Incident-aware release-risk agent

Code review finds bugs.
Rakshak remembers what broke you.

Rakshak AI reads every GitHub change and scores it against your own production incident history and known CVEs — mapping blast radius, weighing the exact files that broke before, then gating the release to ship, review, or block.

GitHub · AI · Jira · Notion · Netlify orchestrated by Swytchcode

Live decisions

What the agent decided this week

All analyses

How it works

AI reasons. Safety constrains. Gates enforce.

01github.fetch_pr

Ingest

Every pull request, commit, and repo event arrives from GitHub with its full diff, labels, and linked issue references.

02jira.search · notion.knowledge

Learn

Rakshak AI pulls related Jira issues, architecture notes, release reports, and the incident postmortems that matter for the touched files.

03llm.reason

Understand

An AI reasoning layer explains the change in plain language, names the affected systems, and forms concrete failure hypotheses from the raw evidence.

04blast-radius · risk-engine

Constrain

A deterministic safety engine maps files to modules, walks the downstream dependency graph, and scores release risk from scope, fragility, coverage, history, and known CVEs.

05jira.create · notion.write · netlify.deploy

Act

The decision becomes an enforced side effect: a Jira blocker, a Notion report, a Netlify preview, or an approved ship — and the merge gate follows it.

Not a code review clone

It answers the questions that matter before a deploy

What can break?

Blast radius mapped from the real dependency graph — not a file list.

Which parts of the product are affected?

Module-level impact with severity, coverage, and downstream services.

Has this area caused incidents before?

Postmortems and incident history are loaded from Notion and weighed into the score.

Is this safe to ship now?

A single release risk score with a hard decision on top.

What should happen next?

The agent files the blocker, writes the report, or triggers the preview itself.

Decision system

Three states. Every nuance underneath.

Ship

Contained, tested, low risk — approved for production.

Review required

Human sign-off required; routed through preview and a Jira follow-up.

Block release

Do not merge. A blocker is filed, the release report updated, and the gate stays red.

Built on the surfaces your team already runs

GitHub

PRs, commits, diff context

AI Agent

LLM reasoning over every change

Jira

Blockers & follow-up issues

Notion

Architecture, reports, incidents

Netlify

Preview & deployment actions

Swytchcode

Orchestration layer

Stop finding out after the deploy. Start deciding before it.

Try it on the demo workspace — analyze a PR and watch the agent map the blast radius and take action.