Catch a partner file feed that stopped arriving—or arrived wrong.
Run one local check after your existing SFTP, S3, or cron download. Agentsor alerts if the expected CSV or Parquet receipt misses its rolling cadence plus grace, cannot be read, is stale, violates the configured schema, or falls outside volume limits. Rows, filenames, paths, and transfer credentials stay in your environment.
Agentsor is not a file-transfer service; it runs after your existing pickup or export. Early access does not guarantee uptime, error detection, or downstream outcomes. A passing result proves only the local checks that ran.
After pickupKeep the downloader you already operate
No file uploadFiles and transfer access stay local
Contract-awareCheck more than process availability
How it works
Add a file-content-free proof step after the download you already trust.
Your existing job still collects the partner feed. The local verifier checks the completed file, and Agentsor watches for the next authenticated receipt.
01
Keep your pickup
Continue using your customer-owned SFTP, S3, managed-transfer, or export job. It must exit nonzero on failure and expose only a completed file to the next step.
02
Define “right”
Choose CSV or Parquet, required columns and compatible types, row and byte bounds, optional event-time freshness, and the expected cadence.
03
Check and report locally
The MIT-licensed CLI reads the completed file where it landed, returns a typed result, and submits a bounded receipt authenticated to your monitor.
04
Catch silence or a bad feed
A rolling deadline starts at activation and advances after each accepted receipt. It catches silence only after the selected cadence and displayed fixed grace. Transition email distinguishes missed, failed, and inconclusive runs; exact reason codes remain in local CLI output.
Data boundary
Your file is not the product.
The verifier is designed so rows and values stay where the job runs. The bearer token selects the server-side monitor; the hosted service stores the fixed receipt, detects a missing receipt, and suppresses an exact duplicate submission. There is no customer dashboard in free early access.
Envelope version, run identifier, start and finish time, overall status, file format, observed row/byte counts, fixed check and reason codes, and project-keyed contract/schema/output HMAC-SHA256 fingerprints.
Agentsor combines local CSV/Parquet contract checks with rolling missed-receipt detection. It is not a replacement for every transfer, heartbeat, or data-observability tool.
Good fit
Partner files outside your data platform
A scheduled raw CSV or Parquet file lands in your environment, rows must stay local, and you want schema, freshness, volume, and silence evidence together.
Use a heartbeat instead
Job liveness is enough
If command exit status and a missed ping fully describe success, keep your existing cron, transfer monitor, Healthchecks, Cronitor, or equivalent.
Use a data platform instead
You need deeper observability
Use dbt, Soda, or a supported warehouse-observability platform for value/anomaly tests, managed lineage, dashboards, long history, or 24/7 SLA support. Agentsor does not provide those today.
Initial contract checks
Know a scheduled data file failed before the dashboard lies.
Define the evidence your partner feed must satisfy. Unknown or unobservable evidence is marked inconclusive instead of being presented as success.
01
Rolling receipt deadline
Detect silence after the selected interval plus its displayed fixed grace. This is not a wall-clock partner SLA or scheduled time-of-day check.
02
Readable structure
Parse CSV structure or Parquet metadata locally and fail closed on malformed or unsupported input.
03
Schema contract
Compare the locally observed CSV or Parquet fields and compatible types with the explicit contract you configured.
04
Freshness and volume
Evaluate configured event-time freshness plus row and byte bounds without transmitting record contents.
05
Idempotent receipt
The server accepts an exact replay of one retained run identifier and envelope without creating a second run. Each new CLI invocation creates a new run identifier.
06
Transition email
Retain redacted receipts for seven days and send typed email when an incident opens or resolves. There is no customer-visible history screen.
Local quickstart
Prove one real partner feed locally before creating a monitor.
The MIT-licensed CLI runs without an account for local checks. It creates a private project fingerprint key, an editable TOML contract, and one fixed JSON result. No file, path, column name, or value is sent.
Python 3.11 or newer is required. Review the generated contract before relying on a result.
No checkout, automatic trial, renewal, or charge is currently available. The Pro price below is a validation target, not a purchasable plan.
Available in source
$0early access
Each verified request creates one monitor tied to one operational-email recipient. Submitted redacted run receipts have seven-day retention.
Planned · not available
$29 USD/ month
File Contracts Pro is planned at $29 USD per month for up to 10 monitored feeds. The planned bundle includes 30-day redacted run history, two alert recipients, and self-serve ingest-token rotation, monitor pause, and monitor deletion. It would be month to month and cancellable before the next renewal. It does not currently include a customer dashboard or paid support SLA. No checkout, paid entitlement, or charge is available today.
Current boundary
No dashboard or paid plan
After activating a free monitor, you may make one non-binding Pro early-access request. It creates no order, trial, charge, subscription, or marketing permission.
Create a free monitor
Start with one real partner feed.
Use a business email you control. Give the feed monitor a non-sensitive label and select its expected cadence. We send one expiring verification link; the raw ingest token is shown once after verification.
Do not enter a customer name, filename, storage path, hostname, personal data, credential, or confidential project identifier.