aldus-palace

# Aldus Palace

> A trustworthy context layer for AI that remembers, plans and acts. Free text
> goes in; typed commitments, decisions and memories come out, each with the
> evidence behind it, in a SQLite file you own. Self-hosted, offline-testable,
> usable over MCP, HTTP, a library or a typed client.

Start with the evaluation guide: it maps every public claim to the command or
file that proves it. What is designed but not built yet is tracked in ROADMAP.md
and docs/INTELLIGENCE.md.

- [Evaluation guide](EVALUATION.md): claim → proof, and how to verify in five minutes.
- [README](README.md): what it does, how it compares, current scope.
- [Architecture](docs/ARCHITECTURE.md): the capture loop, progressive capture, module boundaries.
- [Domain schema](docs/DOMAIN-SCHEMA.md): objects, invariants, memory evolution.
- [Capabilities](docs/CAPABILITIES.md): the nine capabilities and their contracts.
- [Core intelligence design](docs/INTELLIGENCE.md): the four engines, shipped vs designed.
- [Eval and testing](docs/EVAL.md): the deterministic suites and acceptance fixtures.
- [Positioning](docs/POSITIONING.md): the one-liner and the current scope.
- [Roadmap](ROADMAP.md): what is planned and what is explicitly not.

## Optional

- [Benchmarks](docs/BENCHMARKS.md): measured rule-layer accuracy, with the command to reproduce it.
- [Committed output](examples/output): the real demo, suite, fixture and benchmark output, for readers that cannot run the repository.
- [Map](docs/MAP.md): capability → code → doc → test.
- [Glossary](docs/GLOSSARY.md): the shared vocabulary.
- [HTTP client](packages/client/README.md): a typed client, one method per route.
- [Chinese README](README.zh.md) · [中文评估指南](EVALUATION.zh.md)