Best Way to Sync HubSpot to Snowflake in 2026
HubSpot data reaches Snowflake three ways, through HubSpot’s own Data Share, Snowflake’s Openflow connector, or a replication tool such as Fivetran, Airbyte, or Filament. As of November 2026 the choice turns on four things, whether deletes and property history must land, which HubSpot plan you hold, how fresh the data must be, and who operates the pipeline.
Method | Best for | Main limit |
|---|---|---|
HubSpot Snowflake Data Share | Read-only analytics with property history and no pipeline | Needs Data Hub Enterprise and a supported Snowflake region |
Snowflake Openflow connector | Teams that want ingestion to live inside Snowflake | Pre-GA, private app token only, no documented delete handling |
Replication tool you run | Typed tables you own, on your compute, no per-row meter | You schedule it, and HubSpot deletes need extra work |
Managed ELT service | Broadest schemas, property history, webhook deletes | Billed by rows, credits, events, or gigabytes |
Every tool in the last two rows shares one constraint. HubSpot publishes no change feed for CRM records, so each sync polls the CRM Search API by modified timestamp. HubSpot’s search docs state that “The search endpoints are limited to 10,000 total results for any given query.” How a tool handles that cap is the first thing to check.
What decides a HubSpot to Snowflake sync
Five criteria separate the options, and throughput is not one of them, because HubSpot’s API budget caps every tool at the same ceiling.
HubSpot API fact | Effect on the sync | Source |
|---|---|---|
Search returns at most 10,000 results per query | Busy intervals must be sub-paged by record ID | |
New and updated records take a few moments to index | Cursors need a lookback or they skip rows | |
Archived records never appear in search | Deletes need list calls, webhooks, or snapshot diffs | |
Daily budget of 250,000 to 1,000,000 calls is shared by every app on the account | Two tools on one portal fight over the same quota | |
Legacy private app creation ended October 26, 2026 | New integrations authenticate with Service Keys or OAuth |
Coverage decides first, because default object sets range from eight core objects to a schema with associations and property history.
Cursor correctness means paging past the 10,000 result wall and overlapping for the indexing delay.
Deletes and merges matter because the Search API alone cannot tell you a record left HubSpot.
Freshness has a price, since Snowflake bills warehouses “per-second, with a 60-second (1-minute) minimum” on every resume, per its compute cost docs.
Pricing unit and license close the list, from monthly active rows, credits, events, and gigabytes to your own compute.
Which tools can sync HubSpot to Snowflake?
Ten paths run this route, and they split on who operates the process and whether property history and deletes arrive.
Tool | HubSpot read | Snowflake write | Pricing unit | License |
|---|---|---|---|---|
CRM objects and custom objects, incremental on modified_at | Typed tables, key-pair auth, MERGE per batch | None in the tool | Apache 2.0 | |
Everything incl. history, events, associations | Read-only secure views, 15 minute refresh | Snowflake compute only | Data Hub Enterprise | |
33 object types, full then incremental merge | Raw tables plus flattened views | Snowflake compute | Snowflake, pre-GA | |
Core objects, history tables, associations, webhook deletes | Typed tables, key-pair auth | Monthly active rows | Proprietary | |
About 40 streams incl. property history and email events | Direct-load typed tables since 4.0.0 | Credits on Cloud | ELv2 | |
25 bindings, continuous capture, OAuth only | Standard or delta, Snowpipe Streaming option | GB plus per connector | BSL 1.1 | |
Standard objects with archived flag, OAuth | Managed load, 30 minute minimum | Events | Proprietary | |
19 tables, 14 key-based incremental | Managed load | Rows per month | Proprietary | |
Core objects re-read each run, optional history | merge, scd2, upsert | None in the library | Apache 2.0 | |
23 tables, merge on hs_lastmodifieddate | One CLI command | None in the tool | FSL 1.1 |
The native share is the path with the least to run. HubSpot’s Data Share requires Data Hub Enterprise, and its query guide describes a V2_LIVE schema “which is updated every 15 minutes” beside a daily one. Deleted records vanish from the objects view, and HubSpot warns of breaking changes.
Snowflake’s own Openflow Connector for HubSpot has been in preview since April 15, 2026. It runs a full load then incremental merges, but its setup still asks for a private app token that new accounts can no longer create.
The managed services win on schema breadth. Fivetran’s connector keeps property history tables and uses webhooks to capture deletes for companies, contacts, and deals. The free plan covers 500,000 monthly active rows, and one minute syncs need an Enterprise plan.
Airbyte’s certified source lists about 40 streams and accepts a Service Key. Its November 2026 release fixed silently missing contacts by paging by record ID, and Cloud starts at 20 dollars a month with hourly syncs.
Estuary’s native capture runs continuously and can push deltas through Snowpipe Streaming, billed per gigabyte plus per connector under the Business Source License. Hevo flags archived records but runs no more often than every 30 minutes, and Stitch’s HubSpot v4 bills by rows.
The code-first tools are cheapest and narrowest. dlt’s verified source re-reads CRM objects in full on every run and merges them, and ingestr covers 23 tables in one command under a source-available license. Sling gates HubSpot behind a CLI Pro token, and Zapier moves rows one event at a time.
Debezium and PeerDB read database logs, not APIs, so neither runs this route.
Where Filament fits on HubSpot to Snowflake
Filament reads HubSpot CRM objects into typed Snowflake tables on your own infrastructure, scoring on cursor correctness, cost, and freshness control while documenting its gaps on coverage and deletes. Both the HubSpot source and the Snowflake sink carry the preview flag in the connector overview, which the docs clear only “after live validation confirms the supported behavior.” Galaxy builds Filament.
On cursors, the source page handles both HubSpot traps inside the engine. “Reads advance by object ID within the interval to avoid Search’s 10,000-result paging limit,” and each incremental run uses “a lookback to account for HubSpot’s indexing delay. The default lookback is five minutes.” A watermark advances only after the destination commits, so an interrupted read replays from the previous one.
On the write side, the sink page stages Parquet under the service user’s @~ stage and loads it with COPY INTO. “No external storage integration is required.” Upserts land through one MERGE per batch, and a CRC32-C checksum covers every batch before the write, per the integrity guide.
On cost there is no meter. The engine is Apache 2.0 and runs as one process for evaluation or on Kubernetes through the Helm chart. Each run is a bounded catch-up, so the cron schedule rather than the tool drives Snowflake compute.
The limits are written on the same pages. “Associations, archived-record feeds, marketing, and CMS resources are not included.” “Sensitive properties and property history are not included.” Incremental reads do not propagate deletions, owners are full-only, and properties land in one VARIANT column rather than typed columns. Schema evolution is add-only, and the sink accepts only unencrypted RSA keys.
The public benchmark cohort covers Postgres and MySQL routes only, so this page makes no throughput claim for HubSpot or Snowflake.
How to sync HubSpot to Snowflake with Filament
One Service Key, one Snowflake service user with a key pair, and one incremental pipeline cover a CRM. The Snowflake blocks come from the sink page and the Postgres to Snowflake how-to. The HubSpot flags follow the CLI guide’s field-to-flag rule, since the docs carry no HubSpot CLI example yet, so confirm them with --help.
Create a Service Key in HubSpot with read scopes on the objects you want, then generate the Snowflake key pair.
Create the role, warehouse, database, and service user. The warehouse suspends after 60 seconds, which is what keeps a frequent schedule affordable.
Install the CLI, then create the sink and the source and discover the resources.
Create one incremental upsert pipeline for the objects that carry modified_at and a full pipeline for owners, which have no cursor. Direct runs keep no checkpoints, so incremental reads need a saved pipeline.
Column | What lands there | Note |
|---|---|---|
| HubSpot record ID and archive state | Typed columns on every record |
| HubSpot timestamps |
|
| Every standard and custom property by internal name | One VARIANT column, not one column per property |
| Run ID and start time | Lets a model exclude a failed run |
Verify the load, then read a property out of the VARIANT column.
Add a cron schedule in the web app or the Pipelines API, where every 5 to 15 minutes is common. The how-to notes that “the schedule drives most of the Snowflake compute cost.” When an indexing delay outruns the lookback, name the resource in a run request to resync it, per the recovery guide.
Pitfalls that hit every HubSpot sync
Four problems show up regardless of tool, and each has a documented cause.
The contacts cursor is
lastmodifieddatewhile nearly every other object useshs_lastmodifieddate, per the search docs, and Airbyte’s November 2026 fix for missing contacts came from that corner.Calculated properties change without touching the modified timestamp, so Estuary refreshes them nightly and Airbyte recommends a 30 day property history lookback.
Deleted records leave HubSpot’s recycle bin after 90 days, per Hevo’s docs, so a tool that diffs snapshots has one quarter to notice.
The daily API budget is “shared across all apps within the same HubSpot account,” per the usage guidelines, so one tool’s backfill can starve another.
Which HubSpot to Snowflake tool fits which team?
A Data Hub Enterprise portal in a supported region should take the Data Share first, and everyone else chooses between renting a schema and owning a pipeline.
Team setup | Pick | Deciding fact |
|---|---|---|
Data Hub Enterprise, analytics only, no pipeline wanted | HubSpot Data Share | Property history and events included, 15 minute live views, Snowflake compute only |
Own compute, typed tables, no row meter, current values suffice | Filament | Incremental cursor sub-paged by ID, MERGE per batch, key-pair auth, Apache 2.0 |
Property history, email events, and deletes without rebuilds | Fivetran | History tables and webhook deletes for the three core objects, billed by MAR |
Widest open stream list, self-hosting acceptable | Airbyte | About 40 streams, Service Key auth, direct-load typed tables |
Ingestion must stay inside Snowflake | Openflow | Snowflake-native with 33 objects, but pre-GA and private app token only |
Minutes-fresh deltas through Snowpipe Streaming | Estuary Flow | Continuous capture, BSL license, per gigabyte and per connector pricing |
Python repo, Apache 2.0, small portal | dlt | merge and scd2 dispositions, full re-read of CRM objects each run |
Fivetran wins when property history and deletes must arrive without a scheduled rebuild, and Airbyte wins on stream breadth for teams already running it. The Data Share wins whenever the plan allows it. Filament takes the team that wants current CRM state in typed tables on its own compute and can live without history and deletes for now.
The same sink appears in the Postgres to Snowflake, MySQL to Snowflake, Stripe to Snowflake, and Notion to Snowflake guides. The write modes explainer defines upsert and replace, and Filament vs Fivetran goes deeper on the managed trade.
Frequently asked questions
Does HubSpot have a native Snowflake integration?
Yes, two. The HubSpot Snowflake Data Share exposes your portal as read-only secure views inside Snowflake, refreshed every 15 minutes, and needs a Data Hub Enterprise subscription. Snowflake also ships an Openflow Connector for HubSpot, in preview since April 2026, that polls the API and merges records into tables you own.
How often does HubSpot data update in Snowflake?
It depends on the path. The Data Share’s V2_LIVE schema updates up to every 15 minutes and V2_DAILY once a day. Replication tools run on a schedule you set, from one minute on Fivetran’s Enterprise plan to hourly on Airbyte Cloud Standard. Every warehouse resume bills at least 60 seconds, so very short schedules cost real compute.
How do deleted HubSpot records get into Snowflake?
Not through the Search API, which never returns archived records. Tools either call the list endpoints with archived set to true, subscribe to webhooks, or compare snapshots. Fivetran uses webhooks for companies, contacts, and deals. The Data Share simply drops deleted rows from its objects view. Filament’s incremental reads do not propagate deletions.
Why are records missing after an incremental HubSpot sync?
Three causes cover most cases. HubSpot’s search index lags a few moments behind writes, so a cursor without a lookback skips late arrivals. Any single query stops at 10,000 results, so a busy interval must be sub-paged by record ID. Calculated properties change without updating the record’s modified timestamp, so they only refresh on a full read.
Should a HubSpot sync use a private app, a service key, or OAuth in 2026?
New integrations should use a Service Key. HubSpot disabled legacy private app creation for existing accounts on October 26, 2026, and existing private apps keep working until September 2027. Service Keys carry the same object scopes and limits but cannot authenticate webhooks, so a tool that captures deletes through webhooks still needs an app or OAuth.
Can I sync HubSpot property history to Snowflake?
Yes with Fivetran, Airbyte, Estuary, dlt, and the Data Share, each of which exposes property history as separate tables or views. Expect it to cost more, because Fivetran counts every historical change as a monthly active row. Filament states that property history is not included, and ingestr’s table list carries no history table.
How does Filament load HubSpot data into Snowflake?
A HubSpot source reads CRM objects through the 2026-09 APIs and advances an incremental cursor on modified_at with a five minute lookback. The Snowflake sink writes each resource as Parquet to the service user’s internal stage, loads it with COPY INTO, and applies upserts as one MERGE per batch. Standard and custom properties land in one VARIANT column.
Is Filament’s HubSpot connector ready for production?
Both the HubSpot source and the Snowflake sink carry the preview flag in Filament’s connector overview, which means they have not yet been confirmed stable across every supported object and environment. The engine itself is pre-1.0. Teams running it today should expect add-only schema changes, no delete propagation, and no property history.
More articles
Questions
Answered
FAQ
What does Galaxy do?
What is Filament?
What is enterprise context management?
What does working with Galaxy look like?
How do you handle security and compliance?
Why does Galaxy build in the open?
Company
Copyright © 2026 Galaxy. All rights reserved.
