Deepgrip From $89/mo · Free plan 9.3 Visit QMedia
github.com
What it is
QMedia is an open-source multimedia AI search engine for content creators. It searches and extracts information from text, images, and short videos, combining those formats in a multimodal retrieval and question-answering system. Its web service displays content and sources as cards with image, text, and short-video breakdowns. The project can run locally for offline search and question answering over private data. The web service, RAG search and Q&A service, and image, text, and video model service can be deployed separately. The standalone model service offers an API for image and text encoding, video transcription, and image OCR. Documented model options include CLIP, BGE, Qanything OCR, LLaVA, and Faster Whisper, with local Ollama model switching. Google content search is configurable but disabled by default, as is video service use in the RAG server configuration. The RAG service depends on a separate mm_server service. QMedia is self-hosted, available through its web service, and licensed under MIT.
Spec sheet
- What it does
- QMedia is an open-source multimedia AI search engine for creators that extracts and queries text, images and short videos.Read 15 Oct 2026 · github.com
- Product
- QMedia is an open-source multimedia AI content search engine aimed at content creators.Read 8 Oct 2026 · github.com
- Deployment
- Its web service, RAG search and Q&A service, and image/text/video model service can be deployed separately.Read 8 Oct 2026 · github.com
- API
- The standalone multimodal model service can provide image and text encoding, video transcription, and image OCR through an API.Read 8 Oct 2026 · github.com
- License
- The repository is released under the MIT License.Read 15 Oct 2026 · github.com
- API use case
- The model server can operate independently for image and text encoding, video transcription and OCR through an API.Read 15 Oct 2026 · github.com
- Changelog
- The changelog identifies version 0.1.0 and lists multimodal model deployment, API calls, Ollama, image, video and embedding models.Read 15 Oct 2026 · github.com
- Content cards
- The interface presents image, text and video material as content cards containing sources and extracted breakdowns.Read 15 Oct 2026 · github.com
- Content types
- It supports searching and extracting information from text, images, and short videos.Read 8 Oct 2026 · github.com
- Custom results
- Search results can be tailored to the user's interests and needs.Read 15 Oct 2026 · github.com
- Dependencies
- The RAG service depends on the separate mm_server service to run.Read 8 Oct 2026 · github.com
- Documentation
- Documentation consists of a top-level README plus separate installation and configuration guides for each service.Read 15 Oct 2026 · github.com
- Frontend requirements
- The web frontend requires Node.js, pnpm, dependency installation and an environment file before running pnpm dev.Read 15 Oct 2026 · github.com
- Google integration
- Google content search is supported and can be enabled through the mmrag_server configuration.Read 15 Oct 2026 · github.com
- Image embeddings
- Images are embedded with a CLIP encoder.Read 15 Oct 2026 · github.com
- Independent deployment
- The three services can be deployed separately according to available resources and embedded into other systems.Read 15 Oct 2026 · github.com
- Installation requirements
- The model and retrieval services require separate Python 3.11 Conda environments and pip dependency installation.Read 15 Oct 2026 · github.com
- Integration
- The RAG service supports Google content search, which the installation guide says is configurable and disabled by default.Read 8 Oct 2026 · github.com
- Language models
- Local Ollama switching is supported, with llama3 8B and 70B models documented.Read 15 Oct 2026 · github.com
- Limit: deployment complexity
- DEV.co says each service needs its own environment and that production use requires substantial infrastructure and dependency coordination.Read 15 Oct 2026 · dev.co
- Limit: hardware guidance
- DEV.co and SourcePulse note that hardware requirements and performance baselines for local multimodal models are not documented.Read 15 Oct 2026 · dev.co
- Limit: ingestion workflow
- DEV.co says initial ingestion relies on manually prepared asset files and database cleanup, with no bulk-import tooling shown.Read 15 Oct 2026 · dev.co
- Limit: operational readiness
- DEV.co reports that authentication, encryption, health checks, retry logic and vulnerability-disclosure procedures are not documented.Read 15 Oct 2026 · dev.co
- Limit: release packaging
- GitHub shows that QMedia has no published releases to pin or download.Read 15 Oct 2026 · github.com
- Limit: unfinished analysis
- Hysen Labs reports that visual understanding and several short-video analysis entries are marked unfinished in the repository.Read 15 Oct 2026 · hysenlabs.com
- Local use
- The project says it can be deployed locally for offline search and Q&A over private data.Read 8 Oct 2026 · github.com
- Model technology
- The backend uses Python and LlamaIndex applications for retrieval, question answering and multimodal processing.Read 15 Oct 2026 · github.com
- Models
- The project documents local Ollama model switching and lists CLIP, BGE, Qanything OCR, LLaVA, and Faster Whisper among its model options.Read 8 Oct 2026 · github.com
- Multimodal Q&A
- It combines text, image, and short-video information in a multimodal RAG system for content search and question answering.Read 8 Oct 2026 · github.com
- Multimodal RAG
- Its retrieval-augmented system answers questions using extracted attributes such as image style, text layout, video transcription and summaries.Read 15 Oct 2026 · github.com
- Notable news
- Hysen Labs reports that the repository's most recent push was dated April 9, 2026.Read 15 Oct 2026 · hysenlabs.com
- OCR
- Image text recognition uses a QAnything-based OCR service.Read 15 Oct 2026 · github.com
- Offline operation
- The project supports local deployment for offline search and question answering over private data.Read 15 Oct 2026 · github.com
- Praise: creator use case
- SourcePulse describes QMedia as tailored to creators and highlights its content-card presentation for extracted information.Read 15 Oct 2026 · sourcepulse.org
- Praise: local privacy
- DEV.co highlights private, offline media search and question answering without uploading creator assets to third-party services.Read 15 Oct 2026 · dev.co
- Praise: modularity
- Hysen Labs praises the separable services because model and retrieval components can be deployed on different hardware and models can be swapped.Read 15 Oct 2026 · hysenlabs.com
- Privacy deployment
- The README describes local deployment for offline search and Q&A on private data; it does not state a security certification or compliance status.Read 8 Oct 2026 · github.com
- Public API
- The model and retrieval services expose interactive API documentation through local /docs endpoints.Read 15 Oct 2026 · github.com
- Support and community
- The README links to issue reporting and requests, and invites users to join its Discord and WeChat communities.Read 8 Oct 2026 · github.com
- Support channels
- The project README invites users to join Discord and WeChat communities.Read 15 Oct 2026 · github.com
- Text embeddings
- Text embeddings use the multilingual BGE encoder, with BAAI/bge-small-en-v1.5 shown in configuration.Read 15 Oct 2026 · github.com
- Three-service architecture
- QMedia separates a multimodal model server, a retrieval and question-answering server, and a web frontend.Read 15 Oct 2026 · github.com
- Video setting
- Video service use is configurable and is off by default in the RAG server configuration.Read 8 Oct 2026 · github.com
- Video transcription
- Faster Whisper is used for video transcription and is described as runnable on a local CPU.Read 15 Oct 2026 · github.com
- Web technology
- The web service uses TypeScript, Next.js, Tailwind CSS and shadcn/ui.Read 15 Oct 2026 · github.com
Plans and prices
QMedia Free Open-source software · MIT license Read 8 Oct 2026 · github.com
Strong points and limits
Strong points
- Searches text, images, and short videos
- Supports multimodal question answering
- Offers local offline deployment
- Provides image and video model APIs
- Uses an MIT open-source license
Limits
- RAG service depends on mm_server
- Google search is disabled by default
- Video service is disabled by default
- Separate services require deployment planning
Who it suits
QMedia suits content creators who need multimodal search across text, images, and short videos. It also fits users seeking local, offline question answering over private data or an API for media processing.
Verdict
QMedia brings multimodal retrieval, cards, local deployment, and media-processing APIs into one open-source project. Its separate services and default-disabled integrations require configuration before use.
Where Specifiction places it
- AI Video Search Tools#7 of 248.4
