---
title: "Proofline | Official Site"
description: "Proofline is a production agent for AWS and Google Cloud. Check changes, investigate production and read answers with the evidence attached."
url: https://proofline.sh/
---

# Prevent and explain deployment failures with production evidence

Proofline checks changes, verifies deploys and investigates alarms on AWS and Google Cloud.

[Start free for one month](/free-code-review/start/)

Trusted by

-   ![ARC Intelligence](/customers/arc-intelligence.svg)
-   ![Argon AI](/customers/argon-ai.svg)
-   ![Arctic Health](/customers/arctic-health.svg)
-   ![thirdface](/customers/thirdface.svg)
-   ![Codemod](/customers/codemod.svg)
-   ![Events in Minutes](/customers/events-in-minutes.svg)

## An agent with deep knowledge of your production

Proofline connects code, infrastructure and telemetry to explain what can fail and what changed in production.

-   ### Check a change before it ships
    
    Review the exact commit against its target environment. Findings explain the failure mechanism, cite the evidence and suggest a fix.
    
-   ### Verify a deploy after it lands
    
    Compare the deployed change with observations from production. Each answer names the signals and time windows it examined.
    
-   ### Investigate an alarm or a question
    
    Follow an alarm or ask about production in Slack or on a pull request. Read the supported findings and ranked leads where your team works.
    
-   ### Look for risks between changes
    
    Scheduled scans investigate latent production risks against a pinned commit and recorded infrastructure evidence.
    

[How Proofline investigates production](/how-it-works/)

Explore [AI code review against your infrastructure](/code-review/).

## Kian, a world model of your cloud runtime

Kian is Proofline's model of what runs where, who can reach what and which limits bind in your AWS and Google Cloud environments. Every answer is checked against it, says how fresh its evidence is, and says what it does not know.

[Meet Kian](/product/kian/)

## A whole different kind of code review

-   
-   
-   
-   

### Capacity overruns

A one-line change to a task count. The failure lives in the database’s connection limit, which the diff never mentions.

Scale API to 12 tasks #42

Open **you** want to merge 1 commit into `main` from `scale-api`

**infra/ecs/api.tf****+1** _−1_

`@@ -17,4 +17,4 @@`

17 17 `resource "aws_ecs_service" "api" {`

18 18 `name = "api"`

19`− desired_count = 4`

19`+ desired_count = 12`

**proofline** bot now

#### Database connections run out before all tasks become healthy

Each task opens 20 connections before passing its health check. With 12 tasks, the service needs **240 connections**. The production database allows **197**.

View evidence

Code

`infra/ecs/api.tf:19` sets 12 tasks. The application opens its 20-connection pool before reporting healthy.

Environment

The pinned snapshot of `orders-prod` records `max_connections = 200` and 3 reserved connections.

The extra tasks can fail their health checks and stall the deployment. **Reduce the task count or pool size, or add a connection pooler.**

20 20 `}`

### Permission errors

The new action reads as stricter and safer. The running code calls the old one, so the API cannot start.

Restrict access to application config #43

Open **you** want to merge 1 commit into `main` from `restrict-config`

**infra/iam/api.tf****+1** _−1_

`@@ -22,4 +22,4 @@`

22 22 `effect = "Allow"`

23 23 `resources = ["arn:aws:s3:::app-config/prod.json"]`

24`− actions = ["s3:GetObject"]`

24`+ actions = ["s3:GetObjectVersion"]`

**proofline** bot now

#### The API loses access to its startup configuration

The startup code calls `GetObject` without a version ID. The replacement permission allows versioned reads only, so the request is denied and the API cannot start.

View evidence

Code

`infra/iam/api.tf:24` replaces `s3:GetObject`. `src/config.ts` reads `prod.json` without a version ID before starting the server.

Environment

The pinned environment snapshot links the API task to this role. No other applicable policy grants `s3:GetObject` on the configuration object.

**Keep `s3:GetObject` scoped to `app-config/prod.json`.** The API needs that action to load its configuration.

25 25 `}`

### Network outages

A security group swap that is right for the worker. It removes the only path the API has to the database.

Move database access to the worker group #44

Open **you** want to merge 1 commit into `main` from `database-ingress`

**infra/network/database.tf****+1** _−1_

`@@ -29,4 +29,4 @@`

29 29 `from_port = 5432`

30 30 `to_port = 5432`

31`− security_groups = [aws_security_group.api.id]`

31`+ security_groups = [aws_security_group.worker.id]`

**proofline** bot now

#### New API connections cannot reach the database

This replaces the database’s only ingress rule for the API security group. API tasks still connect to `orders-prod:5432`, so new connections time out after the rule changes.

View evidence

Code

`infra/network/database.tf:31` replaces the allowed source group. The API still opens PostgreSQL connections during startup and when its pool reconnects.

Environment

The pinned snapshot places API tasks in the API security group and the database in the group changed here. No other ingress rule allows that source on TCP 5432.

**Retain the API security group in the database’s allowed sources.** Add the worker group alongside it if both services need access.

32 32 `}`

### Breaking migrations

The rename is sound on its own. The release that is running still reads the old column.

Rename the order status column #45

Open **you** want to merge 1 commit into `main` from `rename-order-status`

**migrations/018\_order\_status.sql****+1** _−0_

`@@ -0,0 +1 @@`

1`+ ALTER TABLE orders RENAME COLUMN status TO state;`

**proofline** bot now

#### The running API still queries the old column

The migration renames `orders.status` before the API rollout. The currently deployed release still selects `status`, so order requests fail with a missing-column error while that release is running.

View evidence

Code

`migrations/018_order_status.sql:1` renames the column. The deployment configuration runs migrations before replacing API tasks.

Environment

The recorded production deployment identifies the running image and its commit. At that commit, `src/orders.ts` queries `orders.status`.

**Use an additive migration first.** Add and backfill `state`, deploy code that supports the transition, and remove `status` after old readers are retired.

Read the [database connection limit example](/learn/connection-limits/) or use the [database migration review checklist](/learn/database-migration-review/) to plan a schema transition.

## Grounded in your infrastructure.

Read-only access to infrastructure and bounded telemetry queries, through permissions you grant.

-   [GitHub](/integrations/github/)
-   [Amazon Web Services](/integrations/#amazon-web-services)
-   [Google Cloud](/integrations/#google-cloud)
-   [GitHub Actions](/integrations/github-actions/)
-   [Terraform](/integrations/terraform/)
-   [OpenTofu](/integrations/opentofu/)

[Explore all integrations](/integrations/)

All integrations are read-only

## Before you connect.

What access does Proofline need?

Install the GitHub App for the repositories you choose, then grant read-only access to the infrastructure in the target environment. The required permissions depend on the services you connect. [Read about security and data access.](/security/)

Can it change my infrastructure?

Proofline never mutates your infrastructure or merges changes. Your team decides how to act on its findings.

What happens when evidence is unavailable?

Proofline keeps the assessment incomplete and puts evidence gaps in console setup tasks. Supported findings still reach your team. An absence of findings does not establish that a deployment will succeed.

Does it replace tests or staging?

Keep your tests and staging environment. Proofline adds a review of the change against evidence from its target infrastructure, helping you investigate risks that depend on that environment.

Can I start for free?

Yes. New organizations get one calendar month of hosted code review from GitHub App installation. Private and public repositories are included, with no card or cloud connection required. Fair use applies. [Read the free-month terms](/free-code-review/). Existing signup and environment-ready credit grants are separate. Every plan includes unlimited people. [Compare workspace plans](/pricing/).

## One month of free code review

Try hosted review on your GitHub pull requests, in private or public repositories. No card or cloud connection required. [For new organizations; fair use applies](/free-code-review/).

1.  ### Connect GitHub
    
    Sign in and install the GitHub App on the repositories you choose. Your free month starts at installation.
    
2.  ### Choose a repository
    
    Start with one private or public repository. You do not need to connect a cloud account.
    
3.  ### Open a pull request
    
    New organizations start with automatic code review enabled for their selected repositories.
    
4.  ### Inspect a review
    
    Read findings and their cited evidence on your pull request. Use the console to check connection health and setup tasks.
    

[Start free for one month](/free-code-review/start/)

[Compare Proofline and CodeRabbit](/compare/coderabbit/).
