impera 0.1.0

Universal vector database client — query across multiple vector databases with different embedding models

Impera: Universal Vector Database Client

A universal client for querying across multiple vector databases that use different embedding models, returning a unified result to the caller.

The Problem

Vector databases are built around a single embedding model. When an application spans multiple vector databases — whether by design or circumstance — there is no standard way to query across them. Each database lives in an incompatible vector space, requiring the caller to manage embedding, querying, and result merging manually.

What This Is

A universal client that, given an intention as input:

1. Identifies which embedding model each target database uses
2. Embeds the input accordingly, once per model
3. Queries each database in its own embedding space
4. Returns a normalized, reconcilable result set to the caller

The loop, analysis, and reconciliation logic belong to the calling application. This client handles the embedding and search layer only.

What This Is Not

• A vector database
• A reconciliation or ranking engine
• An agent or reasoning system
• A wrapper around any single vector database or embedding provider

Specifications

This project is specification-first. The client behavior is defined by a set of open, versioned, machine-readable specs before any implementation. See spec/SPEC.md for how specs are written and governed.

The v1 spec set covers:

Spec	Purpose
Registry	Maps each source to its embedding model and access method
Query	Format for an intention passed to the client
Result	Normalized schema returned to the caller
Auth	Credentials and access configuration per source
Error	Error types and fallback behavior
Versioning	How embedding model version changes are declared and handled

Status

Specification in progress. No implementation yet.

Contributing

Contributions to the specs are welcome via PR. See spec/SPEC.md for the process.

License

Apache 2.0