Skip to content
← Tools / Skills
ToolMorphik/10 trustVerified

Morphik

Open-source multimodal RAG that reads the images inside your PDFs, with knowledge graphs and KV caching.

Visit official site →

About

Morphik is an open-source RAG infrastructure project aimed at the documents that break naive chunk-and-embed pipelines: technical PDFs with tables, plots, and diagrams. It uses ColPali-style visual embeddings so search understands page images directly, extracts entities and relations into knowledge graphs for cross-document questions, and adds persistent KV caching so repeated queries over long documents skip redundant attention computation. A natural-language rules engine pulls structured metadata at ingest, and MCP support lets agents query it directly.

The core repo holds about 3,700 GitHub stars with commits days old, and the project states MIT licensing. A hosted cloud with a free tier sits alongside self-hosting via Docker Compose. The Show HN thread that introduced it framed the origin story well: pharma testers asking which EGFR inhibitors showed 30% tumor reduction across reports, and getting back the right tables instead of text snippets.

What it can do

  • ColPali-style multimodal document search
  • Knowledge-graph extraction for cross-document QA
  • Natural-language rules engine for metadata extraction
  • Persistent KV caching for long documents
  • MCP server for agent access; self-hostable

Setup

  1. Self-host: git clone https://github.com/morphik-org/morphik-core, then python quick_setup.py && python start_server.py; or sign up at morphik.ai for the hosted free tier.
  2. Ingest documents (PDFs, images, video) via the API or SDK.
  3. Query with natural language; toggle knowledge-graph mode for cross-document questions.
  4. Create persistent KV caches for long documents you query repeatedly.

Prerequisites

Docker (self-host) or a morphik.ai account; documents to ingest.

Use cases

ragmultimodal-searchknowledge-graphsopen-sourcedocument-ai

Works with

Trust score — 7/10

/10 trustVerified

Scored 7/10 (verified tier, metrics-v1). Evidence: 3,716 GitHub stars (GitHub API), pushed 2026-10-14, not archived; MIT license stated in Show HN launch; morphik.ai verified live with docs. Strengths: strong early adoption, genuinely differentiated multimodal plus knowledge-graph RAG. Docked for: (1) GitHub license detection inconclusive (NOASSERTION), relying on project statement; (2) young project, production hardening unproven. License: MIT (per project announcement).

★ 3,716 GitHub stars

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
7.0/10
Maintenance · recency, releases, issue load
8.0/10
Openness · license + docs
8.0/10
Manual review · sanity read + red-flag check
7.0/10
  • · 3,716 GitHub stars, pushed 2026-10-14, not archived (GitHub API, morphik-org/morphik-core)
  • · MIT license stated in Show HN launch (news.ycombinator.com/item?id=43763814)
  • · GitHub license detection returns NOASSERTION; relying on project statement
  • · morphik.ai verified live with docs; hosted free tier available

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.