Introducing TimeBF
Built for blazing-fast time-series analytics. A portable core,
designed for your hardware, your precision, and your AI workflows.
A familiar query. A different approach.
Explore the idea behind precision on demand.
The right work.
At the right resolution.
Time-series data has structure. TimeBF aims to turn that structure
into less data movement, less computation, and clearer answers.
Cache-conscious layouts. Vectorized kernels. Parallel batches. An execution engine designed around how modern processors actually work.
Ask for the precision you need. Hierarchical summaries guide refinement, with explicit error bounds as part of the query contract.
{
interval: 'last_24h',
context_budget: 2048,
precision: 'adaptive',
include: ['bounds', 'provenance']
}Planned tools make query precision and AI context budgets explicit. Retrieve compact temporal evidence, refine promising candidates, and give agents results with provenance. The goal: faster retrieval and less unnecessary context in model calls.
A focused, embeddable core designed to move between servers, laptops, and edge devices. Keep analytics close to the data and leave room for the rest of your system.
// Proposed browser interface import { TimeBF } from 'timebf/wasm'; const db = await TimeBF.open(); const answer = await db.query({ series: ['a', 'b'], error: 0.02 });
The roadmap brings the same core to notebooks, interactive applications, and the browser. Your data stays on your device.
Design targets for the TimeBF engine. See the roadmap for the planned implementation.
Start with compact summaries. Refine where the question needs detail. Return the answer together with its bounds.
Time-stamped streams.
Multiresolution state.
Your precision budget.
The result, with context.
Built for time-series data in general—from application metrics and market data to industrial signals and scientific measurements.
Interactive dashboards, trends, aggregates, and comparisons across historical windows. Spend computation where more precision changes the answer.
Let agents request time-series evidence through structured tools. Planned controls make accuracy, returned context, and provenance explicit before an AI call.
Explore approximate search over temporal summaries to select relevant windows before retrieving detail. A path toward faster RAG retrieval and smaller prompts, with exact refinement when needed.
Application directions under development. The interactive example above illustrates correlation; it does not define the scope of TimeBF.
TimeBF starts with a simple question: how much computation does a useful answer really need?
Created by Vinh Ngo, a Marie Skłodowska-Curie PhD researcher at Chalmers University of Technology, working on continuous and concurrent data summarization.
The project is ongoing, with promising early results. Funding and design partners will help develop the engine, validate performance across workloads, and bring the first product to developers.
An ongoing, privately developed project. The next steps turn promising research into a portable database, with reproducible evaluation and practical AI interfaces.
Temporal queries, aggregation, comparison, compact summaries, and explicit precision.
Develop execution kernels and validate speed, memory use, and accuracy across workloads.
A portable core, language bindings, and examples that run on devices and in the browser.
Claude tool use, context budgets, approximate retrieval, and provenance for time-series RAG.
Seeking funding and collaborators to bring TimeBF to life.
Let’s talk TimeBF