critical
+12	Recent rollbacks2 rollbacks on these services in 30 days	2 rollbacks on these services in 30 days
+10	Sensitive pathschanges in payment	changes in payment
+10	No tests changed1 code files, 0 test files	1 code files, 0 test files
+10	AI-authoredcommit authors or trailers indicate a coding agent	commit authors or trailers indicate a coding agent
+8	Small diff120 lines changed	120 lines changed

Rollout plan

5%
25%
50%
100%

canary · 4 steps · 30m of bake time · 10m bake per step

With --fail-above 85 the check becomes a merge gate. This change passes.

Progressive delivery

Confidence at every step.A way back, built in.

Compare your canary with the baseline on real error rates and latency. Watch a healthy release progress, or see a regression caught at just 5% of traffic.

checkout-api · registry/checkout:1.42 · risk 62 · medium plan

5%bake 10min progress
25%bake 10mpending
50%bake 10mpending
100%promotepending

error_ratecanarybaseline

latency_p95canarybaseline

Verdicts per stage
Stage	Metric	Canary	Baseline	Δ	p	Result
5%	error_rate	0.208%	0.22%	−5.3%	p=0.988	pass
	latency_p95	177 ms	182 ms	−2.6%	p=0.963	pass
25%	error_rate	0.197%	0.21%	−6.2%	p=0.910	pass
	latency_p95	176 ms	179 ms	−1.7%	p=0.915	pass
50%	error_rate	0.205%	0.215%	−4.7%	p=0.935	pass
	latency_p95	177 ms	181 ms	−2.2%	p=0.964	pass

Developer first

The engine is a CLI.At home in your pipeline.

Risk scoring, rollouts, verification and rollback all run inside one Go binary, usually as a single CI step. The output is built for humans, and the exit codes are built for CI.

$ curl -fsSL https://raw.githubusercontent.com/alrors/alror/main/scripts/install.sh | sh

License: Apache-2.0
Runtime: Single Go binary
SDKs: Go · TypeScript
Exit codes: 2 rolled back · 3 gate

github.com/alrors/alror

Fits your workflow

A safer pipeline.Not a new pipeline.

Keep your CI, your registry, and your cluster. Add a check to your pull requests and one step to your deploy job. You're on your way.

Step 01

Install the CLI and initialise

One binary. alror init writes alror.yaml and a .alror/ folder. Try it with no infrastructure: the default target and metrics are simulated.

Terminal

$ curl -fsSL https://raw.githubusercontent.com/alrors/alror/main/scripts/install.sh | sh
$ cd path/to/your/repo
$ alror init      # alror.yaml + .alror/
$ alror risk      # score your current change
$ alror deploy -s checkout-api -i registry/checkout:1.43 --regress error_rate=1.8   # watch a rollback

Step 02

Score every pull request

The check action posts an Alror / change-risk check run and one sticky comment with the score, the factors and the rollout plan. Set fail-above to turn it into a merge gate.

.github/workflows/alror.yml

jobs:
  risk:
    if: github.event_name == 'pull_request'
    runs-on: ubuntu-latest
    permissions:
      contents: read
      checks: write          # the "Alror / change-risk" check run
      pull-requests: write   # the sticky risk comment
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
      - uses: alrors/alror/actions/check@main
        with:
          comment: "true"
          fail-above: "85"   # block very risky PRs

Step 03

Deploy through the gate

Keep your build. Replace the deploy command with one step. The score picks the plan, and a rolled-back release fails the job, so the bad release is visible.

.github/workflows/alror.yml

  deploy:
    if: github.event_name == 'push'
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
      # … build and push your image here …
      - uses: alrors/alror/actions/deploy@main
        id: alror
        with:
          service: checkout-api
          image: registry.example.com/checkout-api:${{ github.sha }}
          shadow: "false"    # "true" recommends rollbacks, never acts

Step 04

Verify against your metrics

Point Alror at Prometheus or Datadog. Each stage compares canary and baseline; a regression past the limit with p below α rolls traffic back on its own.

alror.yaml

services:
  - name: checkout-api
    paths: [services/checkout/]
    target: kubernetes      # Argo Rollouts, or ecs (beta)
    critical: true
metrics:
  provider: prometheus      # or datadog
  url: http://prometheus:9090
policy:
  max_regression:
    error_rate: 0.25        # +25%
    latency_p95: 0.15       # +15%
  alpha: 0.05
  auto_rollback: true
notify:
  slack_webhook: https://hooks.slack.com/services/…

The whole picture

One home for every release.Hosted on your terms.

Live rollouts, shared policies, and the story behind every deployment. Connect the CLI to your self-hosted workspace and bring your team into the loop.

Every release across your services: risk, rollout progress and verdicts.

Alror console, Deployments: every release with its status, risk score, stages, traffic and duration

Built in the open

Your infrastructure.
Your platform.

Release safety should be something you can inspect, run, and make your own. Alror is open source, from the CLI to the console. Read the code. Shape what comes next.

Explore the source

Run it on your termsSelf-host the workspace in your infrastructure.Build with usYour first issue. Your next pull request.Understand every decisionExplore the engine, policies, and architecture.

A few things to know

Good questions.
Straight answers.

Dig into the details, or start a conversation with the project.

Join the discussion

Do I have to replace my CI?

No. Alror runs as a step after your build. On GitHub Actions, use the check and deploy actions; anywhere else, run the alror binary. Your build, tests, and registry stay as they are.

How is the risk score calculated?

Alror reads the diff against your base branch: change size, affected services, critical services, sensitive paths, test coverage changes, AI authorship, and recent rollbacks. Every point is listed with its reason. You can inspect the rules and tune your policy.

What makes a rollout roll back?

A stage fails when the canary exceeds your per-metric regression limit and the Mann-Whitney test agrees. Alror compares error rate and latency against the baseline. Start in shadow mode to see its decisions before enabling automatic rollback.

Which platforms are supported?

Kubernetes through Argo Rollouts, and Amazon ECS in beta. Verification reads Prometheus or Datadog, and notifications go to Slack. GitHub Actions integrates directly; other CI systems can run the CLI.

Does my code or data leave my environment?

In local mode, nothing phones home: state is stored as JSON under .alror/. In connected mode, the CLI writes deployment events to your self-hosted workspace. Metrics are queried from your network by the CLI or runner.

What can I self-host?

The CLI, workspace console, and runner are part of the open-source platform. Run the CLI on its own or host a workspace for shared policies, deployment history, API keys, and an audit log. The repository includes Docker setup instructions.

Your next release.
A little more fearless.

Start with a single change. Build confidence from there.

Get started with Alror View on GitHub

Pts

Factor

+12

Critical servicecheckout-api is marked critical

+12

Recent rollbacks2 rollbacks on these services in 30 days

+10

Sensitive pathschanges in payment

+10

No tests changed1 code files, 0 test files

+10

AI-authoredcommit authors or trailers indicate a coding agent

Small diff120 lines changed

Stage

Metric

Canary

Baseline

Result

The engine is a CLI.At home in your pipeline.

Risk scoring, rollouts, verification and rollback all run inside one Go binary, usually as a single CI step. The output is built for humans, and the exit codes are built for CI.

$ curl -fsSL https://raw.githubusercontent.com/alrors/alror/main/scripts/install.sh | sh

License

Apache-2.0

Runtime

Single Go binary

SDKs

Go · TypeScript

Exit codes

2 rolled back · 3 gate

github.com/alrors/alror

$ curl -fsSL https://raw.githubusercontent.com/alrors/alror/main/scripts/install.sh | sh $ cd path/to/your/repo $ alror init # alror.yaml + .alror/ $ alror risk # score your current change $ alror deploy -s checkout-api -i registry/checkout:1.43 --regress error_rate=1.8 # watch a rollback

jobs: risk: if: github.event_name == 'pull_request' runs-on: ubuntu-latest permissions: contents: read checks: write # the "Alror / change-risk" check run pull-requests: write # the sticky risk comment steps: - uses: actions/checkout@v4 with: fetch-depth: 0 - uses: alrors/alror/actions/check@main with: comment: "true" fail-above: "85" # block very risky PRs

deploy: if: github.event_name == 'push' runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 with: fetch-depth: 0 # … build and push your image here … - uses: alrors/alror/actions/deploy@main id: alror with: service: checkout-api image: registry.example.com/checkout-api:${{ github.sha }} shadow: "false" # "true" recommends rollbacks, never acts

services: - name: checkout-api paths: [services/checkout/] target: kubernetes # Argo Rollouts, or ecs (beta) critical: true metrics: provider: prometheus # or datadog url: http://prometheus:9090 policy: max_regression: error_rate: 0.25 # +25% latency_p95: 0.15 # +15% alpha: 0.05 auto_rollback: true notify: slack_webhook: https://hooks.slack.com/services/…

Ship at the speed of AI.Keep production safe.

Every release. Under control.

Works with the stack you already run

More velocity.Less crossed fingers.

Understand the change.

Release at the right pace.

Recover automatically.

Every change has a story.Know it before you ship.

Confidence at every step.A way back, built in.

The engine is a CLI.At home in your pipeline.

A safer pipeline.Not a new pipeline.

Install the CLI and initialise

Score every pull request

Deploy through the gate

Verify against your metrics

One home for every release.Hosted on your terms.

Your infrastructure.Your platform.

Good questions.Straight answers.

Your next release.A little more fearless.

Ship at the speed of AI.Keep production safe.

Every release. Under control.

Works with the stack you already run

More velocity.Less crossed fingers.

Understand the change.

Release at the right pace.

Recover automatically.

Every change has a story.Know it before you ship.

Confidence at every step.A way back, built in.

The engine is a CLI.At home in your pipeline.

A safer pipeline.Not a new pipeline.

Install the CLI and initialise

Score every pull request

Deploy through the gate

Verify against your metrics

One home for every release.Hosted on your terms.

Your infrastructure.Your platform.

Good questions.Straight answers.

Your next release.A little more fearless.

Ship at the speed of AI.
Keep production safe.

More velocity.
Less crossed fingers.

Your infrastructure.
Your platform.

Good questions.
Straight answers.

Your next release.
A little more fearless.

Ship at the speed of AI.
Keep production safe.

More velocity.
Less crossed fingers.

Your infrastructure.
Your platform.

Good questions.
Straight answers.

Your next release.
A little more fearless.