What's changed.
Every shipped agent, fix and improvement — newest first. The team works in the open; this is the log of it.
- Fixed
Publishing that doesn’t flinch
A reliability pass on the last mile. If a platform hiccups, your post no longer goes down with it.
- Fixed caption drift on 120 fps footage exported to vertical formats
- Failed platform uploads now retry three times before asking you anything
- Scheduled posts survive a re-render — the schedule follows the new cut
- Export queue position is now visible while you wait
- New
MCP server beta — agents can hire EditorBuddy
Point Claude (or any MCP client) at mcp.editorbuddy.com and it can plan, edit and publish through the same orchestrator the web app uses.
- One-line config: add the server URL, get the full toolset
- Same job graph, same memory, same cost caps as the web app
- SSE progress events so the calling agent can narrate the run
- New
Memory across projects
The memory agent now carries what it learns from one project into the next: caption style, pacing, banned words, the way you like your hooks.
- A visible memory panel — see exactly what EditorBuddy knows about your style
- Edit or delete any remembered preference; nothing is locked in
- Your 10th video comes out more ‘you’ than your 1st
- New
Magic Clips with virality sub-scores
Every clip the shorts agent cuts now ships with four scores — hook, flow, engagement, trend — so you know which one to post first, and why.
- Cross-modal moment detection: visuals and audio dynamics, not just the transcript
- Clips ranked, not just listed — the best opener floats to the top
- One-tap reframe to 9:16, 1:1 and 16:9 per clip
- Improved
Transcription moved to Groq Whisper
The transcription agent now runs on Groq-hosted Whisper. A 40-minute podcast transcribes in under a minute, with word-level timestamps intact.
- Roughly 14× faster than the previous pipeline
- Word-level timing feeds straight into captions and silence removal
- Speaker labels are steadier across long recordings
- New
Captions in 48 languages, timed to the word
The subtitle agent grew up. Word-level timing in every language it speaks, plus a style system that follows your brand instead of fighting it.
- 48 caption languages at launch, all with word-level timing
- Brand styles: pick fonts, colors and emphasis once — every video after that matches
- Profanity and banned-word handling you control per project
- New
All 12 feature agents on the floor
The full crew shipped: cleanup, clips, translation, dubbing, editing, b-roll, audio, publishing and friends — every one watchable while it works.
- The agent activity feed: a live log of who is doing what to your video
- Jobs run as a graph — independent agents work in parallel
- Per-job cost caps so a long video never becomes a surprise bill
- New
The private alpha: one prompt, four agents
The first version of the idea. You typed a sentence; an orchestrator planned the job and three agents executed it. Rough, slow, and it already felt different.
- Orchestrator plans a dependency graph from a plain-English prompt
- Transcription, cleanup and captions as the founding crew
- First 50 creators invited