LEVERIE

Decision tables your AI agents can call — no code

Developer Tools

Artificial Intelligence

No-Code

Hey Product Hunt Thanks for stopping by. I'm the maker of LEVERIE. # WHY I BUILT THIS I kept seeing the same thing at companies rolling out AI agents: the business rules that should drive the agent's decisions were either buried in code, or scattered across a 200-page Word doc and a dozen Excel sheets. When a vendor said "just send us the rules as structured JSON," the people who actually own those rules — the ops lead, the underwriter, the support manager — had no way to do it themselves. So the logic got translated into code by someone else, revisions waited on the vendor's release cycle, and nobody could explain why the agent made a given call. Two problems underneath that: - LLM agents are non-deterministic, but a lot of decisions (refund eligibility, loan triage, routing) need to be exact, repeatable, and auditable. - The person who knows the rules and the person who can write code are usually not the same person. # WHAT LEVERIE DOES You write your rules as a plain table — like a spreadsheet, no JSON, no glue code. From that table, LEVERIE auto-generates the input/output JSON Schema and exposes the logic as an MCP tool (and a plain HTTP API) your AI agent can call. Every call returns a trace of exactly which row matched, so you can explain any decision after the fact. A few things that make it hold up as the logic grows: - Versioned publishing — publish a specific version, and the agent calls that version. Edit, publish again, and it's live immediately. - Diff review before publishing — see exactly what changed between versions, row by row, before you ship it. - Split logic across tables — when one table gets too dense, break it into linked tables, and LEVERIE draws the connections between them so the overall flow stays readable. - Flowchart view — turn the logic in a single table into a familiar flowchart, so non-technical stakeholders can follow it without reading a grid. Try it without signing up: https://leverie.dev (it comes pre-loaded with real examples — support routing, refund checks, loan triage). GitHub: https://github.com/remi0753/leverie — stars and issues genuinely help us prioritize. Docs are here: https://leverie.dev/docs/introdu... Everything above is live and usable today, and we're actively improving it — your feedback here will directly shape what we build next. What I'd love feedback on: 1. If you build agents — would you trust offloading a decision to an external, deterministic tool like this, or do you keep that logic in your own code? Why? 2. For the MCP integration specifically: what would make you confident enough to wire this into a real agent? I'll be here all day answering everything. Thanks for taking a look.