The integration runs as a System Administrator
How you'll know: It works, and it can also delete anything in the org.
Fix: Run it as a dedicated integration user with a permission set limited to the objects and fields the job touches.
The Salesforce REST API lets an outside system read, create, update and delete records in your org over HTTPS with JSON, including every custom object and field. The calls themselves are the easy part. Getting authorized has changed, the org has a daily call allowance, and large jobs belong on different endpoints. Here's each of those.
Base URL: https://{yourDomain}.my.salesforce.com/services/data/v{version}/. GET /services/data/ lists every version your org supports.
Auth: an OAuth 2.0 access token sent as Authorization: Bearer {token}. New integrations get it from an external client app: creating new connected apps is turned off by default since Spring '26, though existing ones keep working.
First call: GET .../query?q=SELECT+Id,Name+FROM+Account+LIMIT+10 runs SOQL and returns JSON, with nextRecordsUrl when there's more.
Enterprise Edition starts at 100,000 requests per rolling 24 hours and adds 1,000 for each Salesforce license. Other license types add smaller amounts, and other editions differ.
| Base allocation | 100,000 |
|---|---|
| 40 Salesforce licenses × 1,000 | + 40,000 |
| A sync polling 6 objects every minute (6 × 60 × 24) | − 8,640 |
| Per-record updates from a form tool, 3,000 a day | − 3,000 |
| Left for everything else that day | 128,360 |
Usage shows under Setup › Company Information (API Requests, Last 24 Hours), in the Sforce-Limit-Info header on every REST response, and in GET /limits under DailyApiRequests. Watching that header in code is how you slow down before calls start failing with REQUEST_LIMIT_EXCEEDED.
/sobjects/{Object}/{Id} Fine for a handful. A loop of these is what drains the allowance./composite/sobjects One call for the batch. Set allOrNone to decide whether one bad record rolls back the rest./composite, up to 25 subrequests Create an Account, then a Contact on it, by passing @{refId.id} between steps. The whole request counts as one call toward the daily allowance.POST /services/data/v{version}/jobs/ingest with the object, operation: upsert and the externalIdFieldName to match on. Upserting on an external ID is what makes a rerun safe.
PUT the CSV to the job's contentUrl (.../jobs/ingest/{jobId}/batches). Column headers are field API names, not labels.
PATCH .../jobs/ingest/{jobId} with state: UploadComplete. Nothing processes until you do.
GET the job until its state is JobComplete or Failed, then pull successfulResults, failedResults and unprocessedrecords.
Watch for: a job that finishes with failed rows still reports JobComplete. Read failedResults every time, and route those rows to someone.
How you'll know: It works, and it can also delete anything in the org.
Fix: Run it as a dedicated integration user with a permission set limited to the objects and fields the job touches.
How you'll know: The job dies overnight with INVALID_SESSION_ID and nobody knows until morning.
Fix: On a 401, get a new token (refresh it, or rerun the client credentials or JWT exchange) and retry the request once.
How you'll know: Calls suddenly return errors after a release.
Fix: Salesforce keeps versions for years but does retire them: versions 21.0 to 30.0 were retired in 2025. Move to a recent version and keep it in config, not scattered through code.
How you'll know: Thousands of calls a day that mostly find nothing new.
Fix: Use Change Data Capture or platform events to be told about changes, and keep a light scheduled query only as a safety net.
Most of what n-frames builds on the Salesforce REST API is the middle layer between your org and a system that doesn't have a deep Salesforce link: a form tool whose submissions should land as clean leads with campaign source and UTM data in the right fields, or a niche scheduling or ticketing tool with no AppExchange listing that has to stay in step with Salesforce records.
Inside Salesforce, logic that should run on save belongs in Salesforce Flow or Apex rather than in an outside script. Storefront APIs for B2C Commerce are a different system with their own auth, covered under Salesforce Commerce Cloud API. For hiring rather than building, see Salesforce developers.
Both work with the same Salesforce data. REST uses JSON over plain HTTP calls and is what almost all new integrations use. SOAP uses XML and a WSDL contract, and still turns up in older integrations and some enterprise tools built before REST was the default.
Not a static one. Salesforce authorizes API access with OAuth 2.0 tokens issued through an external client app (or an existing connected app). Access can be scoped to a user and permission set, and revoked without changing a shared secret everywhere.
The number goes up with each of Salesforce's three yearly releases, and it's now in the high 60s. Call GET /services/data/ on your org to see exactly which versions it supports. Old versions keep working for years, but not forever.
Yes. Salesforce's Outlook integration and Einstein Activity Capture log email and calendar activity against records. They don't push full record data both ways, so showing account details or creating records from Outlook in your own way needs a build on the REST API.
A few project tools have a native Salesforce connector for the basics; see Asana integrations for one example. Most don't, and keeping deals and projects in step usually means a custom integration on both APIs, where you choose exactly which fields and statuses sync.
Enterprise Edition gets 100,000 per rolling 24 hours plus 1,000 for each Salesforce license, with smaller amounts for other license types. Composite requests count as one call, which is the cheapest way to cut usage.
Once a job touches thousands of records. Bulk API 2.0 takes a CSV, processes it asynchronously and gives back success and failure files, using far fewer calls than record-by-record REST.