Bylaw

Bylaw addresses a problem most agent guardrails ignore: the tool call is correct, but the evidence behind it is garbage. A refund agent can call the refund API perfectly—but if it never checked the order status, it just refunded a shipped item. Bylaw catches that by requiring an evidence manifest before execution. The Trace-to-Gate mode is smart: you upload logs from LangSmith, Langfuse, OpenAI, or CRM update logs, and Bylaw surfaces past evidence failures. You then convert the worst ones into runtime gates. This lets teams start with a low-risk discovery phase before enforcing policies in production. Where Bylaw shines is its deterministic policy engine. The LLM helps compile rules from SOPs or approval policies, but the runtime decision is deterministic—no hallucination risk. Signed audit records give compliance teams what they need for SOC 2 or internal reviews. The main downside: Bylaw requires engineering effort. You need to instrument your agent SDK, define evidence requirements per action, and maintain rule packs. It's not a plug-and-play no-code tool. Teams with simple, deterministic agents probably don't need it. Compared to tools like Guardrails AI or NVIDIA NeMo Guardrails, Bylaw is more opinionated about evidence than general output validation. It's closer to a specialized audit layer than a broad safety framework. If you're building autonomous agents in regulated environments (fintech, healthcare, CRM automation), Bylaw is worth evaluating. One caveat: pricing is contact-only and may be high for small teams. The vendor is YC-backed but early-stage, so expect custom onboarding. For teams already running agent loops with high-stakes actions, we'd reach for Bylaw when we need to prove evidence integrity to auditors or prevent costly mistakes.