Mosaic: Agentic Video Editing via Node-Based Canvas

TL;DR

TL;DR: Mosaic is a YC W25 startup that lets you build multimodal AI agents on a node-based canvas to edit video using natural language prompts — from auto-clipping raw footage to generating AI avatars, voiceovers, and B-roll.

Source and Accuracy Notes

This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: mosaic.so
• Source repository: No public GitHub repository found (private/internal codebase)
• License: Proprietary (no open-source license available)
• HN launch thread: news.ycombinator.com/item?id=45980760
• Docs: docs.mosaic.so
• Source last checked: 2026-06-26

What Is Mosaic?

Mosaic is a YC W25 startup founded by ex-Tesla engineers. Their core product is a node-based canvas where you build multimodal AI agents that edit video via natural language. The founding story: recording hours of Cybertruck footage, getting frustrated scrubbing through raw clips in DaVinci Resolve, and building a prototype that could “count Cybertrucks in a video with a single chat prompt.”

The canvas exposes a library of AI tiles you wire together:

• AI Avatar — generate reusable AI presenters from source media
• AI Voiceover — narrate with AI-generated voice
• AI B-Roll — auto-generate or source relevant B-roll matched to spoken content
• AI Music — generate custom music matching video mood and pacing
• Captions / Cinematic Captions — stylized subtitle overlays
• Clips — auto-extract short standalone clips for social platforms
• Rough Cut — auto-create a rough cut by removing filler, silence, dead air
• Silence Removal — strip awkward pauses automatically
• Color Correction — AI color grading
• Reframe — auto-convert 9:16 or 1:1 while keeping subject in frame
• Video Generation — generate video from prompts using Seedance 2
• Motion Graphics, Intro, Outro, Watermark — branding tiles

Setup Workflow

Step 1: Sign Up

Visit app.mosaic.so and create a free account. The free tier includes a set number of credits for trying the canvas.

Step 2: Connect Your Footage

Upload raw video via the Asset Manager. Supported inputs: video files, audio files, images. The upload flow uses signed URLs — see docs.mosaic.so/api/asset-management/upload-flow for the REST upload sequence.

Step 3: Build an Agent on the Canvas

Drag an Input tile onto the canvas and link your footage. Then add AI tiles (Avatar, Voiceover, Captions, etc.) and wire them in a chain. Each tile is a node; the graph defines the editing pipeline.

Step 4: Write a Natural Language Prompt

In the chat panel, describe what you want. Examples from the docs:

• “Find all clips where a car drives by and add a cinematic intro”
• “Add an AI narrator using a British accent, then generate captions”
• “Create a 60-second montage from this raw footage with AI music”

The agent parses the prompt, invokes the relevant nodes, and produces an output video.

Step 5: Export or Publish Directly

Use the Destination tile to export locally or publish directly to YouTube, TikTok, Instagram, or LinkedIn. See docs.mosaic.so/tiles/destination.

Deeper Analysis

The Node-Based Canvas Architecture

Unlike traditional video editors where you manually apply effects, Mosaic treats each AI capability as a node in a directed graph. The canvas UI is at app.mosaic.so — it visualizes the agent’s execution flow, showing which tiles ran, their outputs, and any errors.

The graph is persisted as an agent template. You can duplicate, share, and version-control agents via the API at docs.mosaic.so/api/agents.

API Access

Mosaic exposes a REST API with the following key endpoints documented in the llms.txt index:

• POST /api/agents — create an agent template
• GET /api/agents — list all agents for your organization
• POST /api/agent-runs — execute an agent on a video
• GET /api/agent-runs/{id} — poll run status and outputs
• POST /api/agent-runs/{id}/cancel — cancel an in-progress run
• POST /api/agent-runs/{id}/resume — resume a failed or cancelled run
• GET /api/credits — check credit balance and auto top-up settings
• POST /api/webhooks/events — subscribe to agent run lifecycle events

Credit usage is tracked per tile and per date via GET /api/credits/usage.

Pricing

Mosaic operates on a credit-based system. From the docs, plans are available at docs.mosaic.so/api/plan/get-plan and docs.mosaic.so/api/plan/get-plan-list. The canvas UI shows a credit meter in the bottom-left corner. Auto top-up can be configured via POST /api/credits/settings. Enterprise plans for teams are documented at docs.mosaic.so/more/enterprise.

Models

Mosaic uses multiple AI models across its tiles, billed by generation seconds rather than per-call. The full model list is at docs.mosaic.so/models.

Practical Evaluation Checklist

• [ ] Node-based canvas renders correctly with tile library visible
• [ ] AI Avatar tile generates a presenter from uploaded media
• [ ] AI Voiceover synthesizes speech from a text script
• [ ] Rough Cut tile removes silence and filler from raw footage
• [ ] Clips tile extracts short social clips from a long video
• [ ] Destination tile exports to at least one social platform
• [ ] API key authentication works with the REST endpoints
• [ ] Credit meter reflects actual usage after a run
• [ ] Webhook notifications fire on run completion

Security Notes

• Asset uploads use signed URLs with short-lived expiry — raw API keys are not exposed in client-side upload flows
• API keys are scoped to your organization; key management is via the dashboard at app.mosaic.so
• Social account connections (YouTube, TikTok, etc.) use OAuth — Mosaic stores refresh tokens but these can be revoked individually via DELETE /api/social/connections/{id}
• Enterprise tier adds SSO and audit logging per docs.mosaic.so/more/enterprise

FAQ

Q: Does Mosaic have a public GitHub repo? A: No. Mosaic is a closed-source commercial product. No public source repository exists as of 2026-06-26.

Q: Can I use Mosaic programmatically without the UI? A: Yes. The REST API at docs.mosaic.so/api/introduction covers agent creation, execution, cancellation, and resume. Webhooks provide async notifications.

Q: What video formats does Mosaic accept? A: Standard formats accepted: MP4, MOV, AVI, MKV for video; MP3, WAV, OGG for audio; PNG, JPG, WEBP for images. Upload is handled via the signed URL flow documented at docs.mosaic.so/api/asset-management/upload-flow.

Q: How is Mosaic different from CapCut or Adobe Premiere Pro with AI plugins? A: Mosaic’s agent paradigm means you describe the outcome in natural language and the AI decides which operations to apply. Traditional editors expose individual tools you operate manually. The node graph is also externally accessible via API, enabling programmatic video pipelines.

Q: Is there a free trial? A: The free tier includes a set of starting credits. Plan details and top-up rates are available at docs.mosaic.so/api/plan/get-plan-list.

Conclusion

Mosaic is a compelling entry in the AI video editing space — the node-based agent canvas gives you both a visual workflow and a programmable API, making it useful for one-off creators and teams building automated video pipelines alike. The YC W25 backing and credit-based pricing model lower the barrier to experimentation. If you are tired of manually scrubbing footage and applying effects one-by-one, Mosaic’s agent-first approach is worth exploring.

Visit mosaic.so to get started.