GENYS

GENYS · Memory

Semantic Search

Find the right truth fast. GENYS searches Capsules, Documents, and Events with hybrid retrieval, policy‑aware filters, and reranking—so answers are relevant, scoped, and safe.

Request Access Read the Docs

GENYS powers DesignAdvertise.ai campaigns today. This page shows how its search layer grounds generation and analytics across industries.

Simple

  • Ask in natural language—get answers from your approved memory and docs.
  • Filter by account, app, time, or tag to narrow instantly.
  • Results are ranked for usefulness, not just keyword hits.
  • Only what you’re allowed to see shows up. Everything is auditable.

Technical

  • Hybrid retrieval: dense vectors + BM25 + metadata filters over Capsules/Docs/Events.
  • Policy‑aware query planner applies scope (tenant/app/user), retention, and legal holds before ranking.
  • Reranking uses cross‑encoder signals + recency/authority weights; optional domain‑specific scorers.
  • Answers can cite sources and feed RAG++ context for generation with guardrails.

Try it (demo)

Toggle ELI5 vs Expert and scope by tenant. Results are mocked to demonstrate UX.

Scope:
Decision log · Q3 LaunchCapsulecap_01HY · v7

Approved tone = bold-power; target CPL ≤ $35; channels: Meta + LinkedIn; next step: retarget legal vertical.

Campaign brief · Q3 LaunchDocdoc_09ab

Budget: $45k; timeline: July–Sept; ICP: SMB legal. Owner: Growth.

Event · followup_sentEventevt_01HZ

Email sequence A/B sent; CTR 6.2%; next step scheduled.

Indexing pipeline

  • Normalize
    Chunk docs; extract entities; standardize fields across Capsules/Docs/Events.
  • Embed + Lexical
    Compute dense vectors and store term indexes (BM25) for hybrid search.
  • Metadata
    Attach scope (tenant/app/user), tags, createdAt, authority, and TTL.
  • Provenance
    Record source, actor, signature; enable audit trails.

Query pipeline

  1. Parse intent (question vs lookup vs summarize).
  2. Plan stores to search (Capsules/Docs/Events) + filters from scope, tags, and time.
  3. Retrieve via dense + lexical; merge candidates.
  4. Rerank with cross‑encoder + recency/authority weights.
  5. Cite sources; optional compaction for RAG++.

Quickstart

Search first, then (optionally) generate with guardrails using the retrieved context.

// 1) Semantic search with scope + filters
const results = await genys.search({
  query: 'What did we decide about Q3 Launch?',
  stores: ['capsules','docs','events'],
  scope: { tenantId, app: 'designadvertise', user: currentUserId },
  filters: { tags: ['campaign:Q3'], since: '2025-06-01' },
  options: { k: 12, hybrid: true, rerank: true, cite: true }
});

// 2) Use results to ground generation
const answer = await genys.generate({
  subject: currentUserId,
  input: 'Summarize the Q3 Launch decisions and next steps.',
  context: results, // compacted
  guards: ['toneguard:brand_voice_prime','facts:capsule_consistency']
});

Governance & Privacy

Scope‑first search (RBAC): tenant/app/user enforced at query plan time
Retention & legal holds respected by the planner and filters
Audit entries for queries (actor, scope, time, reason)
Right‑to‑be‑forgotten removes data from indexes and adds a redaction note
Model‑agnostic: exportable data; bring your own reranker if desired
SIEM streaming for query logs (optional)

FAQ

How is this different from plain vector search?

Hybrid retrieval adds lexical signals and strict policy filters. Reranking uses cross‑encoders and authority/recency weights so you get the right truth, not just the nearest text.

Can we limit search to an account or time range?

Yes. Use scope (tenant/app/user) and filters (tags, since/until). Legal holds and retention are honored automatically.

Does search feed generation?

Yes. Results compact into a context bundle for RAG++ generation with ToneGuard and Brand Vector Hash checks.

Get started Explore the docs