Knowledge#Extraction#RAG

Morphik

Morphik Core is an open-source multimodal retrieval engine that embeds charts natively, 3,715+ GitHub stars, free under BSL 1.1 below $2,000/mo revenue.

Product facts

Commercial Suite
Official sitedev.morphik.ai
Pricing
Free tier + usage-based
Data checked
2026-10-06

Capabilities

  • Native format multi-modal extraction
  • Technical chart and schematic parsing
  • Native MCP integration
  • Cross-document Research Agent reports

Snapshot figures reflect the check date and may change over time.

Morphik is an open-source multimodal retrieval engine built by Morphik, Inc. Its GitHub repo morphik-org/morphik-core sits at 3,715 stars as of 2026-10-06 and is still being actively pushed (last commit October 5). The company’s main business has since pivoted to back-office automation for skilled nursing and senior living operators (AP, billing, nurse documentation), so the retrieval engine itself now ships as a standalone product at dev.morphik.ai — a trap worth noting, since the root domain morphik.ai now serves that unrelated senior-living SaaS page, not the knowledge base.

Core Purpose

Traditional RAG pipelines chunk documents into text, OCR them, and embed the pieces — a process that loses spatial relationships in charts, schematics, and multi-column technical specs, and frequently mangles tables. Morphik Core instead embeds each page’s raw content directly into its retrieval layer. A benchmark on Morphik’s own tech blog (evaluated 2025-07-08, using TLDC’s question set) put Morphik at 96% accuracy versus 67% for a custom RAG setup and 13% for dumping the same documents into OpenAI’s file search — numbers now over a year old, worth re-verifying on your own corpus. Beyond retrieval, it adds knowledge-graph visualization, a cross-document Research Agent that synthesizes multi-document reports, native MCP support (works with Claude, Open Web UI, or any MCP client), and a white-labelable web UI you can embed in your own product. Licensing is Business Source License 1.1: free for production use as long as the resulting monthly revenue for your organization stays under $2,000; past that threshold you need a commercial agreement, which is a step short of a pure OSI open-source license. Compared with the ontology-first OpenCrab, which structures documents into reusable ontology packs before querying, Morphik skips that modeling step and goes straight to multimodal vector retrieval.

Key Scenarios

  1. Teams with technical-document-heavy corpora: engineering drawings, datasheets, and research reports where chart and layout context must survive retrieval.
  2. MCP agent developers: wiring an external, graph-backed memory layer into Claude Code, Cursor, or any MCP client instead of hand-rolling a vector store.
  3. Product teams embedding knowledge search: white-labeling the retrieval UI directly inside their own application.

Trying it and its limits

Signing up at dev.morphik.ai drops you into the free tier: 200 pages/month, 3 Research Agent calls/month, and a shared GPU ingestion queue, no credit card required (checked 2026-10-06). Paid tiers run Pro at $59/month (2,000 pages, $0.03/page overage) and Team at $799/month (10 GB storage, unlimited queries, a dedicated instance); Enterprise is custom and adds SOC 2, HIPAA, and bring-your-own-cloud or on-prem deployment. Self-hosters can pull the GitHub repo directly, but read the BSL 1.1 revenue cap before putting it into production. Two concrete catches: old links can still land you on morphik.ai, which is now a completely different senior-living product; and the 96%-accuracy comparison dates to mid-2025, so validate it against your own dataset rather than taking it at face value.