Overview

Aletheia Overview

Aletheia is memory infrastructure for agents and applications that need more than a flat vector index.

What it is built to do

Fuse semantic search with lexical search and reranking.
Keep fresh context visible through time-aware ranking.
Track fact supersession so stale claims stop competing with updated truth.
Support the same developer workflow locally and in the cloud.
Expose an API surface that can power SDKs, benchmarks, and production apps.

Why teams reach for it

Most memory systems can retrieve. Fewer can update beliefs cleanly, handle mixed fact and episodic recall, or ship a local-first workflow that is still production-shaped.

Aletheia is designed for that gap.

Core principles

Hybrid retrieval

ANN alone is not enough. Aletheia combines semantic search, BM25-style lexical search, and reranking so exact phrases and latent meaning both matter.

Memory classes

Not every memory should decay or rank the same way. Facts, summaries, preferences, and episodic traces can behave differently.

Local-first development

Developers should be able to run the same engine locally as a sidecar before they point the SDK at a hosted environment.

Operational readiness

Signed binaries, scoped auth, release manifests, and compatibility-aware SDKs are treated as part of the product surface.

Repositories

Public repositories for platform, SDK, and model adapter surfaces:

Core engine repository is private: Aletheia (https://github.com/SharjeelAbbas014/Aletheia).