Search 64.6M knowledge entries or bring your own data. Tokenized with 3.8M whole words. No LLM embeddings. No inference costs.
// 1. Create an index — no embedding pipeline
await fetch("https://cold-api.coldstate.ai/v1/indexes", {
method: "POST",
headers: {
"Authorization": "Bearer cs_live_...",
"Content-Type": "application/json"
},
body: JSON.stringify({
name: "acme-docs",
domain_preset: "general",
documents: [
{ id: "doc_001", content: "Parental leave policy..." },
{ id: "doc_002", content: "Remote work guidelines..." }
]
})
});
// 2. Search — deterministic, Ψ-scored, sub-3s
const res = await fetch(
"https://cold-api.coldstate.ai/v1/indexes/idx_.../search",
{
method: "POST",
headers: { "Authorization": "Bearer cs_live_..." },
body: JSON.stringify({
query: "parental leave policy for remote employees",
limit: 5
})
}
).then(r => r.json());
// res.results[0].state: "CRYSTALLINE"
// res.results[0].score["Ψ"]: 0.94
// res.diagnostics.execution_time_ms: 847A real, read-only index running on the live API. Run a query, expand any result to see the per-term scoring, then verify the same query returns a byte-identical ranking every time.
A deterministic search API returns the same ranked results for the same query, every time. ColdState computes relevance from whole-word tokenization (a 3.8M-word vocabulary) and transparent scoring instead of LLM embeddings — so rankings are reproducible, explainable, and cacheable. There is no model inference, no temperature, and no drift between runs.
Vector search embeds your query with a neural model and finds approximate nearest neighbors — results can shift between model versions, and every query pays an inference cost. ColdState indexes documents once and scores them deterministically at query time: no embedding step, no approximate search, no GPU. The explain endpoint shows exactly why a document ranked where it did, token by token.
Yes. ColdState ships a Model Context Protocol (MCP) server with six read-only tools — search, list indexes, browse documents, explain scoring, global knowledge search, and domain listing — so Claude and other MCP-compatible AI assistants can search your indexes directly. It runs over stdio for desktop clients or HTTP for remote deployments.
Agents built on stochastic retrieval are hard to debug: the same prompt can fetch different context on every run. Deterministic retrieval makes agent behavior reproducible — evals are stable, failures replay exactly, and every ranking decision can be audited after the fact.
Alongside your own indexes, ColdState exposes a curated knowledge base of 64.6 million entries across 35 domains — science, medicine, technology, law, history, and more — searchable through the same deterministic API and MCP tools.
Yes. IaaS (Indexing as a Service) mode builds your index server-side and delivers it as a portable SQLite file you can query offline — no ongoing hosting relationship. Hosted mode keeps the index on ColdState with encrypted-at-rest storage.
Search our knowledge base or bring your own data. Get your API key and start in under a minute.
Get API Key