Getting Vidsy ready

github

Jan Bogaerts • Peerlist

MD² is an AI orchestration tool for developers that organizes AI-assisted software development around features instead of conversations. Use MD² to work faster with multiple agents, reduce repeated prompting and token usage, and keep better control over how features move through development<ul><li>Feature cards keep track of what every AI agent is working on. Cards can optionally be linked to Git worktrees, updated automatically by agents, and managed from a single dashboard.</li><li>Feature-centric history. A feature usually spans multiple prompts, conversations, commits, and coding sessions. MD² connects all of them back to the feature instead of leaving you with one long list of disconnected chat histories.</li><li>Local Markdown knowledge base. Cards, chatlogs, design notes, and documentation live inside your repository as Markdown files. Agents can read and update them directly, giving sessions access to the entire development history.</li><li>Reusable actions. Instead of cluttering context windows with large sets of unused skill references, MD² uses agent-actions which automatically inject the right project context through placeholders like {{card-file}}.</li><li>Automate the boring parts. Run commands when cards change state, update diagrams, commit changes, execute scripts, or trigger custom workflows. For recurring tasks, let an AI generate the script once, then reuse it without spending tokens every time.</li><li>Built-in token cost tracking. Automatically see AI cost per action, per feature, and across the entire project so you always know where time and money are being spent.</li></ul>

<a target="_blank" rel="noopener noreferrer nofollow" class="text-green-500 rich-link break-all link-break" href="http://Vidsy.ai">Vidsy.ai</a> is a tool for indexing, searching and analyzing visual data. It uses an LLM to convert text searches into structured queries that are run against the results of various CNNs.Key features:<ul><li>real-time camera analysis: multiple cameras can be analyzed in sync</li><li>record video on multiple tracks in sync</li><li>re-analyze recordings</li><li>search through local images and videos</li><li>share your findings with friends and colleagues</li></ul>As this release is still a beta, there are some limitations:<ul><li>only supports a local file system based db, for smaller datasets</li><li>the number of included models currently used for analysis is limited and is only related to faces.</li><li>LLM usage is currently limited to 10 queries a day (a previously retrieved query can be re-used unlimited)</li><li>max nr of bytes shared is limited at 20GB a month</li><li>probably still a few bugs</li></ul>What's coming next:<ul><li>more models for object search and such<ul><li>stage 1 will be the more conventional model types (most likely Resnet based)</li><li>in a second stage I'm planning to work out another idea round CNNs in the hopes to make it easier to teach the system how to recognize things.</li></ul></li><li>support for external db systems (mongodb), which will also allow multiuser use-cases (share indexes)</li><li>flesh out some more features that make things more convenient.</li><li>cloud support for data processing so that things can scale with large camera pools.</li></ul>