Skip to content
Concept demo

A fictional brand, designed and built by Get Ranked On Top. Not a real company.

Skip to content
SHUNT(1)Shuntline Manual · rev. 3.12.0 · 2026-10-14SHUNT(1)
[00] Name — shunt · copy-on-write branches for Postgres

Branch your database like your code.

Every pull request gets its own writable copy of production — masked, isolated and ready in under a second. Diff it, merge it with a plan, or throw it away. Your seed scripts can retire with dignity.

curl -fsSL https://get.shuntline.example | sh
Synopsis
0.8sto branch 1.84 TB. Median, not best run.
0 Bcopied until you write something.
1.9Mbranches created last month, 1.7M of them deleted on schedule.
PGPostgres 14–17. MySQL 8.4 in beta, with appropriate warnings.
Fig. 1 shunt tree — the branches of prod-eu-1, right now
$ shunt tree --db prod-eu-1

main                      ● 4f2a91c  1.84 TB   protected
├─ feature/payments       ● 9be04d7  +412 MB   +3 tbl ~1 col -1 col
│  ├─ preview/pr-482      ● 2d71aa0  +6.1 MB   +1,204 -18 rows
│  └─ ci/run-88213        ○ 9be04d7  0 B       suspended
├─ fix/invoice-round      ● a0c3f19  +88 KB    ~312 rows
└─ scratch/kenji          ○ 4f2a91c  0 B       ttl 3h 51m

6 branches · 4 live · 506 MB written · 11.04 TB not copied
Fig. 2 Timetable, Friday 16 OctoberGit-style graph: feature/payments branches from main, preview/pr-482 branches from feature/payments, and feature/payments merges back into main.09:0010:0011:0012:0013:0014:00mainfeature/paymentspreview/pr-4824f2a91c7c03e1dmerge+3~1-1schema+1,204-18rows
In production at 4,100 engineering teams, includingnone of whom were paid to be here
  • Paraday
  • Quillon Health
  • Marrow & Finch
  • Obel Ledger
  • Tessary
  • Kallio Freight
[01] Terminalno signup, no network, no consequences

Three commands. That’s the product.

Click a command to run it against a pretend copy of production. The output is real in the sense that it’s exactly what you’d see; it’s fake in the sense that we didn’t actually drop your column.

branch create
pointer to the parent’s pages, fresh compute, masked.
diff
schema and rows, against any ancestor.
merge
a plan, a drift check, a lock timeout, then the work.
acme-api — zsh — shunt 3.12.0idle
Last login: Fri Oct 16 09:12:44 on ttys004
~/acme-api (main) $ shunt branch create feature/payments --from main
→ resolving main @ 4f2a91c  (prod-eu-1 · postgres 16.4)
→ snapshotting 212 tables, 1.84 TB logical
  copy-on-write: 0 bytes written. we checked twice.
→ provisioning compute  shu-c-2x  (2 vCPU · 8 GB)
✓ branch feature/payments ready in 0.81s

  postgres://app@feature-payments.eu-1.shuntline.example/app
  ttl 72h · suspends after 5m idle · masking: pii-default
~/acme-api (main) $ 

[02] How it works3 steps · 0 YAML files · 1 line of CI

Git semantics, for the part of your stack that never had them.

Step 01: Point it at Postgres

Shuntline attaches as a read replica. It never writes to your primary, and it doesn’t need superuser — just REPLICATION and a coffee’s worth of patience for the first sync.

step 01
$ shunt init \
    --from $PROD_URL
→ replica attached (lag 120ms)
✓ 212 tables tracked

Step 02: A branch per pull request

One line in CI gives every PR its own writable database, with PII masked by rule and a TTL so nobody pays for last quarter’s experiments.

step 02
$ shunt branch create \
    preview/pr-$PR --ttl 72h
→ masking 14 columns
✓ ready in 0.79s

Step 03: Diff, merge, forget

See schema and row-level changes against the parent, generate a migration plan you can actually read, then merge it — or delete the branch and pretend it never happened.

step 03
$ shunt merge --plan
+ table payment_intents
- column legacy_ref
1 destructive. --yes to proceed
[03] Benchmarksmedian of 30 runs · postgres 16 · eu-1 · sept 2026
Branch create0.8s

p50 on a 1.84 TB database. p99 is 1.6s, mostly DNS.

Written at create0 B

Copy-on-write at the page level. You store what you change.

Read overhead+0.4ms

p99 added latency on hot pages versus the parent.

Storage saved38×

Median across customers versus full nightly clones.

Time to a writable copy of a 500 GB database

pg_dump → pg_restore3h 41m
Logical replica + promote1h 12m
Volume snapshot restore27m
shunt branch create0.8s

Our bar is drawn at 1px because 0.0006% of the width did not render. Methodology and raw numbers in docs/benchmarks.

Fig. 3 Copy-on-write, drawn honestlymainfeature/payments000102030405060708098 pointers, 2 pages

A branch is a list of pointers to its parent’s pages. Write to a page and the branch gets its own copy of that one page. Here: two of ten. You pay for the orange ones.

[04] Changelogshunt changelog --follow

We ship on Tuesdays and Thursdays.

Forty-one releases this year. Semantic versioning, written by humans, with breaking changes in red so you can skim for the ones that matter.

Full history, back to v0.1 →
  1. v3.12.0perf

    shunt diff --data samples tables over 100M rows instead of scanning them. 41× faster, and it tells you it sampled, because we are not monsters.

  2. v3.11.2fix

    Branch names with a trailing slash no longer create a branch called, effectively, nothing. Thanks to the four of you who found this.

  3. v3.11.0new

    TTLs now inherit from the parent branch. preview/* branches default to 72h unless you say otherwise in shunt.toml.

  4. v3.10.0new

    Postgres 17 is generally available on every plan, including extensions pgvector 0.8 and PostGIS 3.5.

  5. v3.0.0breaking

    shunt merge now requires --yes for destructive migrations in non-interactive shells. Your CI will notice. That is the point.

[05] Keyboardshunt ui · works over ssh · 80 columns minimum

Mouse optional.

shunt ui is a full-screen TUI for browsing, diffing and merging branches. It respects $EDITOR, $PAGER and your muscle memory. There is a web dashboard too. We use it mostly for screenshots.

Keyboard shortcuts in shunt ui
KeyActionNotes
bNew branch from selectionprompts for a name; tab-completes from git
dDiff against parentschema first, data on enter
mMerge into parentalways shows the plan first
rReset to parentdiscards writes; asks once
sSuspend / resume computestorage stays, billing stops
/Filter branchesregex, because of course
g gJump to mainfor vim people
ctrl zUndo last mergewithin 15 minutes, before anyone notices
?Show this tablein case you forget it exists
[05½] Field reportsunedited, apart from the swearing

What engineers say

We deleted 1,900 lines of seed-data scripts the week we switched. Nobody has asked for them back, including the person who wrote them.
Ingrid Valtonen
Staff Engineer, Paraday
Every PR gets a real copy of production with the emails masked. QA stopped filing bugs that only reproduce on Tuesdays.
Tomás Arriaga
Head of Platform, Quillon Health
The merge plan caught a NOT NULL on a 400-million-row table at 4:55 on a Friday. I would like to thank it personally.
Priya Raman
Database Lead, Obel Ledger
[06] Pricingusd · billed monthly · cancel with one command

A table, not a funnel.

Every plan includes the CLI, the TUI, copy-on-write storage and branching that takes under a second. The differences are below. There are no hidden rows.

Plan comparison
FeatureSolo$0Team$24 / dev / momost teams start hereScale$480 / mo + usageEnterpriseLet’s talk
Live branches10UnlimitedUnlimitedUnlimited
Included storage10 GB250 GB2 TBCustom
Extra storage—$0.12 / GB-mo$0.09 / GB-moCommitted
Compute per branchShared2 vCPUUp to 16 vCPUDedicated
PII masking rules✓ included✓ included✓ included✓ included
Data diff— not included✓ included✓ included✓ included
Merge plans + undo— not included✓ included✓ included✓ included
SSO / SCIM— not included— not included✓ included✓ included
Audit log retention—30 days1 year7 years
Bring your own cloud— not included— not included— not included✓ included
SupportForumEmail, 1 business daySlack, 4 hoursNamed engineer, 1 hour
 Install the CLIStart 14-day trialStart 14-day trialBook a call
[07] Frequently askedman shunt-faq

Questions, answered plainly.

Something missing? engineers@shuntline.example reaches the people who wrote the code, not a ticket queue.

Is a branch a real Postgres database?

Yes. It is a full Postgres instance with its own connection string, roles and extensions. Your ORM, your migrations tool and psql will not be able to tell the difference, which is the point.

Do you copy production data, PII and all?

Branches are masked at creation using rules you define in shunt.toml — hash, redact, fake or keep, per column. The default ruleset catches emails, phone numbers and anything named like a secret. Unmasked branches require an explicit flag and are logged.

Can I merge data, or only schema?

Both. Schema merges produce a migration plan. Data merges are opt-in per table and resolve by primary key, with conflicts listed rather than guessed at.

What happens when main moved while I was branching?

shunt merge checks for drift against the commit your branch started from. If anything changed underneath you, it rebases the plan and shows you the difference before doing anything.

How is copy-on-write storage billed?

You pay for pages a branch has written, not for the size of its parent. A fresh branch of a 2 TB database costs nothing to store until you change something.

Where does it run?

Managed regions in eu-1, eu-2, us-1 and ap-1, or inside your own cloud account on Enterprise. Your data stays in the region you pick; we put that in the contract too.

[08] Documentation214 pages · searchable offline with shunt docs
[09] See also

Read the manual.

The quickstart takes eleven minutes, four of which are waiting for your first replica to sync. Or skip it, start a trial, and read the docs when something surprises you. Most people do.

No card. No sales call unless you type `shunt sales --please`.

NAME
    shunt-branch — create, list or delete branches

SYNOPSIS
    shunt branch create <name> [--from ref]
        [--ttl dur] [--mask rules]

DESCRIPTION
    Creates a copy-on-write branch of
    ref (default: main). No data is copied.
    Pages are duplicated on first write.

BUGS
    None known. Several suspected.