Documentation
Rendered from the repository's docs/ directory on every build, so what you read here is what is on main.
-rw-r--r-- CLI.md Install, provider setup, environment variables, every command and flag, troubleshooting.-rw-r--r-- DIALECT.md The language: relational core, sessions, model calls, tools, delegation, planner rules.-rw-r--r-- ARCHITECTURE.md The crates, the path of one statement, sessions, the store and trace tables, the daemon protocol, the planner.-rw-r--r-- PLAN.md The design and its rationale, milestone by milestone.-rw-r--r-- BENCHMARKS.md The packs, the learning / frozen / plain modes, the oracles, what bench report and the plots measure.-rw-r--r-- WRITEUP.md The claim, the algebra, what has been measured and how to reproduce it.-rw-r--r-- RESEARCH.md Recursive language models, SQL as a model interface, complexity results, benchmarks.
Start here
- Install and point it at a model.
- Read the dialect to see what the model can write: the relational core, model calls, tools, delegation, planner rules.
- Read the architecture for the crates and the path a statement takes.
- Then the benchmarks and the write-up for what has been measured.
Runnable demos and the shipped task packs live in the repository under demos/ and tasks/. Working rules for contributors and coding agents are in CLAUDE.md.