Writing
How Runmeg learns your voice
Short answer: Runmeg reads your site and the public LinkedIn, X, and Instagram posts you connect. It stores that in memory you can edit or delete. Create and ReelGen use that memory. You approve a plan before a video is made, and you publish the file yourself. Runmeg does not autopost, reply, like, follow, or DM, and it does not invent a customer quote or a metric you did not supply.
The voice is a profile you can open. It is not a hidden prompt, and it is not permission to publish.
What it reads
The site is the offer. The public posts are the habits: sentence length, words you repeat, words you never use, whether you write as I or as we. Onboarding asks you to connect the accounts you want included. A private message, a draft you never published, and a customer's closed ticket are not part of this. If it is not public, it is not the voice.
You should read memory after it is built, the way you would read a new hire's notes. Delete a sentence that overstates the product. Add a ban. If you skip this, the error becomes the house style, because the next video will imitate it. Memory is editable on purpose. A brand you cannot see is a guess.
- The pages on your site that state the offer and the limit.
- Public posts on the networks you choose to connect.
- Your later edits: bans, deletions, and posts you mark as the ones to imitate.
- Not your inbox, not private customer data, not a number you have not published.
What memory is used for
Memory is the brief. Create uses it to write a plan for one video on a clip you upload. Nothing renders until you approve that plan. ReelGen uses it to caption five moments from one upload. None of that is publishing. Publishing is you, after you have watched the file. If a plan drops a limit you wrote into memory, put the limit back or do not approve.
The file is your footage with a line on it. You download it and you decide. A render finishing is not an approval, and it is not a new shot. Create does not invent footage. ReelGen does not invent footage.
What you should correct
Correct claims, not vibes. 'Do not say we replace the CRM' teaches the next video. 'Make it punchier' teaches nothing, and the next line will get louder. When a post performs, add the fact to memory only if you would still sign it. A post that performed because it overclaimed is a bad example. It will be imitated.
Correct the pile, too. If a product video said I, memory mixed the founder with the company. Split them. The next plan will not guess the split from one scolding in a chat if the stored fact is still mixed.
A useful edit to the profile Delete: "Runmeg replaces your marketing stack." Add to never-say: replace, unlock, game-changer. Add as a limit: we do not post, email, or DM for you. Keep: the sentence on the homepage about the video using the brand. Do this before the next Create plan. Do not try to out-prompt a stored sentence.
What it never does
It never publishes. Approving a plan is not a scheduler that fires. It never replies, likes, follows, or sends a DM. It never invents a customer, a logo, a shot, or a metric to make a video more impressive. If the proof is not in memory or in the footage, the line should not contain it. Treat a missing proof as a reason to cut the sentence, not as a gap the model should fill. A fact you would not say on a call does not belong in the next caption.
No autopost
No engagement bots
No invented proof
No hidden voice
Where this is the wrong product
If you want a tool that posts on a schedule, Runmeg will lose that comparison on purpose. If you want an avatar, HeyGen is the closer aisle. If you want to edit the recording yourself, Descript is the closer aisle. Runmeg turns a clip you already have into a marketing video, using memory you can edit. It is not trying to be the whole stack.