Platform

Resources & Knowledge

Centralize documents, embeddings, and structured data so workflows can reason over trusted knowledge sources.

Knowledge source types

Documents and files

PDF, DOCX, Markdown, and HTML with metadata, tags, and owners.

Structured datasets

Tables, CSV, SQL views, and analytics exports for retrieval.

External APIs

Live context from SaaS tools, CRMs, and internal services.

MCP resources

Attach MCP servers and resource URIs for managed retrieval.

Ingestion pipeline

  1. Collect sources, assign owners, and define retention policies.
  2. Normalize formats, extract metadata, and chunk content.
  3. Generate embeddings and index into vector or hybrid stores.
  4. Validate retrieval quality with test queries and scoring.
  5. Schedule refresh or event-driven reindexing.

Session memory

Short-lived context captured per run or conversation.

Workspace memory

Persistent facts shared across workflows and agents.

Entity memory

Customer or project profiles with versioned updates.

Governance and relevance

Access controls on sources with per-block retrieval filters.
Citations and provenance for knowledge-backed outputs.
Freshness signals and expiration windows for time-sensitive data.
PII detection and redaction before indexing.
Similarity thresholds to prevent irrelevant context injection.

Chunking strategy

Tune chunk size and overlap to maximize recall.

Metadata filters

Filter by tags, owner, region, or language.

Ranking and rerankers

Apply LLM or heuristic reranking for relevance.