JSON Repair & Parse for LLM Output

Use this endpoint when handling LLM-generated JSON that frequently contains common syntax errors like trailing commas, single-quoted strings, unquoted keys, embedded markdown code fences, JavaScript comments, or Python-style literals (True/False/None). It is especially suited for production pipelines that consume raw LLM output and need reliable, deterministic parsing without writing custom sanitization logic. Prefer it over manual regex cleaning or ad-hoc parsers when you need a robust, maintained repair solution.

• Input is so malformed it cannot be salvaged into any valid JSON structure — returns error indicating unrecoverable input
• Input is not JSON-like at all (e.g. plain prose with no JSON tokens) — may return error or unexpected result
• Ambiguous repair heuristics may occasionally produce a different valid structure than intended when multiple repair paths exist
• Payment failure or insufficient USDC balance results in 402 response before any processing occurs