COMMERCIAL ACCESS · LOCAL PRODUCTION

Turn one idea into a course video.

Lecturecast is an AgentMesh360 commercial course-video product. Your Agent verifies an active monthly pass and universal API key, the cloud Director turns structured choices into a signed production plan and narration script, and your machine produces two MP4s plus two matching covers. Original media is never uploaded; narration text goes to your selected TTS provider. (Bilibili and Xiaohongshu are the major Chinese video platforms — landscape and portrait respectively. The pipeline produces both formats whether or not you publish to them.)

Install Lecturecast → Get an AgentMesh360 pass
Lecturecast DirectorActive monthly pass · 10 credits / generation

Cloud direction, local production.

The Agent verifies the universal API key, active monthly pass, and shared credits, then completes the native choice cards and Creative Brief approval. Once 10 credits are confirmed, the cloud Director issues a signed ProductionManifest and a complete signed narration, both immutable to the host Agent. The complete signed narration is reviewed and approved section by section; TTS and rendering execute locally only after approval.

~ run-this-locally 
$ curl -fsSL https://raw.githubusercontent.com/jiyangnan/AgentMesh-Lecturecast/main/scripts/install.sh | bash
PS> irm https://raw.githubusercontent.com/jiyangnan/AgentMesh-Lecturecast/main/scripts/install.ps1 | iex
# After installation, start a NEW host-Agent task; the fresh Skill returns the exact command
$ lecturecast onboard --adapter codex --host-contract 1.0.0 --json # Codex example
# Only if requires_user_action asks for an API key: lecturecast auth login
# Rerun onboard after the fix, then execute only workflow.next_action
Why hand it off

AI can write a script. Production is the actual pain.

Getting an AI to write a script is easy. What's hard is everything after — voiceover, timeline alignment, rendering, subtitle burn-in, cover design, and producing landscape + portrait versions each time.

Lecturecast gives your AI agent an executable control flow: Director owns the signed creative plan; your machine owns TTS, timing, and rendering. Any access, asset, timing, or QA failure stops delivery instead of producing plausible text or an incomplete video.

Capabilities · cloud plan, local production

From one topic to two finished videos.

i.

Director-generated signed script

Direction and facts are confirmed in the Creative Brief. After 10 credits are approved, Director generates the immutable signed ProductionManifest and complete narration script; the host Agent cannot rewrite them.

ii.

Local voiceover and alignment

Only your approval of every signed narration section unlocks TTS. Measured audio drives scenes, subtitles, and composition duration; the build stops if any section or total leaves the 75–125% timing envelope, or final A/V coverage differs by more than one second.

iii.

Dual-platform rendering

One script, rendered locally with Remotion into both Bilibili landscape (16:9) and Xiaohongshu portrait (9:16), each with its own layout. No second pass of editing required.

iv.

Subtitles + covers

Local ffmpeg burns in Chinese subtitles and creates two matching PNG covers. Delivery waits for full decode, audio coverage, subtitle-glyph, and dimension QA; you make the final publish decision.

Real delivery · first-customer run

One topic. Two real videos.

Produced in a real first-customer run on 2026-07-21. The same signed script and measured timeline became two native layouts on the customer's machine—not one crop in two containers.

SourceFirst-customer run
Run date2026 · 07 · 21
ProductionLocal render
DeliveryQA verified
BILIBILI · 16:9 · 04:48 QA PASS

Bilibili landscape

1920 × 1080 · H.264 + AAC
8,653 FRAMES

XIAOHONGSHU · 9:16 · 04:48 QA PASS

Xiaohongshu portrait

1080 × 1920 · H.264 + AAC
8,653 FRAMES

08chapters
1,409script characters
91subtitle events
288.433sper final video
0missing glyphs
DELIVERY NOTE

The validation run preserved its full issue and repair record. These are the final files that passed audio-coverage, subtitle-glyph, dimension, and full-render QA.

Local three-piece stack

All on your own machine.

Native macOS and Windows are supported; Linux and WSL are not. There are no cloud media workers or remote renderers. The cloud handles commercial access, creative decisions, and a signed Manifest; narration text goes to the selected TTS service. install.sh / install.ps1 checks what the local machine is missing.

RENDER

Node + Remotion

Code-driven animated scenes that render both aspect ratios. Install Node 20+, then let npm install prepare the project.

VOICE

Python + edge-tts

Audio files are generated and stored locally. The client sends narration text to Microsoft Edge speech; with MiniMax BYOK, it sends the signed narration text to your own MiniMax account.

MUX

ffmpeg (with libass)

Burns Chinese subtitles and stitches the final video. Use ffmpeg-full on macOS or a native Windows ffmpeg build with libass; doctor verifies the actual capability.

How to use

Designed to be agent-friendly.

The installed Skill drives one control flow. Brief approval, 10-credit Manifest-generation approval, and complete signed-script approval are three independent human gates before local TTS and rendering.

01

Install the CLI

Use install.sh on macOS or install.ps1 in Windows PowerShell. Both install the commercial CLI, signature verification, and Agent Skill. Then start a new Agent task; its current Skill runs onboard with exact --adapter and --host-contract arguments.
See on GitHub →

02

Your Agent completes commercial onboarding

The new task attests the installer-owned current Skill, then checks the account, monthly pass, credits, and renderer. It executes only the CLI's returned next_action; missing or stale Skill state stops before project mutation or rendering.
See AGENTS.md →

03

Director plans; your machine ships

Following Director choices and Creative Brief approval, 10 credits are consumed for the signed ProductionManifest. The Agent presents the complete signed narration—still immutable—and waits again for approval; only then do local TTS, landscape and portrait renders, subtitles, and two covers proceed. A QA failure stops delivery.

FAQ

FAQ.

Does it cost anything? Do I need an account?
Lecturecast requires an active AgentMesh360 monthly pass and universal API key. It reuses shared credits and needs no separate LectureCast pass. Choices and Brief creation do not deduct credits; 10 credits are deducted only upon approval of ProductionManifest generation.
Does the video get uploaded to the cloud?
Original media is not uploaded. A bounded source summary, stable option IDs, the Brief, and client capabilities go to the cloud Director; narration text goes to the selected TTS provider. Original media, local paths, generated audio, subtitles, rendering, and outputs stay local.
What does the voiceover use? Can I get a richer voice?
Edge TTS sends narration text to Microsoft's Edge speech service. MiniMax BYOK sends the signed narration text to your own MiniMax account. LectureCast does not intentionally persist MINIMAX_API_KEY; use secure process-environment injection because typing a literal shell export may put it in shell history.
Is production fully automatic? What are the human gates?
Production is not unattended. Delivery requires three human approvals: the Creative Brief, a 10-credit Manifest generation, and section-by-section review of the complete signed narration. Once signed, the host Agent may not edit it; failure at any gate stops delivery before TTS or rendering.
What do I need to install?
Three local pieces: Node + Remotion, Python + edge-tts, and ffmpeg with libass. macOS install.sh and Windows install.ps1 both distinguish CLI installation from renderer readiness; doctor is the capability source of truth.
Which platforms are the outputs for?
Four standard files: a Bilibili landscape MP4 (1920×1080), a Xiaohongshu portrait MP4 (1080×1920), and two matching PNG covers.
Is the public client open source?
The public CLI, local execution tools, and website are Apache 2.0 on GitHub. The complete product has one commercial path: an active monthly pass, universal API key, shared credits, and the cloud Director are required. There is no account-free route.