Give your agent its own footage locker.
An agent that generates or edits video needs where to put it — not your camera roll. Agent workspaces are isolated WeaverClip vaults with their own API key, transcript hours, and storage. They start free and scale through the exact same plans humans use, so one billing model covers both.
Free never bills. Uploads pause at 10 GB; files stay on the agent host until deleted or upgraded. Same 30-day retention as human Free, 180 days on paid.
agent-workspace
Free scalable trialGenerate → Upload → Transcribe → Clip → Render — all isolated
- 1 Agent renders draft MP4 with Remotion
- 2 Uploads to its vault — verified byte-count + MD5
- 3 MCP transcribes, ranks 20 moments/hour
- 4 Creates top clips → helper renders → signed URL
Four loops that are useful on day one
Autonomous daily recap
Nightly agent watches your 2-hour stream, transcribes 40 minutes worth saving, ranks 20 moments, renders the top 3 vertical clips — you wake to drafts, not a timeline.
Generated video QA
Remotion/FFmpeg pipeline renders a draft, uploads to its own vault, agent checks transcript alignment, re-generates if captions are sparse, then marks ready.
Podcast → Shorts while you sleep
Agent ingests the week’s raw, finds complete thoughts that survive without context, and queues 10 captioned exports per hour up to 200 per recording.
Layout testing at scale
Agent burns 30/60 layout gens finding the best 9:16 plate pair for each scene — first regeneration free, then packs if it wants more.
One key, own vault, same pricing
No separate agent SKU. The billing-catalog.json that powers human pricing is the source of truth for agents too — so we can’t drift.
Security that survives unattended runs
The agent never receives raw video in chat. Tool responses contain session IDs, timecodes, transcripts, and short-lived signed URLs (minutes). Storage keys are namespaced by workspace; R2 never makes anything public.
Revoke the agent key and the helper’s Keychain session stays valid for you. Storage remains until its 30/180-day date unless you delete it.
Cost that an agent can reason about
Every plan states: storage in GB, transcript in hours, renders in count, retention in days, overage in $/GB and $/render. The agent can read PLAN_LIST and decide “this 90-minute build needs Pro 8h — Creator 4h won’t fit.”
Free never auto-bills. Paid overage only above the vault; transcript hours simply pause. No seat fees.
What this unlocks as agents get better
Today an agent can store a few tests free and graduate to $15 when it ships weekly. Tomorrow, the same vault can hold an agent’s entire history — searchable transcripts, signed playback links, and durable renders — without a new system. The “autonomous video” startup you described just needs a footage locker that charges like storage, not like seats.
Byte count + MD5 before local delete, same as human pipeline. An unattended upload that fails simply stays and retries.
Renders run on the helper, not on WeaverClip GPUs, so agent renders don’t price you out.
Free 30 days, paid 180 days, shown on each recording, enforced daily. Perfect for an agent that creates 100 tests and keeps 10.
Every Free run can leave a trail — if it actually stored bytes
Agents call weaverclip_submit_review after each use — rating 1–5, task, model, success. One per session, max 5/day, 500 chars. Only the workspace that stored the session can review it, so Free counts and spam doesn’t.
Public aggregate: avg rating + last notes. Private rows stay per-workspace. Powers the “Agents rated WeaverClip 4.8/5 over 147 runs” badge on vs OpusClip once you have volume.
How agents keep your disk clean — the vault math they use
Every agent workspace reads billing-catalog.json via lib/product/calc.ts — gbPerHour, planFit, overageCost, clipYield — so a future $15→$16 change is one JSON edit, not 25 find-replaces. An agent that does 3 podcasts a week at 6 Mbps (~4 GB each, 1.5h) uses 12 GB and 4.5h of transcript — Free 10 GB/2h holds one, Creator 75 GB/4h holds the week. The agent checks GET /api/usage before transcribing, sees 1.2h left, picks 1h not 2h, and never asks you. That's why the vault math asterisk is on every page: *Hours at 6 Mbps ≈2.7 GB/h, actual varies with bitrate.
See the 3 moments we'd pull before you sign up
MCP weaverclip_search_transcript — free taste.
What does your Claude clipper cost per month?
*Hours are at 6 Mbps (1080p30 webinar, slides, sermons) ≈2.7 GB/hour. Actual hours vary with bitrate — 5 Mbps ~2.25 GB/h, 12 Mbps ~5.4 GB/h, 25 Mbps ~11.25 GB/h. Vault is always bytes; hour label is a guide. Hard limit = 2× vault (e.g., Creator 75→150 GB), overage $0.04/GB-month. Transcript never bills overage.
The asterisk on every vault table is why biggest streamers trust the page: we state the vault in GB (the truth) and the hour guide at 6 Mbps (the intuition), plus the four-row table so you translate before you record. Hard limit 2× vault pauses uploads, files stay on Mac. Nothing is deleted to make room.
What agents actually save you — not just time, but disk and dollars
An agent that does 3 podcasts a week at 6 Mbps (~4 GB each, 1.5h) would, with local ffmpeg, pile 12 GB in ./outputs and 4.5h of transcript in ./transcripts, then ask you to clean it. With Weaver, it writes to /tmp, uploads to its isolated vault (scoped API key + R2 prefix), helper verifies byte count + MD5, deletes /tmp, and leaves your Mac at 14% disk. The vault holds 12 GB, transcript 4.5h — Free 10 GB/2h holds one, Creator 75 GB/4h holds the week. The agent checks GET /api/usage before transcribing, sees 1.2h left, picks 1h not 2h, and never asks you. Over a month, that's 48 GB and 18h of transcript — Pro 150 GB/8h covers it with 102 GB headroom, Studio covers the quarter. The 25 tools the next coder builds will all read lib/product/calc.ts so a future $15→$16 change is one JSON edit.
The asterisk on every vault table — and why we put it there
Every vault table on this site shows storage in GB (the truth) and hours at 6 Mbps (the intuition) with the asterisk: *Hours at 6 Mbps ≈2.7 GB/h. At 12 Mbps ~5.4 GB/h, at 25 Mbps ~11.25 GB/h. Vault is always bytes; hour label is a guide. Hard limit 2× vault (Creator 75→150 GB) pauses uploads, files stay on Mac. This single sentence is why a 2-hour Valorant VOD at 12 Mbps doesn't surprise a Free user who bought “~3 hours” and got 10.8 GB. We state both, plus the four-row table, so you translate before you record. That is the sharp nuance biggest streamers notice: we tell you where the line is, before you cross it.
Give your agent somewhere to put the next video.
Create a free workspace now — 10 GB and 2 hours to prove the autonomous loop. When it needs more, Creator $15, Pro $29, Studio $119 are waiting with the same vault, not a new product.
Agents use the same billing policy and unit economics as humans. No hidden seat fees.
The asterisk that keeps us honest — and why biggest streamers notice
Every vault table on this site shows storage in GB (the truth) and hours at 6 Mbps (the intuition). The asterisk is not a footnote to hide behind — it's the promise that we state the vault in bytes and the hour guide at a single, documented bitrate, plus the table to translate your actual OBS settings in seconds. A 2-hour 1080p60 Valorant VOD at 12 Mbps is 10.8 GB, not 5.4 GB. The “Free 10 GB” you bought holds one of those, not two. The same vault at 6 Mbps sermons holds two. If we only said “~3 hours” you'd buy the wrong plan and blame us at 2am when OBS says “upload paused.” So we show both: the vault in GB, the hour guide at 6 Mbps, and the four-row table (5 Mbps ~2.25 GB/h, 6 Mbps ~2.7 GB/h, 12 Mbps ~5.4 GB/h, 25 Mbps ~11.25 GB/h) so you can translate before you record. Hard limit is 2× vault — Creator 75→150 GB, Pro 150→300 GB, Studio 1500→3000 GB — uploads pause there, files stay on your Mac and retry with backoff. Nothing is deleted to make room. Overage is linear $0.04/GB-month, transcript never bills overage. That's the nuance that makes a 24h subathon at 12 Mbps (~129 GB) fit Studio 1.5 TB with headroom, while a 2h sermon at 6 Mbps fits Free 10 GB with 1.9 GB left. We reuse the same lib/product/calc.ts (gbPerHour, planFit, overageCost, clipYield) on every page, so a future $15→$16 change is one JSON edit, not 25 find-replaces. That's top 0.005% system thinking, not just pretty pages.