license	cc-by-4.0
task_categories	other
language	en
tags	webhooks saas api integration team-ops developer-tools reference-data
pretty_name	SaaS Webhook Event Catalog
size_categories	n<10K

SaaS Webhook Event Catalog

A structured, machine-readable catalog of webhook events emitted by 30 team-ops SaaS products. Built to be loaded as training reference, queried at integration time, or browsed via the companion site at automatelab.tech/webhooks/.

One row per (vendor, event_name) pair. Each row carries the event identifier, a paraphrased trigger description, the payload schema (JSON Schema Draft 2020-12, field names + types only — no example payloads), auth method, signature header, retry policy, and a link back to the canonical vendor docs.

Dataset summary

Rows	1,119 events across 30 vendors (v1)
Format	Parquet (primary), JSONL (debug/fallback)
Schema	JSON Schema Draft 2020-12, 11 required + 10 optional fields per row
Update cadence	Monthly auto-refresh; diff alerts on schema drift
Source	Public vendor developer documentation, linked per row via docs_url
Source code	github.com/AutomateLab-tech/saas-webhook-catalog

Vendors covered (v1)

Tier 1 — manually extracted (10): Slack, GitHub, Stripe, HubSpot, Linear, Notion, Calendly, Intercom, Zendesk, Salesforce.

Tier 2 — LLM-assisted extraction (20): Asana, Jira Cloud, Microsoft Teams, Discord, PagerDuty, ClickUp, Greenhouse, Mailchimp, Twilio, Zoom, Loom, Front, Help Scout, Freshdesk, Pipedrive, Close, Attio, BambooHR, Gusto, Ashby.

See vendors.yaml for per-vendor docs URLs and scoping notes.

Row schema

The authoritative schema is schema.json. Summary:

field	type	required	notes
vendor	string	yes	Kebab-case slug. Stable identifier.
vendor_display_name	string	yes	Human-readable name.
category	enum	yes	collaboration / dev-tools / payments / crm / support / scheduling / ops / communications / marketing / ats / hris / ecommerce
event_name	string	yes	Vendor's canonical event identifier. Never normalized.
event_namespace	string	no	Resource grouping or product line within a vendor.
trigger_description	string	yes	Paraphrased trigger summary. ≤500 chars. Facts only.
payload_schema	object	yes	JSON Schema Draft 2020-12 fragment. Field names + types only.
required_oauth_scopes	string[]	no	OAuth scopes required (if applicable).
required_subscription_event	string	no	Subscription identifier if different from event_name.
auth_method	enum	yes	hmac-sha256 / hmac-sha1 / hmac-sha512 / rsa-signature / bearer-token / shared-secret-header / mtls / basic-auth / none / other
signature_header	string	no	HTTP header name carrying the signature.
signature_algorithm_detail	string	no	Free text for nuances the enum can't capture.
retry_policy	object	no	{ max_attempts, backoff, total_retry_window }.
max_payload_size_bytes	integer	no	Documented size cap.
idempotency_key_header	string	no	Header name for per-delivery idempotency keys.
event_id_header	string	no	Header name for per-delivery event ID.
delivery_guarantees	enum	no	at-least-once / at-most-once / exactly-once / best-effort.
delivery	enum	no	webhook / websocket / sse / polling. Null = HTTP webhook (default).
docs_url	uri	yes	Canonical vendor docs URL. Always present.
last_introspected_at	date-time	yes	UTC timestamp of last successful extraction.
source_extractor_version	string	yes	Version of the per-vendor extractor that produced this row.
extraction_method	enum	yes	manual-html / manual-api-introspection / llm-assisted / vendor-openapi / vendor-asyncapi / vendor-graphql-schema
extraction_confidence	number	no	0.0 - 1.0. Set for LLM-assisted rows; null for manual.
notes	string	no	Free-text caveats (deprecation, beta, schema drift).

Loading

from datasets import load_dataset

ds = load_dataset("automatelab/saas-webhook-catalog")
# Filter to one vendor:
slack_events = ds["train"].filter(lambda row: row["vendor"] == "slack")
# Find HMAC-signed events across vendors:
hmac_events = ds["train"].filter(lambda row: row["auth_method"].startswith("hmac-"))

What this dataset is for

• AI agents writing webhook integration code. Today, agents hallucinate webhook payload field names from stale docs. Loading this dataset at inference time grounds the response in current facts.
• Building integration tooling. A webhook router, event normalizer, or queue worker can iterate this dataset to enumerate every event a vendor emits.
• Developer reference. Anyone integrating against <vendor> can query the catalog for the full event list, signature header, retry policy, etc., without scrolling 30 different docs sites.
• Schema-drift monitoring. The monthly diff highlights when a vendor adds, removes, or changes the shape of an event.

What this dataset is NOT

• Not a replacement for vendor docs. The docs_url per row is the authoritative source. The catalog is an index, not a redistribution.
• Not example payloads. We extract field names and types, not example bodies. If you need a sample payload, follow docs_url.
• Not real-time. Monthly refresh cadence. Use the vendor's own docs for breaking changes.

Dataset creation

Vendor selection (sub 1.1)

30 team-ops vendors selected by: public webhook docs, no auth wall, ≥5 distinct event types, active product. Vendor list locked in vendors.yaml; selection rationale in dataset-spec.md.

Licensing / ToS audit (sub 1.2)

Per-vendor Terms of Service audit completed 2026-05-13. Verdict: 26 cleared / 4 cleared-with-restrictions / 0 excluded / 1 dropped (Rippling, replaced with Gusto). Full audit in tos-audit-summary.md; per-vendor notes in tos-notes/.

Extraction (subs 1.5 + 1.6)

• Tier 1 (10 vendors): Per-vendor manual extractor — bespoke HTML/JSON parsing for stable, structured docs. High confidence (≥0.95).
• Tier 2 (20 vendors): LLM-assisted extraction with a JSON Schema output gate and confidence score per row. Lower-confidence rows (<0.7) flagged for the quality audit pass.

Pipeline code: pipeline/. Throttle config (per audit obligations): pipeline/throttle.yaml.

Quality audit (sub 1.8)

After extraction, every row hand-spot-checked against its source docs_url. Threshold gate: tier-1 rows must hit 100% accuracy; tier-2 rows must hit ≥0.7 extraction confidence and 90% spot-check accuracy.

Monthly refresh (sub 1.12)

A scheduled job re-runs every extractor monthly. Output is diffed against the previous run; additions, removals, and breaking schema changes are surfaced in a diff report.

Source data

Each row's docs_url field links to the public vendor docs page that authoritatively documents that event. Extraction respects:

• robots.txt for every vendor docs domain
• Per-vendor throttle caps from the ToS audit (see LICENSE.md)
• A descriptive User-Agent header identifying the catalog

No vendor-side credentials are used. No account creation. Public docs only.

Personal and sensitive information

The catalog contains no personal or sensitive information. It describes the shape of webhook events, not their contents. No real payloads, no example user data.

Considerations and limitations

• US-centric vendor selection. v1 is heavily weighted toward US-based SaaS. v2 may expand to EU/APAC vendors.
• Dev-tool slant. Selection skews collaboration / dev-tools / support. Adjacent categories (security, observability, finance ops) are out of v1 scope.
• Schema drift. Vendors change webhook payload shapes without notice. Monthly refresh narrows the lag; integrators should still validate against the live docs_url.
• LLM extraction errors on tier-2 rows. The confidence score helps, but a non-zero error rate is unavoidable. Spot-check before relying on a low-confidence row in production.
• Vendor-scope decisions. Some vendors emit events across multiple mechanisms (e.g., Salesforce Platform Events vs Change Data Capture vs Outbound Messages). v1 scope decisions per vendor are documented in vendors.yaml.
• Known v1 verification gaps. The 2026-05-13 audit (sub 1.8) could not fully verify these vendors against live docs; rows are published in v1 with a planned re-extraction in v1.1:
  • Gusto (16 rows) — tier-2 LLM-assisted; ~82% spot-check accuracy. Event names may drift from Gusto's current canonical names.
  • Zendesk (57 rows) — tier-1; docs event-type-list URL returned 404 during audit, rows were offline-reconstructed and not live-verified.
  • BambooHR (13 rows), Twilio Conversations (subset of 16 twilio rows), Attio (15 rows), Ashby (21 rows) — docs URLs returned 404/redirect or were too thin to cross-check at audit time; data is plausible but unconfirmed.

Bias

• Selection bias toward English-language docs and US/EU developer-focused SaaS.
• No bias in the data itself — these are structured facts about public APIs, not opinions or judgments.

Licensing

CC-BY-4.0 on the catalog itself; the underlying vendor facts are not relicensed. See LICENSE.md for the full stance, attribution requirements, and per-vendor compliance obligations.

Citation

@misc{automatelab_saas_webhook_catalog_2026,
  title  = {SaaS Webhook Event Catalog},
  author = {automatelab},
  year   = {2026},
  url    = {https://huggingface.co/datasets/automatelab/saas-webhook-catalog},
  note   = {Companion site: https://automatelab.tech/webhooks/}
}

Contributions

Issues, schema corrections, vendor additions: open an issue on the HuggingFace dataset discussion or on github.com/AutomateLab-tech/saas-webhook-catalog. New vendors are reviewed against the v1 selection criteria (team-ops category, public webhook docs, ≥5 event types, active product) and a ToS audit before inclusion.