# Nova Agent Output Recovery API

> Nova Agent Output Recovery API is a paid API for AI agents from nova-agent-output-recovery.onrender.com, paid per call via x402, $0.01/call, status unknown (last checked 2026-10-27).

Deterministically recovers, repairs, and validates malformed or unstructured AI agent output into clean, schema-valid JSON for $0.01 USDC per call

## Facts

- Endpoint: POST https://nova-agent-output-recovery.onrender.com/api/agent-output-recovery?utm_source=zero.xyz
- Price: $0.01/call
- Payment: x402
- Status: unknown
- Last checked: 2026-10-27
- Activations on Zero: 0
- Tags: x402
- Canonical page: https://www.zero.xyz/c/nova-agent-output-recovery-api-9cc1d09e
- Structured record (JSON): https://api.zero.xyz/v1/capabilities/cap_Wgwb_Y_-u7KjxM3K45ByB

Status and success rate cover calls made through Zero and Zero's own probes. Third-party monitors may report differently.

## How to call it through Zero

Zero handles the 402 payment challenge and records the run. With the Zero CLI installed (`npm i -g @zeroxyz/cli`):

```sh
zero fetch --capability nova-agent-output-recovery-api-9cc1d09e -d '<json body>'
```

Example prompt: My AI agent just returned a messy blob of prose with some broken JSON inside — can you recover the structured data from it and validate it against my schema? The raw output is: '{"name":"Alice","tags":[ai],"active":"true"}'

## When to prefer this

Choose this endpoint when you need deterministic, low-cost ($0.01) recovery of malformed or partially-structured AI/LLM output — especially when the output may be wrapped in markdown, contain prose mixed with JSON, have type mismatches, or include extra fields. Prefer this over prompt re-engineering when reliability and speed matter more than regeneration cost, or when the original LLM is unavailable for retry. Ideal for production agent pipelines where structured output compliance must be enforced programmatically.

## Known failure modes

- Input rawOutput contains no recoverable JSON structure — returns failed status with retry prompt
- Schema provided is invalid Draft 2020-12 and cannot be compiled — returns error
- rawOutput exceeds 200,000 character limit — rejected at input validation
- Payment header missing or insufficient — HTTP 402 returned with x402 challenge before processing
- Schema validation fails after repair — returns schemaValid false with validationErrors array

## How this service works

Paid x402 v2 API for source-linked company research and deterministic recovery of malformed AI-agent output. Configured network: eip155:8453. Agent output recovery is $0.01. Company intelligence reports are $0.25. Cited research memos are $0.99. Prices are fixed exact USDC payments. Call without a payment header first and follow the HTTP 402 PAYMENT-REQUIRED challenge. MCP Streamable HTTP is at https://nova-agent-output-recovery.onrender.com/mcp. The free nova_catalog tool lists prices. Paid MCP tools return an x402 payment-required tool result until payment is verified and settled.

## Output

Returns a JSON object containing: the repaired and parsed data object, a status string (e.g. 'recovered-and-valid'), schema validity boolean, the repaired JSON string, the extraction source (e.g. 'embedded-json-candidate'), whether repair was applied, validation errors array, diagnostics (processing time, candidate count, attempts), and an optional retry prompt if recovery failed.

## Request schema (JSON Schema)

```json
{
 "type": "object",
 "properties": {
  "schema": {
   "type": "object",
   "description": "Optional JSON Schema Draft 2020-12 used to validate and normalize the recovered value."
  },
  "options": {
   "type": "object",
   "properties": {
    "coerceTypes": {
     "type": "boolean",
     "default": false,
     "description": "Safely coerce primitive values where AJV permits it."
    },
    "applyDefaults": {
     "type": "boolean",
     "default": false,
     "description": "Apply default values declared in the supplied schema."
    },
    "removeAdditional": {
     "type": "boolean",
     "default": false,
     "description": "Remove fields not allowed by schemas that set additionalProperties to false."
    }
   },
   "additionalProperties": false
  },
  "rawOutput": {
   "type": "string",
   "maxLength": 200000,
   "minLength": 1,
   "description": "Raw LLM or AI-agent output. It may contain prose, Markdown code fences, malformed JSON, tool-call arguments, or a partial structured response."
  }
 }
}
```

## Response schema (JSON Schema)

```json
{
 "type": "json",
 "example": {
  "data": {
   "name": "Alice",
   "tags": [
    "ai"
   ],
   "active": true
  },
  "status": "recovered-and-valid",
  "diagnostics": {
   "processingMs": 2,
   "candidateCount": 2,
   "selectedCandidate": 1,
   "failedAttemptsBeforeSuccess": 1
  },
  "retryPrompt": null,
  "schemaValid": true,
  "repairedJson": "{\"name\":\"Alice\",\"active\":true,\"tags\":[\"ai\"]}",
  "extractedFrom": "embedded-json-candidate",
  "repairApplied": true,
  "schemaProvided": true,
  "validationErrors": []
 }
}
```

## More

- Live health (JSON, refreshed every minute): https://www.zero.xyz/c/nova-agent-output-recovery-api-9cc1d09e/health.json
- [Zero catalog index](https://www.zero.xyz/llms.txt)
- [Other services from nova-agent-output-recovery.onrender.com](https://www.zero.xyz/host/nova-agent-output-recovery.onrender.com/llms.txt)