Symplex
Symplex is an AI-native platform for investigating complex systems under uncertainty.
Core trick
Compiles competing hypotheses into executable simulation models, then a meta-reasoning controller picks the next experiment to run.
Description
Symplex is an AI-native platform for investigating complex systems under uncertainty. It turns a problem into competing evidence-grounded hypotheses, compiles them into executable mathematical and computational models, and uses simulation, inference, and verification to evaluate their consequences. A meta-reasoning controller chooses the next best investigation action—such as gathering data, testing a rival model, or running an experiment—while preserving uncertainty and provenance. Symplex also supports bounded recursive self-improvement, allowing its investigation workflow to improve only through protected, auditable evaluation on fresh tasks.
README excerpt
# Symplex Symplex is a local scientific investigation workbench for complex systems. You start from a decision problem, attach evidence, and run an agent-guided investigation that produces inspectable model artifacts, experiments, and decisions. ## Quick start Requirements: Python 3.11+, macOS or Linux. ```sh uv sync cp .env.example .env # Only for first-time setup # Set OPENAI_API_KEY in .env uv run python -m symplex serve --port 8765 ``` Then open: http://127.0.0.1:8765 > This checkout includes an installed `.venv`, so `.venv/bin/python -m symplex serve` also works. ## How it works 1. Create a problem (not just a dataset). 2. Add context in **Evidence**. 3. Choose investigation depth. 4. Click **Launch solver**. The agent selects actions, while the host validates and executes them. Chat can clarify intent and perform web research, but experiments are never run silently. ## Key capabilities - FastAPI backend and responsive local workbench with persistent chats and problem versions. - Adaptive, bounded solver with focused/balanced/thorough profiles (8/16/24 action envelopes). - Typed solution graphs for subproblems, hypotheses, components, couplings, and decision contracts. - Versioned inputs and explicit problem reframing with inspectable historical context. - Hosted OpenAI Python execution that persists generated code and artifacts (CSV/JSON/PNG). - Frozen experiment protocols with host-side numerical check recomputation. - Deterministic search archive for proto