MomentSearch
- 1Runs onAPI, Linux, Mac, self-hosted, Web, Windows
- 2Visual searchYes
- 3Audio searchYes
- 4Multimodal searchYes
- 5API accessYes
Overview
MomentSearch is open-source software for visual video search and retrieval-augmented answers tied to specific video moments. Users can upload videos or submit YouTube URLs for indexing. Each search checks both visual frames and transcript text, then combines the results. Answers can point to clickable moments with keyframes, timestamps, synchronized transcript text, labels showing whether a match was seen or said, and match scores. A confidence gate can return no generated answer without making an LLM call when neither search branch meets its threshold. Without an LLM key, the system still returns ranked clickable moments, but does not write prose answers. Answer, embedding, reranking, and speech-to-text models can be selected by provider, with hosted and local options documented. Optional speaker recognition uses Gemini video understanding and requires a Gemini API key. The project includes Docker entrypoints and deployment guidance for Fly, AWS, and Google Cloud. Its Apache 2.0 license permits open-source use. The API has no authentication, and the application is single-user by default.
Who it is for
MomentSearch suits developers building or self-hosting video search systems who need both visual and transcript retrieval. It also fits users who want timestamped, clickable results rather than prose answers alone.
What is good
- Search combines visual frames with transcript text
- Citations include keyframes, timestamps, and match scores
- Models can use hosted or local providers
- Apache 2.0 licensed
What to know first
- API has no authentication
- Application is single-user by default
- Speaker recognition requires a Gemini API key
- SocialKit YouTube ingestion becomes paid after 20 free credits
Verdict
MomentSearch offers multimodal retrieval with clickable, timestamped results and configurable models. Its unauthenticated API and single-user default require careful deployment choices.
Compared on AI video search tools
- Visual search
- Yesgithub.com
- Audio search
- Yesgithub.com
- Multimodal search
- Yesgithub.com
- API access
- Yesgithub.com
Facts
- Purpose
- MomentSearch is visual video search and retrieval-augmented generation software that returns exact video moments matching what is seen and said.github.com · 30 Sept 2026
- Input
- Users can upload videos or paste YouTube URLs for indexing.github.com · 30 Sept 2026
- Retrieval
- The system always searches both a visual frame branch and a transcript branch, then fuses their results.github.com · 30 Sept 2026
- Citations
- Answers cite clickable moments with matched keyframes, timestamps, transcript synchronization, seen or said badges, and match scores.github.com · 30 Sept 2026
- Uploads
- Presigned uploads send gigabytes directly from the browser to object storage instead of through the API.github.com · 30 Sept 2026
- Speaker recognition
- Optional per-video speaker recognition uses Gemini video understanding and requires a Gemini API key.github.com · 30 Sept 2026
- Abstention
- A confidence gate can abstain without making an LLM call when neither retrieval branch clears its threshold.github.com · 30 Sept 2026
- Deployment
- The project ships as a neutral Docker image with api, worker, clip, and one-shot seed entrypoints, and documents Fly, AWS, and Google Cloud deployment.github.com · 30 Sept 2026
- Storage integrations
- Object storage supports local storage, AWS S3, GCP storage, GCP native storage, and Fly Tigris, while the architecture uses Postgres, Qdrant, and Prefect Cloud.github.com · 30 Sept 2026
- Security limit
- The API is unauthenticated; anyone who can reach its port can mint upload URLs, ingest, ask, and delete, so the README recommends localhost binding or an authenticating proxy.github.com · 30 Sept 2026
- User model
- The application is single-user by default and ignores X-User-Id and Authorization headers, routing every request through SINGLE_USER_ID.github.com · 30 Sept 2026
- YouTube ingestion limit
- The documented SocialKit YouTube ingestion option provides 20 free credits and then becomes paid, while cookie authentication is free but cookies expire approximately every two to three weeks.github.com · 30 Sept 2026
- License
- MomentSearch is released under the Apache 2.0 license.github.com · 30 Sept 2026
- Maker privacy
- Traversaal.ai says it uses strictly necessary authentication and usage cookies and does not use cookies for advertising, analytics, or tracking.traversaal.ai · 30 Sept 2026
- Model flexibility
- Answer, embedding, reranking, and speech-to-text models are configurable by provider name, including hosted and local options.github.com · 1 Oct 2026
- Integrations
- Documented infrastructure integrations include Postgres, Qdrant, Prefect, and object storage through Tigris, Amazon S3, or Google Cloud Storage.github.com · 1 Oct 2026
- Security limitation
- The API has no authentication, and anyone who can reach the port can upload, ingest, ask, and delete data.github.com · 1 Oct 2026
- Storage protection
- The project recommends keeping buckets private and using presigned URLs for uploads and playback.github.com · 1 Oct 2026
- Retrieval-only mode
- Without an LLM key, the application still returns ranked clickable moments but does not generate prose answers.github.com · 1 Oct 2026
Best MomentSearch alternatives
See all 20
Where it ranks on Specifiction
Is MomentSearch yours?
Claim it for free: prove the domain, then correct facts, plans and screenshots. An editor reviews every change.
Sources
- github.com/traversaal-ai/momentsearch· checked 30 Sept 2026
- github.com/traversaal-ai/momentsearch/blob/main/MO· checked 30 Sept 2026
- traversaal.ai/privacy· checked 30 Sept 2026
- github.com/traversaal-ai/momentsearch/blob/main/DE· checked 1 Oct 2026