Sign in

logst: event intelligence

Concept-first event intelligence: multi‑day reasoning over time‑series, derived signals, and embedding‑aware context management.

Abstract

We focus on concepts over implementation: deriving meaningful signals from raw events, constructing compact contexts from embeddings, and enabling models to reason over multi‑day time‑series to produce actionable insights.

Architecture

  • Ingestion: Producers send events; writes are queued to decouple capture from compute.
  • Event store: Time‑series persistence optimized for aggregation and backfills.
  • Derived signals: Rollups, cohorts, funnels, and semantic features computed incrementally.
  • Embedding index: High‑cardinality and textual facets embedded for retrieval.
  • Reasoning: Models reason over multi‑day windows using compact contexts built from signals and retrieval.
  • Delivery: Insights routed to where work happens; channels are pluggable.

Derived Signals & Embeddings

Signals are computed continuously (and backfilled) to transform raw events into features suitable for retrieval and reasoning.

  • Sessionization and time‑bucket rollups (rates, latencies, ratios)
  • Funnels and cohorts with incremental maintenance
  • Feature extraction from strings (templates, entities, categories)
  • Embeddings for high‑cardinality/text facets; vector index for retrieval
  • Context builders combine top‑K retrieval with guardrailed aggregates

Data Model & API

DATA MODEL & API
- Event: { name: string, properties: Record<string, scalar>, organizationId, accessTokenId?, createdAt }
- POST /api/v1/events: single or batch payloads
- GET /api/v1/events: filters by name substring, createdAt range, limit/offset
- Properties accept only scalars (string | number | boolean | null) for safety and portability

Reasoning & Insights

A reasoning model operates over multi‑day windows using compact contexts built from derived signals and retrieved embeddings. It explains changes (spikes, drifts, regressions), cites underlying aggregates, and adapts to horizon‑length and confidence thresholds.

API

Send events via HTTP or the official JavaScript SDK.

curl -X POST https://logst.nendlabs.com/api/v1/events   -H "Authorization: Bearer $LOGST_API_KEY"   -H "Content-Type: application/json"   -d '{
    "event": "user.login",
    "user_id": "12345"
  }'