What's New in Subscribr

Subscribr's AI integration can now manage your Projects board and research with Intel directly from your AI agent. Ask Claude or ChatGPT to save an idea you just brainstormed, open your interactive Projects board, move work through your pipeline, search channels and videos, inspect results, or save research — without copying everything between tabs.

The expanded integration can also generate ideas, turn an idea into a Script project, manage Tasks and notifications, create custom templates, and safely create or update a channel Voice Profile after validating its exact format and asking for your approval. It's available on all paid plans. Go to AI Integrations in Subscribr to copy the MCP server URL and follow the setup guide. If you connected Subscribr before this update, refresh its tool list or reconnect it, then start a new conversation.

We're also preparing Subscribr for submission to the ChatGPT app directory and Claude Connectors Directory, which will make installation even easier.

GPT upgraded from 5.4 to 5.6 Terra for Canvas script drafts, offering stronger reasoning at the same price. If you already had GPT selected in your channel or script settings, you're automatically using the latest version. GPT 5.2 has also been retired — any scripts still set to it now use 5.6 Terra automatically.

MiniMax upgraded from M2.7 to M3 for Canvas script drafts, offering stronger reasoning and better consistency with story requirements. If you already had MiniMax selected in your channel or script settings, you're automatically using the latest version.

Swipe Mode is a new full-screen way to flip through your channel's video ideas and decide what's worth making, one idea at a time. Keep the ones you love, skip the rest, with a swipe, a click, or your arrow keys. You can flip through dozens of ideas in a couple of minutes instead of scrolling a long list.

It gets sharper every swipe. The ideas you Keep and the ones you Skip quietly teach Subscribr your taste, so the next batch of generated ideas leans toward what you actually want to make. Outlier-backed ideas show the proven video that inspired them, with its view count and outlier score, so you can decide with real context. Changed your mind? Undo brings the last card right back. And the deck refills itself in the background, so you rarely run out.

It's live now. Open any channel's Ideas page and tap Swipe Mode to try it.

You can now rename private chat threads from the sidebar or directly from the conversation title. Give ongoing work a name that makes sense to you, then find it again without guessing which conversation is which.

Team owners can now remove saved Anthropic and OpenAI API keys directly from Team Settings. That makes it easy to rotate a key or switch how Agent Mode connects whenever you need to.

Canvas now moves more smoothly from planning into writing: your latest outline stays in sync, the editor is ready when the script view opens, and explicit script-length requests carry through. Section research also finishes cleanly when there is nothing useful to add, so you can keep writing instead of waiting on a section that appears stuck.

When you regenerate a completed Agent Mode script, you can now choose a different target length from the same menu as your model and template. Keep the draft direction you like while making the next version shorter or more detailed, without starting over.

Copying a full script now gives you clean, ready-to-paste output with the title included once. It is a small detail that makes handoffs to Google Docs, Notion, teleprompters, and production tools feel much tidier.

GLM upgraded from 5.1 to 5.2, offering stronger long-context performance for Canvas script drafts and Agent Mode runs. If you already had GLM selected in your channel settings or Agent Mode, you're automatically using the latest version.

Claude Sonnet upgraded to version 5, offering improved performance for script drafts and Agent Mode runs. If you have Claude Sonnet selected in your channel settings or Agent Mode, you're automatically using the latest version.

Agent Mode now follows the instructions in your chat messages, even when they differ from how your selected template usually structures a script. If you ask it in chat to open with a particular story or skip the intro, it does that — the template still shapes the writing style, but it no longer overrides what you typed.

When you're creating a script and choosing a script template, you can now see what kind of opening it writes, what it's best for, and when to avoid it — so you can pick the right one for the video you have in mind instead of guessing from the name alone.

If a script generated in Agent Mode isn't structured the way you want, you can now pick a different script template right from the regenerate controls and run it again — no need to start a new script from scratch.

Creating a custom template from YouTube videos is more reliable on longer videos. Templates stay focused and save successfully instead of failing or producing overly long output.

Channel voice generation no longer fails when your channel doesn't have unique terms to list in the glossary.

Thumbnail prompt generation for video ideas now works correctly for channels with saved thumbnail style analysis.

Agent Mode scripts now get one more check before they finish. Subscribr reviews the draft against your channel voice after the cleanup passes, so scripts are less likely to drift into a generic AI style before they land in your Script tab.

Chat and Canvas are now faster and steadier. Responses are less likely to stall, repeated replies are gone, and the research, videos, and channels you add reliably stay in context for the whole conversation. Your chat threads also always get a title now, and script generation in Canvas runs more smoothly start to finish.

The Projects board now uses clearer final-stage columns: Ready to Upload for finished videos and Published for upload and published work. You can also archive or restore cards from the board and card details.

The YouTube description and video breakdown tools are more dependable now. If a video cannot be analyzed, Subscribr will show a clearer message instead of getting stuck or showing a crash page.

Agent Mode credit pricing now matches how you pay for scripts elsewhere in Subscribr:

• Open-source models (Kimi, MiniMax, GLM) use the same length-based credits as Canvas script generation.

• BYOK Claude and GPT runs stay at 2 credits per run, flat.

• Deep Research is still +3 credits when a new Perplexity pass runs.

Failed or cancelled runs are still refunded automatically.

Agent Mode is more reliable and easier to understand when a run finishes. We've improved hook generation so the agent evaluates multiple opening options before writing, added clearer run details with the summary, quality report, and research sources in one place, and made the app better at explaining what happened when a run has issues.

We also tightened up failure handling for connected AI models and Deep Research. If a connected AI account is missing credits or a research provider is temporarily unavailable, Subscribr now gives a clearer message instead of leaving you guessing. Longer Agent Mode runs also have more time to complete, which helps complex scripts finish instead of timing out right near the end.

Subscribr started as a place to write scripts. But a script is only part of making a video. You still have ideas to sort through, thumbnails to design, descriptions to write, and the whole thing to get ready for upload. That work used to live in spreadsheets, Trello, ClickUp, and a half-dozen other tabs.

Projects pulls all of it into one board, so you can run your whole channel from one place. Each video is a card you drag across columns: Idea, Script, Thumbnail, Video Gen, Upload.

The best part: drag an idea card onto the Script column and Subscribr offers to write the whole script for you with Agent Mode. Click any card to open it and manage everything from there: generate a thumbnail, write your description, and work through your publish checklist.

Working with a team? Post comments, assign cards to teammates, set due dates, and follow the activity feed. There's a built-in notification system for everything happening on your board, with optional Slack notifications too. (Flying solo? It all still works. You're just a team of one.)

What's next: we're connecting Projects to Subscribr Video, so you'll take a video all the way to a finished, edited cut from the same board. Generate ideas, write scripts, design thumbnails, and automatically edit the video into an upload-ready package.

It's live for everyone today. Open any channel and click Projects in the channel nav.

Agent Mode automates your whole scriptwriting process. Instead of working through the Canvas step by step, you hand off your idea and an agent does the work: it researches your topic, builds an outline, writes and scores 5-7 hook options to find the strongest, drafts the script in your channel's voice, then runs humanizing, retention, and repetition passes before a final quality check.

It all runs in the background. Pick your topic, hit Generate with Agent, and close the tab if you want. Most runs take 15 to 30 minutes, and we'll email you the moment your script is ready. The finished draft lands right in your Script tab to edit and ship.

A few things worth knowing:

• 2 credits per successful run. A flat rate no matter how long your script is. If a run doesn't finish, you're not charged.

• More model choice. Three open-source models are included free for a limited time to celebrate Agent Mode's launch: Kimi K2.6, MiniMax M2.7, and GLM 5.2. Want Claude or GPT? Connect your own API key (BYOK) for more options, including Claude Haiku and Opus.

• Optional Deep Research. For tougher topics, add Perplexity Deep Research for +3 credits.

• It's in the API, too. Trigger Agent Mode from your own tools and automations. (Requires the Automation or Scale plan.)

Prefer to drive each step yourself? The standard Canvas workflow (questions, outline, script) isn't going anywhere. Agent Mode is just the hands-off option when you want a finished draft without babysitting it.

Your AI agent can now use Subscribr.

The new Subscribr CLI installs a skill file into Claude Code, Codex, Cursor, OpenClaw, and any other Agent Skills–compatible agent. Once it's there, you can ask your agent — in plain English — to research a niche, generate video ideas, write a full script, or design a thumbnail. The agent knows which Subscribr endpoint to call, how to authenticate, and how to poll for async results.

Install with one command from your project:

npx @giltotherescue/subscribr-cli

The skill teaches your agent every endpoint in the Subscribr REST API and how to chain them together — for example, researching a niche, outlining a script, and exporting it to markdown without you writing a single line of glue code.

Optional Python CLI: Add --with-cli to also drop in subscribr.py, a zero-dependency CLI that wraps all 45 API endpoints. Useful for cron jobs, n8n workflows, GitHub Actions, or anywhere you want to call Subscribr without an agent in the loop.

REST API access requires the Automation or Scale plan. Grab a token at subscribr.ai/developer. The repo is open source at github.com/giltotherescue/subscribr-cli.

Subscribr now uses the latest versions of DeepSeek, Kimi, and MiniMax for script drafts: DeepSeek V4 Pro, Kimi K2.6, and MiniMax M2.7. We've also improved OpenRouter Nitro routing for faster throughput, so drafts generate through high-speed providers while keeping model costs tracked more accurately behind the scenes. If you already had one of the older versions selected, your scripts will automatically use the newer version.

Your script is ready. Now your thumbnail is too.

Thumbnail Studio brings AI-powered thumbnail creation directly into your Subscribr workflow — no Canva, no Photoshop, no context-switching.

Brainstorm mode: Describe your video idea and get 4 concept sketches across different proven formats. Pick the one that fits, and Subscribr generates a publish-ready thumbnail. 157 formats trained on 25,000+ high-performing YouTube videos — not random templates.

Clone mode: Upload any competitor's thumbnail or paste a YouTube URL. The AI reverse-engineers the visual style — colors, typography, composition, lighting — and generates thumbnails in that style for your video.

AI Feedback & Auto-Improve: Every generated thumbnail is scored against CTR principles: contrast, text readability, emotional impact, mobile visibility.

One click runs an improvement pass that fixes the weak spots automatically.

YouTube Preview: See your thumbnail in context — placed next to real videos on a simulated YouTube page — before you publish.

Style memory: Set your visual preferences once — colors, composition style, reference photos of your face, logo, products — and every generation reflects your brand.

Import styles from any YouTube channel to match a look you like.

Every plan includes free monthly thumbnail generations — Creator gets 5/month, Automation gets 15/month, and Scale plans get 30–60/month.

Need more? Add-on subscriptions and one-time packs are available from the Billing page. Packs never expire while your account is active. Learn more about Thumbnail Studio.

Click the mic. Start talking. Subscribr writes it down. There's now a microphone button in both the chat bar and the planning questions textarea — tap it, speak your thoughts, and text appears in real time as you talk. The button turns red while recording; click it again to stop. Text is transcribed as spoken — no punctuation added. The AI will understand it. Works in Chrome, Safari, and Edge; Firefox isn't supported yet.

GPT-5.4, OpenAI's newest model, is now available as a draft writing model in Subscribr. You can select it from your channel settings under Draft Writing Model. Available on all plans.

Canvas script regeneration now includes 3 free full regenerations per script (shared across Regenerate Script and Humanize). After free regenerations are used, you can unlock additional regeneration batches with credits directly in-app.

Creating a script from an idea no longer charges credits up front. Credits are now charged at the correct time: when you actually generate script content on the canvas.

The Plan (outline) step now uses a cleaner top action bar aligned with Script mode, including compact stats and actions for clearer flow between outlining and scripting.

Resolved intermittent issues where Plan/Script tabs could get stuck and where toolbar controls (like version arrows and export/share) sometimes disappeared after first generation. These controls now reliably stay in sync without requiring a full page refresh.