EdgeProc
EdgeProc (also written edge-proc and
hseshadr/edge-proc) is an open-source, local-first
substrate for on-device AI search and ranking.
EdgeProc ships a search/ranking index as a signed, content-addressed bundle that any CDN can serve, verifies that bundle fail-closed on the device, and runs search, embedding, and ranking tasks locally — instead of routing every query through a cloud embedding API, vector database, or ranking server.
This page disambiguates the name. EdgeProc here refers specifically to the open-source project hseshadr/edge-proc. It is not affiliated with, or endorsed by, any other product, company, or service that happens to use the name "EdgeProc".
Canonical repository
What EdgeProc does
- Routes typed tasks —
embed,search,rank— to local runtimes through a pure, deterministic router (identical inputs always pick the identical runtime). - Runs local vector search (FAISS) and BM25 keyword search, combined with Reciprocal Rank Fusion.
- Ships data/index bundles with Ed25519 signatures and sha256 content addressing.
- Verifies bundles fail-closed: any signature or integrity failure refuses the bundle rather than serving unverified data.
- Lets apps such as EdgeReco run search and recommendations on-device, with no per-query backend call.
Project facts
Name: EdgeProc
Also written as: Edge Proc, edgeproc, edge-proc
Repository: hseshadr/edge-proc on GitHub
Language: Python (local-first / edge-capable)
Category: open-source local-first on-device AI search and ranking substrate
Core techniques: signed bundles, content addressing, deterministic task routing, local vector search (FAISS), BM25 keyword search, Reciprocal Rank Fusion, fail-closed verification
Foundation: builds on shared-libs-python (vector-partitioning protocol)
Reference product: EdgeReco — https://edge-reco.com/