The Ultimate Discovery Engine.
Stop paying Algolia thousands of dollars a month for what you already have. Aeion Search is a native OS module delivering enterprise-grade federated discovery — full-text via PostgreSQL `ts_rank` + `ts_headline`, semantic via `pgvector` HNSW indexes, hybrid via Reciprocal Rank Fusion. Zero sync lag, zero external dependency, zero platform markup. You bring your own embedding provider key.
Intelligence Built into the Database
Semantic Vector Search
Keyword matching is dead. When a customer searches for "something to wear to a summer wedding," Aeion Search uses AI embeddings to return elegant dresses — even if "wedding" isn't in the description. It searches by *meaning*, not characters.
Hybrid RRF Fusion
The best of both worlds. Runs both a full-text Postgres query (for exact SKU matches) and a semantic vector query (for concept matching) in parallel, then blends results via Reciprocal Rank Fusion for mathematically optimal relevance.
Federated Multi-Module Discovery
Search everything at once. Press `Cmd+K` anywhere in Aeion OS. One query federates across CRM Contacts, Commerce Orders, Helpdesk Tickets, Workspace Pages, Files, Forge Issues — natively.
Zero-Sync Architecture
External search tools require brittle API syncing — update a product price, wait 5-30min for Algolia to index. Aeion Search runs natively on Postgres — the millisecond a record commits, it's globally searchable.
Dynamic Text Highlighting
Uses `ts_headline` to extract preview snippets with matching terms dynamically highlighted. Customers see relevance instantly, not just a result list.
Typo-Tolerance + Stemming
Built-in PostgreSQL linguistic capabilities — "shos" → "shoes", "running" → root "run". Multilingual stemming for non-English locales.
What Ships in Aeion Search — Core Engine + Module Telemetry
Core search engine
One unified, federated search behind a single API call — full-text, vector, and hybrid all live in the same query, so you're never stitching together two systems just to answer "find me the thing that matches this."
Module setup
Every collection you want searchable gets registered once — telemetry, analytics, and a `/search` REST endpoint come with it automatically, no per-collection plumbing to write.
Built-in analytics
See exactly what your users are searching for, which queries return zero results (your product gaps, in their own words), click-through rates, and latency at P50/P95/P99 — no separate analytics tool bolted on.
pgvector HNSW indexes
Sub-millisecond approximate nearest-neighbor lookups over millions of embeddings via Hierarchical Navigable Small World graphs — semantic search that stays fast as your catalog grows, not just at demo size.
ts_headline highlighter
Every result comes back with the matching terms already highlighted inline, so customers see *why* a result matched instead of scanning a wall of text to find it themselves.
Hybrid RRF fusion
Full-text and vector rankings are mathematically blended via Reciprocal Rank Fusion (`score(d) = Σ 1 / (k + rank_i(d))`, k=60 by default, tunable per query) — so a document doesn't have to win on keywords alone or meaning alone to rank well; it can win on either, or both.
Embedding pipeline
New and updated records get embedded automatically in the background the moment they're saved — bring your own embedding provider through the AI gateway, and there's no manual re-indexing job to babysit.
The Math: Why Paying for Algolia Is Insane
External search SaaS charges you to search your own data.
Federated Search — One Query, Every Module
The Cmd+K command palette runs a federated query across every search-enabled collection.
Hybrid RRF — Why It's Smarter Than Either Alone
Reciprocal Rank Fusion combines ranked result sets from multiple retrieval methods.
CTO / search-buyer questions
Those are external services — your data syncs to them, you pay per query, there's 5-30min lag between save and searchable. Aeion Search runs *inside* your Postgres — the same database your records live in. Save a product, it's searchable in the same transaction. No sync, no cost-per-query, no lock-in.
Same reason. Pinecone is a separate service with its own billing, latency hop, and sync requirement. pgvector runs in your Postgres — vectors live in the same row as the rest of the record. Cosine similarity is a single SQL query. HNSW indexes give sub-ms ANN over millions.
You bring your own provider via /platform/ai. OpenAI's `text-embedding-3-small` is $0.02 per 1M tokens. 100K typical product descriptions ≈ $0.10. 1M ≈ $1. Zero platform markup.
Re-embed via background job. Schema supports multiple embedding columns side-by-side (A/B test models). Embedding refresh is automatic on every record save.
HNSW is `O(log n)` for ANN queries — typical 0.2-2ms over 1M vectors on modest hardware. Build cost ~30s per 100K vectors; queries stay sub-ms forever after.
Yes. Every search automatically inherits the user's tenant and permission scope. Federated queries apply the same access rules as a regular collection read. Cross-tenant leakage impossible by construction.
Our built-in analytics log captures every query: text, user, timestamp, latency, result count, clicked result, scroll depth. The dashboard and BI-tool export are included with the module.
Yes via SQL boost expressions. Add `+ (age_days * -0.01)` for recency, `+ (purchase_count * 0.1)` for popularity. RRF blends with relevance. Configurable per-collection.