LLM-JSON-guard – Middleware to auto-repair broken AI outputs
JSON repair middleware; several alternatives (Outlines, instructor, Marvin) already solve this better.
Ultra-low-latency reverse proxy that repairs truncated & malformed JSON in LLM streaming responses (OpenAI, Anthropic, Vertex AI, Bedrock) — fixes JSONDecodeError / serde_json EOF on truncated tool calls.
Fixes truncated JSON on the wire in ~10µs without SDK changes or retries.
Backend developers building LLM applications with streaming tool calls
LiteLLM · OpenRouter
JSON repair middleware; several alternatives (Outlines, instructor, Marvin) already solve this better.
Zero-dependency Zod streaming parser when zod-stream requires ecosystem buy-in.
Missing primitive: tolerant JSON parser for streaming LLM output, typed and <1KB.
Replaces O(n²) token re-parsing with true O(n) streaming; Vercel SDK does 4K re-parses on 50KB payloads.
Wire-protocol circuit breaker for agents when LangSmith costs too much.
Subdomain routing means zero /etc/hosts edits—just change base_url and capture everything.