Centauri — Banco de Dados Bitemporal e Orientado a IA
aniljacobv-lab/centauri
Banco de dados que nunca esquece — bitemporal, causal, IA-first
Ver no GitHubUm banco de dados de eventos bitemporal, causal e orientado a IA que registra não apenas valores atuais, mas a história completa: o que era verdadeiro em qualquer momento, o que se acreditava saber, por que mudou e o quão confiável é. Inclui Genesis Engine (cria schemas a partir de descrições em linguagem natural), CeQL (linguagem de query com tempo e causalidade como sintaxe), busca vetorial e server MCP nativo para agentes Claude.
Instalação
npx convez-ai aniljacobv-lab/centauri1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Auditoria e compliance de mudanças (AS OF e AS KNOWN AT)
- Rastreamento de causalidade e proveniência de dados
- Sistemas de registro (flight recorder operacional)
- Cenários com confiabilidade e rastreamento de fonte críticos
- Agentes IA que precisam contextualizar decisões no tempo
Veredito

Centauri é inovador em seu modelo bitemporal e integração nativa com IA, mas está em v0.3 com limitações reais: índices em memória, sem clustering e sem criptografia em repouso. Ideal para quem precisa de rastreamento de causa e confiança como propriedade de primeiro-nível; não é alternativa universal a PostgreSQL/Oracle.
README
A tradução para o português deste README ainda está na fila.
Centauri
A bi-temporal, causal, AI-first event database — in one file. Every fact knows its time, cause, source, and trustworthiness. Nothing is ever erased.
Free and open source under a PostgreSQL-style license. If you've spent years trapped in expensive database licensing — welcome home.
Why Centauri
Ordinary databases remember only the latest value. Centauri remembers the whole story, so it can answer four questions no ordinary database can:
- What was true at any moment? (
AS OF '2026-03-15') - What did we believe at any moment? (
AS KNOWN AT '2026-03-01'— the audit superpower) - WHY did it change? (
WHY— causes are first-class data) - Can we trust it — and did it actually land? (provenance + confidence on every fact;
PENDINGfinds changes that were sent but never applied)
Plus: CeQL (a query language where time, cause, and trust are syntax — with a plain-English helper), CePL stored procedures that are versioned and self-tracing, vector search, an embedded dashboard and full textbook, a tamper-evident hash chain, multitenant environments with snapshot cloning, read replicas, an MCP server so AI agents are first-class clients — and the Genesis Engine: describe your scenario in plain language and Centauri interviews you and builds the entire database, then stores that conversation as facts so the database forever remembers why it exists.
Quickstart (30 seconds)
Download one file from Releases, then:
centauri serve
Open http://localhost:7771 — the dashboard, guided tour, and CeQL textbook (/ceql) are inside the binary. Click 🏗 Build my database, describe what you're building, answer a few questions, done.
Or build from source (Go 1.22+):
go build -o centauri ./cmd/centauri
./centauri seed && ./centauri serve
Python: see sdk/python — zero dependencies, three lines to your first fact.
Commands
centauri serve HTTP/JSON API + dashboard (+ -token, -read-token)
centauri mcp Model Context Protocol server for AI agents
centauri follow live read-only replica of a primary
centauri backup consistent snapshot, chain-verified
centauri verify prove history is byte-for-byte intact
centauri seed synthetic retail demo data
Contributing — humans and AI together
We run an automated pipeline: your suggestion can become a pull request written by Claude.
- Open an issue using the Feature or Bug template.
- A maintainer reviews it and comments
@claudewith go-ahead. - Claude implements it, runs the tests, and opens a PR.
- CI must pass; a human merges.
See CONTRIBUTING.md. Hand-written PRs are equally welcome.
Honest status
v0.3: single-node engine with read replicas. In-memory indexes (your data must fit in RAM), no clustering/sharding, no cost-based optimizer, encryption-at-rest planned (v0.4 with crypto-erasure). The dashboard's ★ Why Centauri table states plainly where Oracle/Postgres/Mongo still win. Centauri's lane: the system of record for what happened, when, why, and how much to trust it — the flight recorder beside your operational stores.