Open-source AI social media manager
The AI social media manager that gets you seen everywhere.
One idea becomes a post for every network you connect, written native to each instead of copy-pasted across them, with the image or video generated on the way. It learns your niche before it writes, and measures what landed after. Run every brand you have, set how much the agent does alone, and drive the whole loop from your own AI client. Open source, so self-host it or let us run it.
One agent · every network
Building is cheap now. Distribution isn't.
AI turned a weekend into a working product. It also turned everyone else's weekend into one.
The bottleneck moved. There are more good things to find than there is attention to find them with, and the ones that get found are the ones that show up in public every week, on every network their audience reads.
Doing that by hand is a full-time job. Most tools take one slice of it: a scheduler wants copy you already wrote, a writer hands you text and stops. This runs the whole loop, for the SaaS you launched on Tuesday or for the account that is the thing you are building.
More on what that means in practice, and where the platform rules draw the line: automated social media accounts.
Three steps to a social presence that runs itself.
No blank box, no content calendar to fill. Connect your accounts, let the agent learn your niche, and open an Inbox that already has the week written.
Connect your networks
OAuth into the accounts you want to grow, one or all of them. Where a network has no OAuth, an app-specific password you can revoke on its own. Your real password never comes near us.
It researches and drafts
The agent studies your niche continuously: what's landing, who's rising, which angles are open. Then it drafts posts, threads, and replies in your voice, sourced and dated.
You decide how much rope
Everything lands in one Inbox: post, thread, reply, every network in the same queue. Approve it now, schedule it for a better hour, edit it, or bin it. Today nothing leaves your accounts without that tap, and as the dial opens, how far you turn it stays your call.
Your posts, written by AI agents.
Most social tools hand you a blank box and a calendar. This one hands you drafts. Hooks, threads, captions, and replies, each written for the network it's headed to and grounded in what the agent learned this week.
Hooks pulled from what's working
Every draft opens on an angle the agent found in your niche this week, not a template it memorised last year. It writes the first line first, then builds the post to earn it.
Written native to each network
One idea comes back as a thread on X, a caption on Instagram, a longer take on LinkedIn, a script for TikTok. Same message, written the way each network actually reads.
Your voice, not house style
Point it at the posts you're proud of and it picks up your rhythm, your vocabulary, the things you'd never say. Drafts come back sounding like you on a good day.
Replies that read the room
For every conversation worth joining, the agent reads the thread before it answers and drafts a reply with your actual take in it. Congratulations and fire emojis grow nothing.
A draft is an offer, never a takeover.
Every AI draft opens in its own editor next to whatever you were already typing, so the agent can never overwrite your words. Keep a line, rewrite the rest, or bin the whole thing. Each version it wrote stays in the log.
The agent that can't grade its own homework.
Every tool in this category tells you it grew your account. Ours logs each action in code the moment it happens, joins your live platform metrics to that log overnight, and builds its reports out of those numbers. Code keeps the record, so the model has nothing to shade.
It can't claim a win it didn't get
Code logs every publish, reply, and approval the moment it happens, with who did it and from where. The model cannot write to that record, so any number it quotes you came off your own account.
It says so when it doesn't know yet
Under 20 measured posts it reports your baselines and stops. You get your typical engagement and your active hours, which is a description of where you are rather than advice. Any cut with fewer than 3 posts behind it comes back as a small sample.
One change, then keep it or bin it
To test an idea it edits exactly one line of how it writes, then drafts half your normal posts that way. Beat the ones written the old way and the change sticks. Fall short and it is thrown out and logged, so the agent stops proposing it.
Your account is the dataset
Live platform metrics come back every night and join the log. The agent reads what happened on your account rather than a general theory about what performs on social media.
It proposes pillars, audience, and people to watch. You accept each one.
Your typical engagement, your active hours, your approval rate.
Findings against your own baseline, and experiments the coordinator scores.
The loop is adapted from Andrej Karpathy's autoresearch (2026): change one thing, run it on a fixed budget, keep it if the number improved and revert if it did not. The gates above come from his Recipe for Training Neural Networks (2019): get a dumb baseline before you draw conclusions. He wrote both for training models. They hold up on a social account.
Everything the agent does, in one place.
Not a stack of tools. One agent that composes, schedules, researches, and watches your niche. These are the product's own screens, running live on this page with sample data, not screenshots of them.
Every major network. And we tell you what's shipping.
One agent, 49 networks on the roadmap, and 31 of them working today. Live means you can connect it, draft to it, and publish. The other 18 are listed anyway, at the status they have actually reached.
Need one that's still planned? It's open source, so build the adapter yourself, or open an issue(soon) and we'll prioritize it.
Not a scheduler. An agent that grows your presence.
A scheduler waits for you to write. This one researches your niche, writes the drafts, and has the week queued before you open it. You approve; it does the work.
Autonomy is a dial you hold
The agent runs the whole loop on its own cadence: it researches, drafts, adapts each draft per network, and queues the week. Today the dial stops one notch short of the send, and that notch is yours. Where it's going is your choice too, per network and per content type: keep tapping approve, or hand it the send once it has earned it. Autopilot will be a switch you turn on, never something you wake up to.
Auto-research
Switch it on and pick how often, from hourly to let-the-agent-decide. From then it studies your niche on that beat: what's landing, who's rising, which conversations are open. It folds all of it back into every draft, sourced and dated, so your feed reflects this week rather than the day you set it up.
Drive it from any AI client.
GrowSocialelion ships as a Model Context Protocol server. Point Claude, your own agent, or any MCP client at it and work your presence in plain language: propose posts, refresh your voice profile, scan for replies to join. Every call goes through the same approval gate and the same audit log as the dashboard, so nothing arrives by a side door. Parity with the full UI is still being built, tool by tool.
API-backed where possible. Manual where required.
The Chrome companion is on a manual-only policy hold for X. It keeps brand voice settings nearby, but it does not read, inject, prefill, click, or submit on x.com.
A clear handoff, not hidden browser automation.
Use official APIs for supported X actions. When an action is unavailable there, review and copy the draft in GrowSocialelion, then perform every input and click on X yourself.
- No X page access, content script, injected controls, or composer prefill
- Official APIs handle supported actions under the scopes you grant
- Unsupported website actions stay manual: copy the reviewed text and act on the platform yourself
- Open source, so the permission boundary is auditable
Yours to run. Nothing hidden.
The agent that speaks for your brand shouldn't be a black box. GrowSocialelion is open source end to end under the AGPL. Inspect the prompts, own the data, self-host it, or let us carry the ops. Same code either way. The one condition: run a modified copy as a service for other people and they get your source too.
- AGPL-3.0: read every line, fork it, run your fork
- Self-host on your own infra, or let us run it managed
- No lock-in: your voice, your data, your accounts stay yours
- MCP server included, so you can drive it from any AI client
Free to self-host. Pay only for hosted.
The open-source agent is free to run forever. The managed tiers are the hosted service. Open them up at launch; you only move up when your usage does.
Run the full agent on your own infrastructure.
- AGPL-3.0: fork it, change it, run your fork
- No caps: the self-hosted plan is every limit set to null
- Your voice, data, and accounts stay yours
- MCP server included
- Every network the agent supports
Everything to start growing one presence.
- 2 connected networks on 1 brand
- 10 drafts, 5 publishes, 20 replies a day
- A daily scan of your niche
- One Inbox, one tap to publish
- Hosted by us, nothing to run
The whole loop, on your real network mix.
- 3 brands, unlimited networks on each
- 30 drafts, 15 publishes, 60 replies a day
- Hourly research and signal scans
- Frontier drafting model, 3 MCP keys
- 90 days of analytics and audit log
Every network, run like a team.
- 10 brands, unlimited networks, 5 seats
- 100 drafts, 50 publishes, 200 replies a day
- A voice per brand, per network, per format
- A year of analytics, 10 MCP keys
- Priority support
Self-hosting is free forever. Hosted plans have no card and no checkout here. You upgrade inside the app, only when you hit a limit.
Questions, answered.
What is an AI social media manager?
An AI social media manager is software that plans, writes, schedules and measures your social posts for you. GrowSocialelion is an open-source one: it learns your niche, writes each post native to the network it is going to, generates the image or video to go with it, and reports what actually landed.
What is GrowSocialelion?
GrowSocialelion is an open-source AI social media manager. It researches your niche, drafts posts and replies in your voice, generates the image or video to go with them, and schedules across every brand you run. You set how much it does alone, and publishing waits for your approval. Self-host it, or let us run it.
Does it write the posts for me?
Yes. AI agents do the copywriting: hooks, posts, threads, captions and replies, each written for the network it's going to and grounded in fresh research on your niche. They learn your voice from posts you've already published, and every draft opens in its own editor so it never overwrites what you wrote.
Which social networks does it support?
X, Instagram, Facebook, Threads, TikTok, YouTube, Mastodon, Pixelfed, Bluesky, Discord, Slack, Telegram, Nostr, Dribbble, Lemmy, Tumblr, Misskey, Twitch, Kick, LINE, Wykop, Plurk, Dailymotion, PeerTube, Dev.to, Qiita, TabNews, Hatena Blog, Hashnode, Ghost and WordPress are live today: connect, draft, schedule, publish. Another 18 are on the roadmap, including Instagram, LinkedIn, Facebook, Threads, TikTok, YouTube and Reddit. The Platforms table on this site lists every one of the 49 with its real status, and no network is marked live before its adapter ships.
How do I know its reports are honest?
Because the agent doesn't write the record it reports on. Code logs every action at the moment it happens, live platform metrics join that log nightly, and the agent can only interpret those numbers. It also refuses to draw conclusions early: under 20 measured posts it reports baselines only, and any cut with fewer than 3 posts is called a small sample instead of a finding.
Is GrowSocialelion really free?
Yes. The full agent is open source under the GNU AGPL-3.0 and free to self-host forever, with no feature held back for the paid tiers. The AGPL's one condition is that if you run a modified copy as a service for other people, they get your source too. The managed service adds a free tier for one network, with paid Pro ($19/mo) and Scale ($49/mo) tiers for more networks and a faster research cadence.
Do I have to approve every post?
Today, yes, every post, and that is deliberate. Publishing is gated in the runtime rather than by instructions to the model, so the agent cannot post or reply without your tap even if it decides it should. Approving is one tap, you can clear a week in a sitting, and scheduling means approved once and sent later. Autopilot is where this is going: the agent proposes, and you choose per network and per content type whether to keep tapping or hand it the send. It will be a setting you switch on, never the default.
Can I control it from my own AI agent?
Yes. GrowSocialelion ships as a Model Context Protocol (MCP) server, so any MCP client (Claude or your own agent) can propose posts, refresh your voice profile, and scan for conversations to join. Those calls hit the same approval gate and the same audit log as the dashboard. Full parity with the UI is still being built.
Let your agent grow your presence, starting tonight.
Connect one network and let it research, draft, and fill the week. You read what it wrote and decide what goes out. Free to start, open source to keep.
Not open for signups yet. The source and the hosted service both land at launch; self-hosting is free forever.