Your infrastructure,
as a database.

Ship from the pull request.
Plans in seconds. Applies in parallel.
SQL across every state.

Same HCL, modules, and providers. Import Terraform state in one command.

Pull request

Humans and agents, together.

pull request #482

stategraphbotcommented

3 to add, 1 to change

✓Policy 2/2 checks passed
!Cost +$147/mo, second approval required

humanapprovedapplied

agentopened itpending

In parallel. Everything recorded.

Database

Your infrastructure becomes a graph.

PostgreSQL

Every resource a row. Every dependency an edge.

Not a state file.

Workflow

Ship from the pull request.

Open a PR, get a plan. Policy, cost, and approvals in the review.

Plans as PR comments
Approvals via CODEOWNERS, OPA policy gates
Drift detection

Explore Orchestration →

Open Add read replica #482

app-team/read-replica into main

stategraph bot commented

Plan: 3 to add, 1 to change, 0 to destroy

✓Policy 2/2 checks passed
!Cost +$147/mo, over threshold, second approval required

Applying alongside #479 and #481. No lock to wait for.

Some checks haven’t completed yet

stategraph apply pending Waiting for status to be reported

Merge pull request merge or comment to apply

Faster

Plans in seconds, not hours.

A plan walks the subgraph your change reaches, and stops there. Work scales with the size of your change, not the size of your state.

How plans get fast →

the resource you changed

Concurrency

Everyone ships at once.

Every change is a transaction with resource-level locks. Changes that don't overlap apply together. Changes that do are rejected at commit.

No global lock. No queue, no force-unlock.
Cross-state changes commit as one atomic transaction.
One graph at any size. You never split state to scale.

How transactions work →

terraform apply

state: production Queued

humanadd read replica applying

cirotate tls cert lock held

agentresize payments db lock held

humanwiden subnet cidr waiting

one at a time1 lock

stategraph apply

state: production Concurrent

humanadd read replica committed

cirotate tls cert committed

agentresize payments db committed

humanwiden subnet cidr committed

no queueone transaction each

Independent changes against one state. Every plan and apply is its own transaction. Overlapping ones are rejected at commit.

Observability

Ask your infrastructure anything.

Your infrastructure is a database. Every resource, dependency, cost, and change is a row you can query, from the console, the CLI, the API, or an agent.

See the console →

stategraph sql

$ stategraph sql query "select type, count(*) from resources group by type"

aws_instance 20

aws_security_group 15

aws_subnet 6

Blast radius

Every resource a change reaches, before you merge.

Cost in the plan

Current versus planned, inline. Attribution and history.

Timeline

Every change an attributed transaction. The audit trail is the record.

Drift and gaps

What drifted, and what nothing manages. AWS and GCP.

AI agents

Built for agents. Safe by construction.

Skills teach any agent the workflows. Its access token grants plan and apply independently, down to a single resource. A model can ignore a prompt. It cannot exercise a capability it was never granted.

$ npx skills add stategraph/skills
Scoped tokens. Plan only, or plan and apply. Down to a resource pattern in one state.
On the record. Every agent action is a transaction on the timeline.

AI agent docs →

agent session · token: plan only

> What depends on payments db, and what would a resize cost?

stategraph-querySELECT … FROM resources JOIN edges …

12 dependents across 3 states

stategraph-cost$412/mo → $588/mo (+$176)

> Plan the resize.

stategraph-changestategraph plan --out resize.plan

✓ plan recorded · transaction 8f2a41 · revision 1042

✗ apply not permitted: token has plan only

→ handing off for human review

Terraform vs. Stategraph

	Terraform	Stategraph
Language	HCL	HCL, same modules and providers
State	A file behind a global lock	A dependency graph in PostgreSQL, with history
Plan	The whole file, every time	Only what your change touches
Changes in flight	One	As many as don't overlap
Across states	In sequence, by hand	One atomic transaction
Questions	grep a JSON file	SQL across everything
Workflow	Bring your own CI	Pull-request orchestration built in
AI agents	Blanket credentials	Skills and scoped tokens

Coming from Terraform?
terraform state pull, then stategraph import tf.
Export back any time.

Full comparison →

Deploy it on your terms

Stategraph Cloud

Managed SaaS. Free tier is shared; paid tiers are single-tenant with your own Postgres.

Start free →

Self-hosted

Your VPC, your PostgreSQL. State never leaves your network.

Deployment docs →

BYOC

We operate Stategraph inside your AWS, GCP, or Azure account.

Talk to sales →

See pricing →

Your infrastructure,as a database.

Ship from the pull request. Plans in seconds. Applies in parallel. SQL across every state.