EditMind
title: edit-mind type: tool created: 2026-06-22 last_updated: 2026-06-22 radar_quadrant: Tools radar_ring: Assess radar_position: outer sources: ["https://github.com/IliasHad/edit-mind"]
edit-mind is a local-first video knowledge base that indexes video libraries using AI — object detection (YOLO), face recognition (DeepFace), emotion analysis, and speech transcription (Whisper) — and enables semantic natural language search across indexed footage. It also generates rough cut suggestions from indexed segments. All processing runs locally on the user's machine. Built with TypeScript; web UI served at localhost:3745. Approximately 1.7k stars as of June 2026. Released under a custom license (post-21 February 2026; prior versions MIT).
The custom Edit Mind License permits use by individuals, non-profits, and for-profit organisations with five or fewer employees without payment. Organisations with six or more employees may evaluate for 30 days, then must purchase a commercial license. Modifying the code for internal use is permitted; building a competing or resold product on top of edit-mind is explicitly prohibited. Pre-February 2026 releases remain under MIT.
edit-mind stores indexed metadata in PostgreSQL (via Prisma) and vector embeddings in ChromaDB. The ML service generates timestamps internally during transcription and object detection, but no documented export endpoint, CLI flag, or structured output format (JSON, CSV, segment list) exists. The backend exposes an undocumented BullMQ + Express.js HTTP API at localhost:3745 — technically queryable but unsupported and subject to breaking changes.
Workarounds for pipeline integration:
- Direct PostgreSQL query via the Prisma schema to extract segment timestamps after a search result.
- File an issue or PR for a
/exportendpoint; the project welcomes contributors and the data is already structured internally. - Manual handoff: copy timestamps from the UI into a downstream editing tool.
Placed in Tools / Assess / outer.
Large personal or professional video archives are effectively unsearchable without manual tagging. edit-mind addresses this by running AI analysis over each video at index time, building a structured metadata layer that natural language queries can target. The local execution model means no footage leaves the machine.
The outer position reflects three constraints confirmed from the live source: the project is explicitly not production-ready (v0.22.0, pre-v1.0 as of May 2026), the custom license introduces commercial use ambiguity for larger teams, and the absence of a documented export API limits composability with downstream editing tools such as radar/tools/FireRedOpenStoryline. The use case is real; the integration friction is the current barrier to moving inward.
