{
  "entity": "usr_42",
  "decay": 0.85,
  "hops": 3
}
Feather DB v0.5.0

The ContextEdition

Embedded vector database + living context engine. Part of Hawky.ai. Zero-server, file-based, powered by Rust and C++.

Explore the Docs
View on GitHub
01 / Architecture

Built natively for
living context.

Performance marketing fails due to missing context. We solve this at the database layer.

Multimodal
Pockets

Each modality gets an independent HNSW index. A single entity ID holds text, visual, and audio vectors simultaneously.

db.add(id=42, vec=text_vec, modality="text")
db.add(id=42, vec=img_vec, modality="visual")

Context Graph

Typed, weighted edges. Link nodes natively within the vector store.

db.link(from=1, to=2)

Context
Chain.

One call combines semantic vector search with n-hop BFS graph traversal. Expand search results effortlessly.

res = db.context_chain( query=vec_q, k=5, # seed vector searchhops=2, # BFS expansionmodality="text")
02 / Adaptive Memory

Living Context & Adaptive Decay.

Records accessed more frequently resist temporal decay. The engine mathematically blends vector similarity, graph hops, age, and stickiness. Older information gives way to newer, more relevant context automatically.

Temporal Decay

Vectors naturally decay over time based on a configurable half-life. Your agents maintain relevance.

Memory Stickiness

Records accessed more frequently build stickiness using a logarithmic scale, causing them to resist age-based decay natively.

Dynamic Scoring

Score = similarity × hop_decay × importance × stickiness

03 / Multitenant

Powerful Metadata &
Filtered Search.

Generic partitioning for multi-tenant, multi-domain use. Filter by namespace, entity, and arbitrary attributes natively.

from feather_db import FilterBuilder f = FilterBuilder() \ .namespace("nike") \ .entity("user_8821") \ .attribute("channel", "instagram") \ .importance_gte(0.5) \ .build() results = db.search(query_vec, k=10, filter=f)
04 / Performance

Developer
Experience.

Extremely Fast Tooling

A high-performance Rust CLI enables lightning-fast data ingestion and vector debugging straight from the terminal.

# Add a record to the database
feather add --db my.feather --id 1 --vec "0.1,0.2"
# Search using the Rust binary
feather search --db my.feather --vec "0.1" --k 5
Install CLI →

SIMD & File Format

SIMD (AVX2/AVX512) runtime optimizations ensure lightning fast similarity computation in C++. The custom binary .feather flat format leverages zero-copy reading footprint.

C++17 Core Engine
Shared Metadata Store
Custom Flat File Format (v5)
05 / Scale

Trusted by
builders.