Kendr.org

Build, share, test, and rebuild hosted knowledge bases.

Kendr Cloud KBs expose the hosted RAG lifecycle behind a stable API: estimate, create, index, inspect artifacts, test retrieval, share access, rebuild, and delete. Every KB records its pipeline config, model map, vector dimensions, run history, retained artifacts, evaluations, and explainable retrieval scores.

Hosted vector DB Endpoint reference OpenAPI JSON

/api/kb/cloud Staged RAG pipeline Team sharing Explainable retrieval

Pipeline

The managed pipeline is Sources to Extraction JSON to Cleaning to Chunking to Embeddings to Kendr hosted vector DB to Retrieval to Reranking to Answer.

Prepare sources

Files, links, notes, and directory-like source manifests are normalized into page and block metadata.

Clean and chunk

Presets and regex rules run before semantic, word, or token chunking so each chunk remains useful on its own.

Embed and store

The active embedding provider, model, dimensions, and vector-store provider are saved on the KB record.

Retrieve and evaluate

Test queries return source chunks, vector score, lexical score, rerank score, final score, diagnostics, and answer metadata.

Configuration model

Configure the hosted pipeline with pipeline_config. Beginners can omit most fields and use Kendr defaults. Advanced callers can tune extraction, cleaning, chunking, embeddings, vector storage, retrieval, reranking, answer model, and artifact retention.

Area	Key fields	Why it matters
Artifacts	artifact_retention: none, preview, or full	Controls whether extraction, cleaning, chunking, and embedding summaries can be inspected later.
Cleaning	presets, ordered regex rules	Removes repeated headers, footers, empty lines, and noisy boilerplate before chunking.
Chunking	strategy, target_words, min_words, max_words, overlap_words	Determines chunk quality, recall, duplication, and reindex requirements when changed.
Embeddings	provider, model, dimensions, mode	Defines the vector dimensions and the model identity stored in the KB model map.
Vector store	provider: kendr-cloud, collection metadata, dimension metadata	Routes chunks into Kendr managed storage and tells clients when a dimension-changing rebuild is required.
Retrieval	top_k, score_threshold, hybrid_vector_weight, source_cap, mmr	Controls how many chunks are retrieved, how hybrid scores are balanced, and how repetitive sources are capped.
Reranker and answer	reranker.mode, answer_model.provider, answer_model.model	Makes it visible which model or fallback is responsible for ordering evidence and producing the final answer.

Endpoints

Method	Path	Purpose
GET	/api/kb/cloud	Lists hosted KBs visible to the caller.
POST	/api/kb/cloud/estimate	Estimates indexing, storage, and source-size credits before creation.
POST	/api/kb/cloud	Creates a KB, saves pipeline config and access policy, and queues indexing.
POST	/api/kb/cloud/{kb_id}/test	Runs retrieval and returns hits, vector score, lexical score, rerank score, diagnostics, and answer text.
POST	/api/kb/cloud/{kb_id}/rebuild	Queues a rebuild from a selected stage after config, source, chunking, or embedding changes.
POST	/api/kb/cloud/{kb_id}/access	Updates team, user, link, and source-access metadata.
GET	/api/kb/cloud/{kb_id}/pipeline/artifacts	Returns retained extraction, cleaning, chunking, embedding, and storage artifacts.
GET	/api/kb/cloud/{kb_id}/runs	Returns pipeline run history with status, timings, warnings, and errors.
GET	/api/kb/cloud/{kb_id}/evaluations	Returns previous test queries and selected retrieval hits.
DELETE	/api/kb/cloud/{kb_id}	Deletes the hosted KB, sources, chunks, artifacts, runs, evaluations, and access grants.

Example

This JavaScript example creates a team-visible KB, stores chunks in Kendr hosted vector storage, and runs a retrieval test with explainable scoring.

const kb = await client.createCloudKnowledgeBase({
  name: 'Support runbooks',
  sources: [{
    kind: 'file',
    label: 'Retention playbook',
    value: 'retention-playbook.md',
    content: 'Revenue retention improved after renewal handoff changes.'
  }],
  pipeline_config: {
    artifact_retention: 'preview',
    chunking: { strategy: 'semantic', target_words: 180, overlap_words: 30 },
    embeddings: {
      provider: 'kendr-managed',
      model: 'kendr-managed-hash-embedding-v1',
      dimensions: 128
    },
    vector_store: { provider: 'kendr-cloud' },
    retrieval: { algorithm: 'hybrid', top_k: 8, hybrid_vector_weight: 0.65 },
    reranker: { mode: 'heuristic' }
  },
  access_policy: { visibility: 'team', role: 'viewer' }
});

const result = await client.testCloudKnowledgeBase(kb.knowledge_base.id, {
  query: 'What changed retention?',
  limit: 4,
  include_answer: true
});

console.log(result.hits.map(hit => ({
  title: hit.title,
  vectorScore: hit.vector_score,
  lexicalScore: hit.lexical_score,
  rerankScore: hit.rerank_score,
  finalScore: hit.final_score
})));

Where to go next

Storage

Hosted Vector DB

Understand managed vector storage, dimensions, model maps, sharing, billing, and current integration boundaries.

Reference

Cloud KB endpoint table

Use the route map when implementing create, estimate, test, rebuild, access, artifact, and delete flows.