CrawlDex
Check

API guide

Read the trust record before web agents act.

Use the CrawlDex API as the checkpoint: fetch the Agent Trust Record, echo the decision, then report only redacted outcome evidence.

Integration path

Agent ready
OpenAPI
contract
JSON
requests
Header
agent keys

What it is for

The checkpoint call

CrawlDex is not a browser-control API. It returns a dated Agent Trust Record so an agent can decide whether to proceed, keep a human present, avoid the route, or choose an alternative.

Echoes measure whether the record changed the agent decision. Reporting is separate evidence intake. Anonymous reports are accepted for review but do not automatically move public scores.

Downloads

Machine-readable contract

Use JSON for client generation or YAML for API tools that prefer human-readable schemas.

Common endpoints

Trust record first, echo second

GET /api/v1/trust-record/:site/:task

Fetch the citable Agent Trust Record before an agent attempts a public website task.

POST /api/v1/echo

Send a decision echo after reading a trust record: followed, overrode, or partially used; add x-crawldex-agent-key for private dashboard attribution.

POST /api/v1/feedback/outcome

Send a free-text-free ATR outcome status with record_id, outcome, task_attempted, and optional redaction_status.

GET /api/v1/echoes/me

Read the authenticated agent key's memory-v0 decision echoes and override rate.

GET /api/v1/resolve

Resolve a natural-language user request to canonical CrawlDex task keys.

GET /api/v1/tasks/:task

Inspect task success criteria, posture, related tasks, and known blocker patterns.

POST /api/v1/preflight

Ask whether an agent should attempt a site/task now and which guardrails apply.

POST /api/v1/runs

Submit redacted task outcome evidence after an agent attempt.

POST /api/v1/observations

Submit a human or agent observation through the same run-report validation path.

POST /api/v1/trust-record/batch

Fetch public trust records in batches with tiered caps: 500 anonymous/free, 2,000 partner, 5,000 paid.

GET /api/v1/accounts/options

Discover account, reporting, and hosted endpoint URLs.

GET /api/v1/accounts/me

Read the current agent account, resolved API tier, effective rate limit, window, and batch cap.

GET /api/v1/usage/me

Read self-scoped in-memory usage counters for the presented agent key.

GET /api/v1/principals/me/agents

List principal-owned linked agent identities with key prefixes only.

Verified read

Fetch an Agent Trust Record

This copy-paste command is the only rendered curl on this page because it can be verified against production without mutating public state.

GET /api/v1/trust-record/:site/:task

bash

curl -i https://crawldex.vercel.app/api/v1/trust-record/netflix.com/subscriptions.cancel

✓ verified working 2026-07-03 · https://crawldex.vercel.app · api-atr-curl · 2026-07-03T10:04:15.000Z

All examples on this page are execution-verified against the origin shown. The same paths will be served unchanged at https://api.crawldex.com once the custom domain is attached.

Tiered keys

Capacity is assigned by humans

Agent keys can resolve to free, partner, or paid API tiers. Tiers change rate-limit capacity and batch size only; scores, verdicts, confidence, rankings, and evidence weighting never read API tier state.

Partner and paid tiers are assigned after human review. Start through the pricing page.

Self-scoped usage

Account and usage responses

`GET /api/v1/accounts/me` returns the account handle, resolved `api_tier`, API key prefix, and limits: `rate_limit_max`, `rate_limit_window_ms`, and `batch_max_pairs`.

`GET /api/v1/usage/me` returns self-scoped counters with persistence `in_memory_v0`, including `total_requests`, `rate_limited_requests`, `batch_requests`, and `batch_pairs_submitted`.

`GET /api/v1/echoes/me` returns self-scoped decision echoes and override rate for the active agent key. `GET /api/v1/principals/me/agents` returns linked identities with key prefixes only.

Add `x-crawldex-agent-key` to `POST /api/v1/echo` when a decision echo should appear in the private dashboard. Attribution never changes scores, verdicts, confidence, rankings, or evidence weighting.

curl -X POST $BASE/api/v1/echo \
  -H "content-type: application/json" \
  -H "x-crawldex-agent-key: $AGENT_KEY" \
  -d '{"record_id":"atr_0123456789abcdef","action_taken":"followed","task_attempted":true}'

Rate limits

429 response body

Rate-limit responses use `rate_limited` with `tier`, `limit`, and `retry_after_ms` so callers can back off without guessing.

Batch caps

413 response body

Over-cap batch requests use `batch_cap_exceeded` with `tier`, `max_pairs`, and `received_pairs`.

Before acting

Compatibility preflight

`POST /api/v1/preflight` remains available for existing clients. New integrations should read the Agent Trust Record first, then treat preflight language as the compatibility path.

After acting

Report request

`POST /api/v1/feedback/outcome` accepts exactly `record_id`, `outcome`, `task_attempted`, and optional `redaction_status`. It accepts no free text. An optional `x-crawldex-instance` value is salted and stored only as a hash; the raw instance is never stored. Add an optional valid `x-crawldex-agent-key` when the accepted response should return `attributed: true`; no actor identifier is returned.

The response contains `status` and `attributed`. This signal never affects scores, verdicts, confidence, rankings, evidence weighting, or reputation. Use `POST /api/v1/runs` or `POST /api/v1/observations` only for fuller redacted evidence, and never submit passwords, cookies, payment data, session tokens, private screenshots, or unredacted user content.