AI that writes in your voice, on your key.
The model is one you choose and pay for.
Bloom does not bundle an AI model and bill you for it by the seat. It calls a key you set on the workspace — Groq, OpenAI, Anthropic Claude, or a local Ollama — which Bloom verifies before saving and refuses to pretend around: drafting returns a 409 if no key is configured. The practical effect is that your content and your choice of model stay yours, rather than passing through a vendor black box you cannot see into or move off.
| Provider | Where it runs | What it costs you |
|---|---|---|
| Groq | Hosted API, your account | Often cents per batch on a fast model. |
| OpenAI | Hosted API, your account | Billed at your own OpenAI usage rate. |
| Anthropic Claude | Hosted API, your account | Billed at your own Anthropic rate. |
| Ollama | Local, on your machine | Free — nothing leaves your machine. |
Verified before it is trusted
A key is checked against its provider when you save it, so a typo is caught at setup rather than at the moment you needed a draft. If no key is set, the drafter does not silently stub a result — it returns a 409 and tells you to add one, which is the honest failure a social tool should have.
Switchable, not locked in
Because the model is yours, you can start on a fast, cheap hosted model and move to a local Ollama later — or the reverse — without exporting anything or leaving Bloom. The drafting behaviour is the same; only the key behind it changes.
Nothing to train on you
Bloom has no model, so there is nothing of Bloom’s to train on your posts. Your text goes only to the provider you chose, for the one draft you asked for. On a local Ollama it never leaves your machine at all.
Ideas from what is trending, not from a black box.
A blank compose box is the hardest part of running a social account, so Bloom suggests what to post about from what is actually trending across nine free, public sources — the same open signals you could read yourself, gathered into one place instead of nine browser tabs. The point is that every idea traces back to a source you can open, which is the difference between a suggestion you can sanity-check and one you simply have to accept.
This is deliberately not a recommendation engine you cannot inspect. Because the sources are public and free, an idea is auditable: you can see where it came from and decide whether it fits the brand before a single word is drafted. Ideas feed straight into the drafter, and from there into the same weekly queue the publishing page describes — so a topic can go from trending signal to a scheduled, approved post without leaving Bloom. When a post goes out and the conversation comes back, it lands in the unified inbox for the seven networks that return messages.
One idea, refitted to every network.
A draft is only useful if it sounds like you and fits where it is going. Bloom reads a brand kit — the name, tone, colours and fonts for each brand — and drafts to that voice rather than to a generic house style. Then it does the tedious part: it refits a single idea to each network, trimming to the character limit and matching the format, so one thought becomes a correct post everywhere instead of the same text pasted five times.
It writes to the brand kit
The brand kit is the voice input: name, tone, colours and fonts per brand. A drafter that writes to it produces copy a client recognises as theirs, which is what makes AI output usable rather than something an editor has to rewrite from scratch every time.
It respects each network’s limit
280 on X, 300 on Bluesky, 500 on Mastodon, thousands on Telegram — the full table is on the networks page. The drafter trims or expands one idea to fit each, and the queue blocks a shape a network cannot take before it is ever scheduled.
Autopilot fills the plan. You still approve it.
Autopilot can do the whole loop — source an idea, draft it, and schedule it on a cadence — so the calendar keeps itself full. What it cannot do is skip you. Every draft it produces enters the same four-state approval chain as a post a person wrote, and nothing reaches a network until it carries the approved state. Automation fills the plan inside your rules; it never becomes a way to publish something nobody read.
Approval is a state, not a convention
A post cannot be queued until it is approved, so “nothing ships unreviewed” is a property the product enforces rather than a norm the team hopes to remember. Autopilot’s drafts start at the beginning of that chain, exactly like a human draft. The full chain and roles live on the social-teams page.
It works inside the queue’s guard rails
An autopilot draft is still bound by the per-account rate governor, the catch-up cap and the duplicate guard. It does not invent new posting times or route around the limits that keep an automated queue from getting an account suspended.
You can always say no
Because every automated draft is reviewable, you can edit it, hold it or reject it like any other post. AI that cannot be reviewed is just a faster way to publish something you did not choose — Bloom keeps the human at the gate.
What Bloom’s AI deliberately is not.
The fastest way to lose a customer in month two is to let them assume a capability in month one. These are the four things Bloom’s AI is not, stated before you sign up rather than discovered after.
Not a house model
Bloom provides no AI model of its own. It calls the key you configure, which is why the quality and the cost are set by your provider, not by us. If you want a different model, you change the key — there is no Bloom model to be stuck with.
Not “free unlimited AI”
The drafting cost is real and it is yours, paid to your provider. On a fast hosted model it is often cents; on a local Ollama it is free because it runs on your hardware. Either way we name the cost instead of hiding it behind a plan tier and calling it unlimited.
Not trained on your data
Bloom does not train any model on your posts — it has none to train. Your text is sent only to the provider whose key you set, for the one draft you requested, and on a local Ollama it does not leave your machine at all.
Not a replacement for your judgement
Ideas and drafts are proposals. They trace to public sources, they fit each network, and they still pass through your approvals. The AI removes the blank page and the retyping; it does not remove the person who decides what the brand says. Questions? Reach us at contact.
AI questions, answered plainly.
Whose AI model does Bloom use to draft posts?
Where do Bloom's content ideas come from?
Does Bloom write in my brand voice?
Can the AI run the whole calendar on its own?
What does the AI drafting actually cost?
Does Bloom train on the posts I write?
Draft on your own key. Approve what ships.
Bloom runs on its own application and its own sign-in — an Autocloz account is not required. Add a Groq, OpenAI, Claude or Ollama key, set your brand voice, and let the drafter turn one idea into a correct post on every network you connect.