AIONBD

Edge-First Vector AI Database for Production

AIONBD is a Vector AI platform built to run production retrieval systems on edge nodes and constrained infrastructure.

It combines deterministic runtime guardrails, explicit durability profiles, and high-throughput Vector AI Search across exact, IVF, and auto modes.

Open Docs Hub Read Vector AI Use Case Open GitHub repository

Project Stage

Public preview.

License Status

Evaluation license. Production use is currently prohibited.

Search Modes

exact, ivf, auto.

Target Environments

Edge nodes, constrained machines, intermittent networks.

AIONBD architecture - Vector AI Edge Database flow
High-level architecture flow for ingestion, retrieval, persistence, and observability.

Overview: Why AIONBD

AIONBD is a Vector AI database focused on production control. The goal is reliable Vector AI Search and AI retrieval in environments where memory, I/O, and operational windows are limited.

  • Single Rust server runtime with local-first persistence.
  • Deterministic guardrails for memory, payload size, request concurrency, and query fanout.
  • Explicit durability profiles with safe defaults and auditable throughput modes.
  • Metrics, runbooks, soak, and chaos workflows for reliability-driven operations.
  • Deployment-ready packaging across Docker, systemd, Kubernetes, and Helm.

System Architecture

Core Modules

  • core/: vector math, collections, persistence primitives.
  • server/: HTTP API, auth, quotas, runtime controls.
  • bench/: reproducible benchmark and regression scenarios.
  • sdk/: JavaScript, Python, Go, and Rust clients.

Reliability Model

  • /live and /ready health endpoints.
  • WAL append and in-memory apply model with snapshots and incremental compaction.
  • Safe default durability profile: AIONBD_WAL_SYNC_ON_WRITE=true. 
Client request
  -> HTTP handlers (server)
  -> Collection engine (in-memory)
  -> WAL + snapshot/incremental persistence
  -> Metrics endpoints (JSON + Prometheus)

API Surface

Health and Metrics

  • GET /live
  • GET /ready
  • GET /metrics
  • GET /metrics/prometheus

Collections and Points

  • POST /collections
  • GET /collections
  • PUT /collections/:name/points/:id
  • POST /collections/:name/points

Vector AI Search

  • POST /collections/:name/search
  • POST /collections/:name/search/topk
  • POST /collections/:name/search/topk/batch
  • Modes: exact, ivf, auto with target recall

Performance Positioning

Numbers below summarize results from docs/optimizations_and_benchmarks.md. Treat them as reproducible local positioning. Validate on your own hardware before external claims.

Internal Benchmark Scenario Key Result
Core dot scan (10k x d128) 6,945,199.54 QPS, p95 1.577 ms
Core l2 scan (10k x d128) 7,030,746.86 QPS, p95 1.660 ms
Collection list_points 50k (cursor vs offset) 713,509.28 pages/s vs 11,823.63 pages/s (60.35x)
Search quality 20k d128 recall@10 = 1.0 in exact and auto; auto p95 0.620 ms
Persistence write path 4096 writes d64 sync every N: 9,189.78 QPS (5.45x vs sync each write)
Open-source 500k (persistent, aionbd-batch-size=128) AIONBD auto: 94.78 QPS, p95 10.55 ms; Qdrant exact: 6.80 QPS, p95 222.41 ms

Operations and SLO Baseline

Suggested SLO Baseline

  • Availability: aionbd_ready == 1 at 99.9% over 30 days.
  • API error budget: 5xx ratio below 1% over 5 minutes.
  • In-flight pressure: typically below 80% of max concurrency.
  • IVF fallback ratio target: below 25% over 10 minutes.
  • Checkpoint and WAL backlog must stay bounded.

High-value Metrics

  • aionbd_http_requests_in_flight
  • aionbd_persistence_wal_size_bytes
  • aionbd_persistence_checkpoint_error_total
  • aionbd_l2_index_cache_hit_ratio
  • aionbd_search_ivf_fallback_exact_total
  • aionbd_auth_failures_total

Security and Multi-Tenancy

  • TLS for non-local deployments with AIONBD_TLS_ENABLED=true and managed certificate rotation.
  • Authentication modes: api_key, bearer_token, jwt, and mixed policies.
  • Tenant protection through quotas and rate limits.
  • Request surface hardening through body, concurrency, and query limit controls.

Deployment and Packaging

AIONBD ships with deployment artifacts for release binaries, OCI images, Docker Compose, systemd, Kubernetes, and Helm.

Build release binary

cargo build --release -p aionbd-server

Compose production package

cp ops/deploy/.env.prod.example ops/deploy/.env.prod

docker compose -f ops/deploy/docker-compose.prod.yml \
  --env-file ops/deploy/.env.prod up -d

Documentation Map

Main resources and support pages for deeper Vector AI and Vector AI Search coverage.

Document What it covers
README.md Project scope, quick start, API overview, configuration.
docs/platform_guide.md Platform scope, architecture model, and topologies.
docs/optimizations_and_benchmarks.md Benchmark matrix, optimization workflow, reproducibility.
docs/performance_tuning.md Runtime tuning knobs for latency, throughput, durability.
docs/production_sizing.md Memory, disk, and capacity planning baselines.

Internal pages: Docs Hub, Blog Hub, Vector AI use case, Vector AI Search at the edge.

FAQ

What is AIONBD in a Vector AI context?

AIONBD is a Vector AI database for production retrieval systems that require deterministic limits and predictable operational behavior.

Can it run Vector AI Search workloads?

Yes. Vector AI Search is supported through exact, IVF, and auto modes with top-k and batch APIs.

How do durability and throughput tradeoffs work?

Safe durability is default with sync-on-write. Throughput-focused profiles can be enabled explicitly with documented acknowledged write-loss risk.

Is AIONBD currently production-licensed?

Not yet. The codebase is production-oriented, but the current license remains evaluation-only.

Authority and Off-Page Signals

Ranking on competitive terms like Vector AI also depends on external authority. The technical site is optimized, but backlinks and community references are still required.

  • Publish benchmark and architecture posts on Dev.to and Medium with links to this site.
  • Share release notes and benchmark updates on Hacker News and relevant Reddit communities.
  • Publish short technical posts from the GitHub repository and link back to the Docs Hub.
  • Ask SDK users and integration partners to reference AIONBD in technical writeups.