Agent Mode is Subscribr's fully automated script generation pipeline. Instead of working through the Canvas step by step, you hand your idea to an AI agent and it researches, outlines, writes, and quality-checks your script for you.
Best For
Agent Mode works well when you have a clear topic and want a complete first draft quickly. If you prefer hands-on control at each stage, the standard Canvas workflow gives you more creative input along the way.
How to start Agent Mode
You can launch Agent Mode from three places:
- Projects board — Drag an idea card into the Script column (or another production column). A confirmation modal appears with model, research, and credit details — click Start writing to create the project and start the agent immediately. See Projects & Production Board.
- Canvas — Open an existing script in the Canvas and use the Agent Mode controls in the chat panel to trigger a run on that script.
- REST API —
POST /v1/scripts/{id}/agent/generate(Automation plan and above; see below).
How It Works
When a run starts, the agent runs through a series of steps automatically:
- Research -- The agent gathers relevant information on your topic using web search and any context you provided.
- Outline -- It builds a structured outline based on the research, your channel voice profile, and your audience.
- Script writing -- The agent drafts several hook options, then expands the outline into a full script, applying your voice and style preferences.
- Quality checks -- The agent runs retention optimization, humanization, and repetition cleanup passes.
- Voiceover preparation -- The script is adapted for spoken delivery.
The entire process runs in the background. You can close the tab and come back later -- Subscribr notifies you both in-app and by email when the script is ready.
How Long Does It Take?
A typical Agent Mode run takes 15 to 30 minutes depending on the length and complexity of your topic, and occasionally longer for very involved topics or when video production is enabled. You can watch the progress bar in real time, or leave and return once you receive the completion email -- the run continues even if you close the tab.
Credits
Agent Mode pricing depends on which writing model you choose. Standard Canvas script generation is unchanged — this section applies only to Agent Mode runs.
Included open-source models (Kimi, MiniMax, GLM): charged the same way as generating a script in Canvas — based on your script's target length. A typical 10-minute script (~1,600 words) is 5 credits per run; longer targets cost more, shorter targets cost less.
BYOK Claude or GPT models: 2 credits per successful run, flat, regardless of length.
You're charged when a run starts. Running the agent again on the same script costs again. If a run fails or you cancel it, credits are refunded automatically (see Canceling a Run).
Perplexity Deep Research adds 3 credits when a new deep-research pass runs. If this script already has a reusable deep-research report from a prior run, that surcharge is not charged again.
Choosing a Model
Agent Mode includes three open-source models with no API key: Kimi K2.6, MiniMax M2.7, and GLM 5.1. They use the length-based credits described above.
Prefer a model from Anthropic or OpenAI? Connect your own API key (BYOK) in Team Settings — Claude models (Haiku, Sonnet, and Opus) with an Anthropic key, and GPT models with an OpenAI key. BYOK runs use the flat 2-credit access fee; provider token usage is billed on your key. If the model you pick requires a key you haven't connected yet, Subscribr will prompt you to add it before the run starts.
BYOK needs a true API key from the provider's developer console — a Claude Code, ChatGPT, or Codex subscription will not work, since those plans don't include API access.
Deep Research
For topics that need deeper sourcing, turn on Perplexity Deep Research in the generation options. The agent runs a dedicated deep-research pass before writing, pulling richer source material into the script. This adds 3 credits to the run.
Monitoring and Reviewing
You can watch the run progress from two places:
- Projects board card — Click the card that triggered the run and open the Script tab in the detail panel. It shows a live status badge ("Writing…") and a Watch progress button that takes you to the Canvas.
- Canvas — The progress panel updates in real time. The run continues even if you close the tab.
When the agent finishes, your script appears in the Canvas. From there you can:
- Read through the full script and make direct edits.
- Use highlight-and-improve to refine specific sections.
- Run a Humanize pass for additional tone polish.
- Continue chatting in the left panel to request targeted changes.
The agent also generates an outline you can review under the Plan tab.
Video Production Pipeline
After the script is complete, Subscribr can continue with automated video production:
- Voiceover -- An AI voiceover is generated from the finished script using your selected voice.
- Avatar video -- A presenter avatar is created to deliver the voiceover on camera.
- Shot list and rendering -- The system plans visual scenes and renders a complete video.
These steps are optional and happen automatically when enabled for your channel. You can track their progress in the same progress panel that shows the script generation steps.
Canceling a Run
If you need to stop an Agent Mode run, open the Canvas and click the Cancel button in the progress panel. Canceling is immediate, and your credits are refunded automatically — you only pay for runs that finish successfully.
Tips for Better Results
Set up your voice profile first. The agent uses your channel voice profile and audience settings to shape the script. Channels with detailed profiles produce noticeably better output.
Add research context. Before launching Agent Mode, attach relevant YouTube videos, articles, or documents using the Add Context feature. The agent uses this material during the research phase.
Review and refine. Agent Mode produces a strong first draft, but the best scripts come from editing the output with your own expertise. Use the Canvas editing tools to add personal stories, adjust pacing, and tighten the language.
Using Agent Mode via the API
Agent Mode is also available through the Subscribr REST API, so you can trigger it from your own tools, scripts, and AI agents. Send a request to POST /v1/scripts/{id}/agent/generate to queue a run, then poll the returned status URL until the script is ready. API runs use the same credit rules as the Canvas (length-based for included models, 2 credits for BYOK Claude/GPT). REST API access requires the Automation plan or above -- grab a token at subscribr.ai/developer.
Next Steps
Want more control over each step? Learn about the standard Canvas workflow. For tips on polishing any script after generation, see Writing and Editing tools.