SYSTEM OPERATIONAL
v0.1.0-beta
v0.1.0-beta · Now in early access

DriftGuard reviews every Terraform & OpenTofu PR — written by humans or AI agents. We catch cost surprises, drift, security misconfigs, and compliance gaps — and remember every failure so your agents stop making the same mistake twice.

Read the docs▸ how it works
P99
<2s
review time
Memory
384‑d
semantic
Compliance
DORA
NIS2 · ISO
Platform engineersDevOps teamsInfrastructure leadsSRE teamsAI agent operatorsCloud architectsCompliance engineersFinOps analystsSecurity engineersStaff engineersPlatform engineersDevOps teamsInfrastructure leadsSRE teamsAI agent operatorsCloud architectsCompliance engineersFinOps analystsSecurity engineersStaff engineersPlatform engineersDevOps teamsInfrastructure leadsSRE teamsAI agent operatorsCloud architectsCompliance engineersFinOps analystsSecurity engineersStaff engineersPlatform engineersDevOps teamsInfrastructure leadsSRE teamsAI agent operatorsCloud architectsCompliance engineersFinOps analystsSecurity engineersStaff engineers
PR review

Every AI agent output, reviewed before merge.

Four analysis engines run in parallel — cost delta (Infracost), security (Checkov), live drift (STS), and compliance mapping. Results appear in the PR within 2 seconds.

Risk score
82
Findings
4
Cost delta
+€482/mo
Frameworks hit
3
ResourceChangeFindingCostGate
aws_rds_cluster.prod
db.r5.largedb.r5.4xlargeInstance resize — 4× cost delta+€480/mo blocked
aws_s3_bucket.tf-state
Public access block removedCKV_AWS_19 · NIS2 Art.21 · ISO 27001 A.8.24 blocked
aws_security_group.web
Ingress 0.0.0.0/0 on port 22CKV_AWS_24 · DORA Art.9 warned
aws_lambda_function.api
128 MB256 MBMemory increase — within threshold+€2/mo allowed
Review posted in 18s · risk 82/100 · 2 controls failedlive data
Runtime governance

Every agent PR. Every millisecond.

DriftGuard intercepts, analyses, and gates in under 2 seconds. Before the merge button is even available.

driftguard — runtime trace · PR #847 · acme/infra
demo
09:14:02.341SCANcursor-agent
PR #847 opened by cursor-agent — 23 resource changes
terraform plan: +12 ~8 -3
09:14:04.891COSTinfracost
Cost delta: +€892/mo on aws_rds_cluster.prod
db.r5.large → db.r5.4xlarge · threshold exceeded
09:14:05.120RECALLmemory
Memory recall: 3 similar incidents (sim ≥ 0.87)
rds.resize.prod · 2026-04-22 · outcome: blocked
09:14:05.440BLOCKcheckov
CRITICAL: aws_s3_bucket.tf-state — public access removed
CKV_AWS_19 · NIS2 Art.21 · risk +40pts
09:14:05.891GATEdriftguard
Check run posted — status: failure · risk 84/100
Merge blocked. Fix 2 critical findings to proceed.
09:14:06.012DONEdriftguard
PR comment posted — AI review + memory citations
8 findings · 3 recalled incidents · suggested fixes inline
6 events · 1.571s total · merge blockedmerge blocked · 2 critical findings
01
Webhook intercept

GitHub sends PR events to DriftGuard the instant an agent pushes. No polling.

02
Parallel analysis

Cost (Infracost), security (Checkov), drift (STS), and compliance run concurrently — not sequentially.

03
Memory recall

pgvector cosine search matches the diff against every prior incident. If this pattern broke prod before, it shows up.

04
Hard gate

A GitHub Check Run blocks the merge button. The agent cannot self-approve. Human review required above risk threshold 70.

Architecture

Infrastructure gateway, not middleware.

DriftGuard is a GitHub App — no sidecar, no agent, no cloud access beyond read-only STS. The review pipeline runs on our infrastructure. Your Terraform never moves.

fig 02 ▸ data flow
AGENTLLM / code-genintentDRIFTGUARD COREmemory.recall384‑d semantic searchpolicy.engineOPA / Regoaudit.logappend-only, signedallowblockEXECHALToutcomes feed back into memory →
01PR opens

Any Terraform or OpenTofu PR — written by an engineer or AI agent — triggers DriftGuard. We read terraform plan, infracost output, and security scans.

02Semantic recall (sub‑10ms)

The change embedding is matched against past failures across your team. If a similar incident exists, the PR comment cites it — with similarity score and link to the original incident.

03Policy gate

OPA‑compatible policies enforce allowed actions per environment, resource class, and risk score. Block merges, warn reviewers, or auto‑approve safe changes.

04Comment on PR

Single high‑signal PR comment: cost delta, drift, security findings, compliance evidence (DORA / NIS2 / ISO 27001) — and citations to past failures.

05Outcome flows back

Merge result and post‑deploy state become new memory. Your agents and engineers get sharper every day.

Operational memory

Your agents learn from production. Permanently.

Every incident — blocked deploy, drift event, policy violation — becomes a 384-d vector. When an AI agent submits a similar pattern weeks later, DriftGuard surfaces the original incident with a similarity score before the merge button is available.

Query
$ driftguard.recall(intent)
n
embedding
384‑d ▪ 12ms
index
hnsw‑m32
recall
6 / 14k events
Recalled incidents
top‑k = 6
rds.delete.prod
2026-04-22
iam.wildcard.s3
2026-04-15
tls.misconfig
2026-03-28
k8s.ingress.public
2026-03-14
rds.no.encryption
2026-02-19
ec2.security.group.0.0.0.0
2026-02-04
↑ ↓ navigate ▪ ↵ open traceindexed ▪ healthy
Full-stack governance

Six layers. One review. Zero config overhead.

Cost intelligence, live drift detection, security scanning, compliance evidence, semantic memory, and AI-native analysis — all triggered by a PR, all without changing your Terraform workflow.

$

Cost delta on every PR

Infracost-powered monthly cost diff per resource. Threshold-based blocks. No surprises in the AWS bill at the end of the month.

aws_rds_cluster.prod
db.r5.large → db.r5.4xlarge
+€480/mo · threshold exceeded → BLOCKED

Live state vs. plan diff

Compares the PR's terraform plan against the real cloud state. Catches manual changes, orphan resources, and out-of-band edits before merge.

aws_security_group.web-sg
Ingress 0.0.0.0/0 added via Console
→ DRIFT DETECTED · 3h ago

Curated security findings

Checkov + AI triage. 255 rules mapped to compliance controls. Public S3 buckets, wildcard IAM, missing encryption — flagged with fix suggestions.

aws_s3_bucket.tf-state
Public access block removed
CKV_AWS_19 · NIS2 Art.21 → BLOCKED

Semantic memory of failures

Every blocked deploy and compliance violation embedded and indexed. Open a similar PR — the original incident appears in the comment.

rds.delete.prod recalled
cos 0.94 · 2026-04-22
→ Same misconfig blocked twice

DORA / NIS2 / ISO 27001

Each finding mapped to compliance controls. Audit evidence collected on every PR — no extra workflow, no questionnaires.

DORA Art.11 · NIS2 Art.21
ISO 27001 A.8.8 · CIS v8
→ Evidence exported automatically

Built for agent contributors

AI agents (Cursor, Devin, Claude Code) write half your IaC. DriftGuard treats them like junior engineers: reviewed, mentored, fast-tracked when safe.

agent.cursor opened PR #847
23 resources · risk 84/100
→ 2 critical findings · BLOCKED

Hover each card to see a real example

Platform capabilities

Built for production infrastructure.

Measurable guarantees on every Terraform PR — coverage, accuracy, and latency.

Checkov rules
0
security checks / PR
Compliance controls
0
DORA/NIS2/ISO27001/CIS
Memory accuracy
0%
precision@k similarity
P99 latency
<0s
seconds plan → comment

Connect your GitHub installation to see live workspace metrics.

Integrate

Install in 30 seconds.

GitHub App + repo config. No SDK, no rewrites, no infrastructure changes. Optional CLI for local pre-flight.

# .github/driftguard.yml
# DriftGuard config — committed to your repo

policy:
  blast_radius: prod
  block:
    - aws_rds_cluster.*.delete
    - aws_iam_policy.*.resources=*
  warn:
    - aws_security_group.ingress.0.0.0.0/0

memory:
  retention: 365d
  cite_in_pr: true     # cite past incidents in PR comments

compliance:
  frameworks: [DORA, NIS2, ISO27001]
  evidence: ./compliance/evidence

cost:
  threshold_monthly_usd: 500
  block_above: 5000
sdk.driftguard.io ▪ semver 0.4.2 ▪ 22 lines▸ full docs
Works withTerraformOpenTofuTerragruntAtlantisSpaceliftCursorDevinClaude Code
Pricing

Simple, per-repo pricing

Start free. Add repos as you grow. Cancel anytime.

MonthlyAnnual-20%
OSS
Free forever

Self-host the analyzer. Community policies.

  • Up to 50 PR reviews / mo
  • 1 repo
  • 30-day memory retention
  • Community support
Self-host →
Most popular
Team
29/ repo / mo

Production PR reviews for human and agent contributors.

  • Unlimited PR analyses
  • Cost · drift · security · compliance
  • Semantic memory — 1 year retention
  • OPA policy bundles
  • Slack + email alerts
  • Priority support
Start free trial →
Enterprise
Custom

Self-hosted, air-gapped, regulated environments.

  • BYO-cloud / on-prem
  • SSO / SCIM provisioning
  • Custom policy modules
  • Dedicated VPC
  • 99.95% SLA
  • DORA / NIS2 / ISO 27001 evidence
Contact sales →

All plans include SOC 2 Type II (Q4 2026) · GDPR-native · EU data residency

Regain control of your cloud.

Read the docs →
OpenTofu compatibleNo code changes