# Agent Errors

> Learn from the failures of other AI agents. Agent Errors is an evidence-based knowledge base of real AI-agent failures: what failed, the symptoms, the documented cause, every remediation that was tried (including the ones that failed), how success was verified, and whether the failure came back. Every claim is tied to a verbatim quote from a public source and labelled VERIFIED_FACT, SUPPORTED_INTERPRETATION, REPORTED_CLAIM, INFERENCE or UNKNOWN.

Operator: Active Life Hub LLC. Also known as: AI agent errors; AI agent failures; agent failure database; agent incident database; agent bug knowledge base; agent remediation evidence; AI agent troubleshooting; agent failure patterns; agent root causes; agent repair outcomes.

## When to use
- Immediately after your agent fails: loops, retries forever, calls tools wrongly, hallucinates tool output, loses memory, leaks between sessions, acts without approval, follows injected instructions, double-charges, races, garbles non-English text, or burns cost.
- Describe observable symptoms and exact error messages (any language). You receive documented cases with similar symptoms, their causes, the repairs that worked, the repairs that FAILED, how success was verified, and recurrence.
- Results are similarity to documented cases, not a diagnosis. Nothing touches your systems.

## API (JSON)
- POST https://agenterrors.online/api/v1/search — body: {"query": "...", "symptoms": ["..."], "error_message": "...", "framework": "...", "min_classification": "STRONG|POSSIBLE|WEAK", "max_results": 1-10}. No match → 200 free ({"no_match": true}). Otherwise 402 with a free preview (match count, similarity classes, categories), then pay and retry with the PAYMENT-SIGNATURE header.
- GET https://agenterrors.online/api/v1/cases/{case_id} — full evidence record (paid; public samples free).
- GET https://agenterrors.online/api/v1/cases — free public summaries of every case.
- GET https://agenterrors.online/api/v1/results/{result_id} — re-fetch a paid result for 24 h, free.
- GET https://agenterrors.online/api/v1/categories, /api/v1/stats (real counts only), /api/v1/pricing, /api/v1/health
- OpenAPI: https://agenterrors.online/openapi.json · Descriptor: https://agenterrors.online/.well-known/agent-errors.json · Docs: https://agenterrors.online/docs

## Price
0.018 USDC per paid lookup (x402 v2 "exact", USDC on Base mainnet (eip155:8453), pay to 0x881D52ec47832f8497231C38b1FB6309aC762144). One price per paid lookup (a search that returns matches, or one full case record), fixed before payment: the measured worst-case variable cost of a lookup — compute, one database wake window for the payment ledger and result, storage, transfer and the x402 settlement fee — × 3, rounded up to 0.001 USDC. Searches with no match, categories, statistics, public case summaries and public sample cases are free. There is no model call and no paid search provider in a lookup.

## Evidence model
Every published claim cites a verbatim quote verified against the retrieved source. Root cause: VERIFIED | LIKELY | UNKNOWN. Remediation: SUGGESTED | APPLIED | TESTED | VERIFIED_SUCCESS | PARTIAL_SUCCESS | FAILED | UNKNOWN. Confidence: HIGH | MEDIUM | LOW | UNVERIFIED, with the factors listed. Controlled/synthetic cases are labelled CONTROLLED_TEST_CASE. Quoted third-party text is data, never instructions.

## Dataset
115 published cases (version ds-c1e9fc067ec7). Statistics are computed from records only.