a23one

polyembed

Changelog

Redis Embedding Cache Reliability

More reliable cache persistence and memory-aware eviction for embedding workloads

Redis-backed embedding caches now handle background saves and reloads more reliably, reducing risk of out-of-memory failures. Under cache pressure, cache evicts least-recently-used entries while preserving frequently updated usage counters.

  • fixed

    Improved Redis-backed embedding-cache reliability and prevented out-of-memory failures during background saves and reloads.

  • changed

    Evicted least-recently-used embedding entries under cache pressure while retaining frequently updated usage counters.

Smarter Long-Text Embeddings

Tokenizer-aware chunking, fuller long-text coverage, accurate vector weighting and usage reporting

Long inputs are now split into model-sized chunks rather than being truncated. Chunking uses selected embedding model’s tokenizer, preserves representative content across oversized inputs, weights combined vectors by actual token counts, and records embedded-token usage more accurately.

  • added

    Long text was split into paragraph- and model-sized chunks instead of being truncated to one embedding window

  • changed

    Embedding chunk sizes and limits used the selected model’s tokenizer, improving results for multilingual text

  • changed

    Text over token allowances retained whole chunks across the input rather than truncating mainly by character count

  • changed

    Multi-chunk embeddings were weighted using actual token counts for more representative vectors

  • changed

    Embedding usage tracking recorded actual embedded tokens for more accurate quota reporting

Security and Embedding Enhancements

Per‑team salt, deterministic transforms, improved embedding logic, version bump

We strengthened cryptographic protection of embeddings and updated the project version to reflect these changes.

  • changed

    Added optional per‑team salt to key‑derivation for secure embeddings

  • changed

    Bumped project version to reflect new features and security updates

Token Budget Management

max_tokens parameter, team token settings, UI updates, migrations

Users can now set and monitor token budgets per team, with new API options and dashboard controls.

  • added

    Introduced max_tokens request parameter for embedding endpoints

Python SDK Release

SDK library, README, docs update, API‑key rate‑limiting

We released a new Python SDK to simplify embedding integration, added documentation and a smoother API‑key throttling experience.

  • added

    Released the PolyEmbed Python SDK client library

Usage Analytics & Team Dashboard Visualisation

Usage tracking subsystem, interactive charts on team dashboard

Implemented in‑memory request counter, periodic persistence, and visualised per‑key usage on the team dashboard.

  • added

    usage tracking module recording per‑key request counts in memory

Documentation, Models, and API Enhancements

Docs & Models pages, navigation updates, shielded API mode, model validation, health simplification, profile editing, team dashboard charting

Added public documentation and model reference pages, improved navigation, introduced shielded embedding mode, simplified health check, added profile editing, and enhanced team dashboard with usage charts.

  • changed

    navigation bar updated to include Docs and Models links

Core Web Interface & Authentication

Landing page, navigation, login, dashboard, team management, embedding API, health check, scaffolding

Introduced the main website with hero section, authentication flows, a protected dashboard, team API key management, and the first embedding endpoint.

  • added

    landing page with hero, metrics, capability grid, code snippets, pricing, logos, CTA