Seek.js

Build-time extraction + hybrid search for static docs and websites.
Seek.js converts shipped content into queryable index artifacts, then runs lexical + vector retrieval with Orama.

TL;DR

• Input: static build output (dist, build, out) or local rendered output.
• Output: normalized manifest + packed search index.
• Runtime: browser-first search, optional edge AI answer layer.
• Goal: remove vector DB ops burden for docs and landing sites.

Quickstart (5 min)

1. Build site normally.
2. Run Seek extractor on output directory.
3. Compile extracted sections into index artifact.
4. Load index in client search UI.

Example shape:

npm run build
seek extract --input ./dist --url-base https://example.com
seek compile --manifest ./seek.manifest.json --out ./public/search_index.msp

Core Pipeline

flowchart LR
    siteBuild[StaticSiteBuild] --> extractor[Extractor]
    extractor --> manifest[SeekManifest]
    manifest --> compiler[Compiler]
    compiler --> indexFile[search_index.msp]
    indexFile --> clientSearch[ClientHybridSearch]
    clientSearch --> optionalAi[OptionalEdgeAiSummary]

Loading

Package Roles

• @seekjs/parser: HTML/content extraction and section normalization.
• @seekjs/compiler: embedding + Orama-ready index compilation.
• @seekjs/client: browser hydration/cache/search primitives.
• @seekjs/ai-edge: optional answer streaming over retrieved chunks.

Repo Map

• [docs/README.md](docs/README.md): user-facing docs boundary and guide index.
• [specs/README.md](specs/README.md): implementation contracts.
• [specs/extractor/README.md](specs/extractor/README.md): extractor spec entrypoint.
• [research/README.md](research/README.md): rationale, tradeoffs, experiments.

Read Order

1. This file.
2. [specs/extractor/01-hybrid-extraction-architecture.md](specs/extractor/01-hybrid-extraction-architecture.md)
3. [specs/extractor/02-seek-manifest-schema.md](specs/extractor/02-seek-manifest-schema.md)
4. [specs/extractor/03-extractor-compiler-contract.md](specs/extractor/03-extractor-compiler-contract.md)
5. [specs/extractor/04-probe-and-pivot-strategy.md](specs/extractor/04-probe-and-pivot-strategy.md)
6. [specs/extractor/05-source-adapter-contract.md](specs/extractor/05-source-adapter-contract.md)
7. [specs/extractor/06-route-discovery.md](specs/extractor/06-route-discovery.md)
8. [specs/extractor/07-chunking-strategy.md](specs/extractor/07-chunking-strategy.md)
9. [specs/extractor/08-performance-budgets.md](specs/extractor/08-performance-budgets.md)

Current Status

• Stage: early architecture + contract hardening.
• Primary focus: extractor correctness, URL fidelity, manifest stability.
• Public docs polish and broad integrations follow after contract freeze.

Internal Doc Rules

Internal instructions stay outside publishable docs/ folder.

• Public docs: docs/
• Contracts/specs: specs/
• Research/rationale: research/

Canonical Terms

• Seek Manifest
• Probe and Pivot
• Local Render Fetch
• static artifact parsing

Spec Status

• Draft: early, not safe to implement against.
• Proposed: review-ready, not final.
• Accepted: implementation source of truth.
• Deprecated: obsolete, kept for history.

Quality Gates

A list of quality gates for the AI agent:

• Root README.md <= 220 lines.
• Spec intro <= 80 lines before first normative section.
• Checklist docs <= 250 lines.
• Keep one source-of-truth per concept; other files link out.