If the server was started with a token, paste it here. Stored only in this browser.

Use a zero-dependency client, plain HTTP, or MCP. Insert = update; reads, time-travel, context, and CDC are one call each.

from centauri import Centauri
db = Centauri(token="…")
db.add("toy:robot", {"price_cents": 500})
db.context("toy:robot")            # the agent bundle
db.run("reprice", item="toy:robot", pct=90)

c := centauri.New(url, centauri.WithToken("…"))
c.Add("toy:robot", map[string]any{"price_cents": 500})
c.Query("FACTS OF item:* WHERE region='EU'")
c.Run("reprice", map[string]any{"item":"toy:robot","pct":90})

const db = new Centauri(url, {token:"…"});
await db.add("toy:robot", {price_cents: 500});
await db.query("FACTS OF toy:robot WHY");

curl -s URL/v1/query -H 'Authorization: Bearer …' \
     -d '{"q":"FACTS OF toy:robot WHY"}'

{ "mcpServers": { "centauri": { "command":"centauri", "args":["mcp"] } } }

PROCEDURE reprice(item, pct)
  LET cur = FIRST FACTS OF ${item}
  WHEN cur IS MISSING: FAIL 'unknown ${item}'
  LET newp = cur.price_cents * pct / 100
  PUT ${item} SET price_cents=${newp}
  RETURN newp
END
-- then:  RUN reprice WITH item='toy:robot', pct=90

Loads a curated dataset across retail, healthcare, education & construction into a dedicated demo database — planted with bi-temporal corrections, causal chains, pending "wedges", cross-source disagreements, an AI enrichment, schemas and aggregates. It never touches your real data.

"Clear" drops the entire demo database file — consistent with Centauri's "nothing is ever erased" (that rule protects facts inside a live log, not a disposable example database).

Upload an image or PDF (e.g. an electrical drawing). Centauri stores the blob locally (content-addressed, not in the log), renders PDF pages, then a vision model you register describes it and embeds it into the vector index — searchable with no external store.

Defaults target Ollama (ollama pull gemma3:12b — the same vision model the AI presets use). One click writes model:vision (+ an embedder) as facts.

Then search them: SEARCH 'E-3 200A panel' · FACTS OF asset:* · SIMILAR TO <asset>

Registering also powers semantic SEARCH (embeds your query) and ASK '…' (plain-English answers from your data, with citations) — they use this embedder + chat model. See docs/ai-features.md.