Bonsai - A TypeScript-first sandboxed expression evaluator
Replaces eval() with typed sandboxing while JEXL and jsonata already exist.
Deterministic boolean algebra engine — evaluates expressions, detects contradictions, audits logic rules. MCP server, NL layer, REST API, CLI, Streamlit UI.
Catches contradictions in AI decision rules that LLMs miss, in under 10ms.
AI agent developers, teams building rule-based decision systems
SymPy · PyEDA
Let me know if you want me to try it out on something.
Replaces eval() with typed sandboxing while JEXL and jsonata already exist.
Finally separates JSON validity from actual value hallucination in LLM outputs.
Zero-dependency parsing plus regex support and implicit-AND is a practical combo — you can run the same parser in the browser via WebAssembly for quick demos. It’s a tidy, usable tool for projects that need boolean filtering without a Lucene stack, but it doesn’t attempt anything novel (no ranking, tokenization pipelines, or scaling benchmarks), so it’s best as a lightweight building block rather than a drop-in search engine.
Deterministic rule extraction from traces — same input always produces same output, no tokens.
Fail-closed guardrails for LLM actions with cryptographic approval and audit chains.
Testing framework for AI agents with LLM judges and SQLite result tracking.