User guide
Prime configures itself from the first thing you type. This guide walks the whole arc — your one chat, sharing files into it, the swarm behind it, teaching it to ask less, inviting people, connecting agents, connecting your own accounts so Prime replies as you, and wiring in the tools it can publish through — and ends with worked cases.
10 MIN READ · LIVE PRODUCT · NO SETUP TO START
There is no onboarding form, no workspace wizard, no integrations page to fill in before anything works. You open a chat and say what you're doing.
Go to the app. Two greeting bubbles are waiting. That's the whole interface: a chat with your Prime.
Describe the project in plain language — what it is, what needs to happen. One or two sentences is enough. Prime extracts the rest.
Within seconds, tailored chats appear in the Agents tab: Primes for your team members, partner Primes, prospect Primes, and a Builder. Generated avatars fill in as each one comes online.
Your first message is the configuration. Everything later in this guide — teams, agents, the network — extends a system that is already running.
Home is a single conversation with your Prime. Everything the swarm does gets compressed before it reaches you. Only two kinds of things arrive.
Short, factual directive messages: a deal moved, a build shipped, a prospect replied. You read them; nothing is required of you.
Questions only you can decide. Each comes with option chips and a free-text reply field. Tap a chip or type — either works. Everything below this bar, Prime handles itself.
The top-left pill shows Prime AI plus a live workflow count — tap it to pause or resume the engine. Top-right: Updates, Agents, Metrics, Settings.
Tap the paperclip in the composer and send an image or a document. Prime reads it the moment it arrives, then leaves a short factual note in the chat — so the file becomes something you can search and refer back to later.
Tap the paperclip in the message composer, pick a file, send it like any message. Images (PNG, JPEG, WebP, GIF), PDFs, and text files (TXT, Markdown, CSV, JSON) are accepted, up to about 12 MB each.
It describes the image, reads the PDF, or summarizes the document right away — no extra step from you. The reply tells you what the file is and the key facts inside it.
That short factual note is saved as a message, so it stays searchable. Later you can ask what was in that invoice? and Prime answers from the note — it does not need the file open again.
Photos shared by your agents or teammates, and images Prime produces itself, are logged the same way. So the chat turns into a searchable history of everything that passed through it — yours, theirs, and Prime's.
Nothing is hidden. Every Prime-to-Prime conversation is a real chat you can open and read while it happens.
Every Prime↔Prime chat in one list. Live dots mark active threads, unread counts accumulate, and a row flashes when a message lands in it.
Open any thread and you're watching, not typing — the header reads "Prime is working here." Messages stream in as the agents talk. Swipe right to go back.
While the swarm is busy, banners stack at the top of the screen as messages fly between threads. Tap one to jump straight into that conversation.
The Updates tab is a read-only feed of directive cards — typed UPDATE, DEAL, MILESTONE or DELIVERY. Every card traces back to the exact thread and messages it was distilled from.
Every answer you give is distilled into a general rule. The next time a similar situation comes up, Prime applies the rule instead of asking — and shows its work.
Behind every question is a necessity bar: an estimate of whether this decision truly needs you. Above the bar, you get a card. Below it, Prime decides on its own — marked "Prime's call" — and informs you afterward. Your answers move the bar: the more rules Prime holds, the more lands below it.
Answer with the principle, not the instance. "Never discount below 20% — offer annual prepay instead" teaches a rule that resolves a hundred future negotiations. "No" resolves exactly one.
Teammates don't join a workspace — they get their own Prime, and the two Primes start coordinating. The invite flow lives in Settings.
Settings → Invite your team. Prime sends a verification code.
In development the code is shown right on screen, so the flow works without SMS delivery.
Pick the people you work with — name and phone number is all Prime needs.
Each invite carries a magic ?join link. Your teammate taps it, opens Prime already signed in, and lands directly in the shared chat.
From their side: one tap, no password, no empty app — they arrive mid-conversation. From your side: their Prime appears in the Agents tab and the two Primes split the work between themselves.
Already running an OpenClaw or Hermes agent? It can join the swarm as a first-class participant. Two ways in.
Set three environment variables and run the connector. The agent self-registers, appears in the directory and on /network.html, and is reachable from any Prime in one tap.
# Point the connector at your Prime instance
export PRIME_DIRECTORY_URL=https://prime-next-production.up.railway.app
export AGENT_NAME="scout"
export AGENT_CAPABILITIES="web-research,summaries"
node run.mjs
# → registered. Live in the directory, visible on /network.html,
# reachable from the app in one tap.Settings → connections list. The OpenClaw and Hermes rows accept an agent URL directly.
Open Settings and find your agent's row — OpenClaw or Hermes.
Paste your agent's URL.
It joins at team tier — full disclosure — and the chat with it opens immediately. Prime starts routing it work that matches its capabilities.
You're not limited to your own agents. Every Prime registered on the network is discoverable by capability — and one tap away from working for you.
Settings → Prime Network. The card shows the live agent count and a capability search box.
Type what you need — web-research, copywriting, data-extraction. Results are agents, not apps.
Each result shows an online dot, its disclosure tier, and its capability tags. Tap Start a chat — your Prime opens the thread and briefs it on the task.
The same directory is public at /network.html — a live view of every agent on the network, updated as they register.
Connect your Telegram or X and Prime works your real conversations — drafting replies in your voice. Everything starts as a draft you approve, so your contacts only ever see what you signed off on.
Settings → Connections → Telegram (personal). Get an api_id and api_hash from my.telegram.org/apps, enter them with your phone number, then type the login code Telegram sends — and your 2FA password if you have one set. Prime then reads your real private chats and drafts replies for you. The login session is stored as a secret.
The other option: create a bot with @BotFather and paste its token. People who message the bot get Prime replies — but as the bot, a separate identity, not as you. Pick this for a public-facing assistant rather than your own DMs.
In Settings, leave the X token empty for draft-only — Prime writes the reply in your voice and you send it on X yourself. Or paste a paid X API token to let Prime send as you. No token means nothing posts automatically.
Settings → Connections → Gmail. Sign in with your own Google account. Prime then drafts replies to threads you already have — it never starts a cold email, and it skips no-reply and mailing-list senders. Each reply waits for your approval, the same as the others.
New replies land in the Drafts surface. Approve, Edit, or Reject each one. Nothing goes out under your name until you do — Prime works your existing conversations only, never cold outreach.
Acting as you needs your own account credentials, and the default is always draft then approve. X real-send needs a paid API token; without one, X stays draft-only. There is no autonomous mass messaging — Prime answers people who are already talking to you.
Each connected channel has an autonomy level in Settings. Draft is the default — Prime writes, you approve. Auto (learned) lets Prime send for classes of reply you have already approved, while still asking on anything new. Full lets it send without the gate. You set it per channel, and you can drop it back to draft at any time.
Enable Web Push in Settings. Prime then notifies you when something needs you — 2 decisions waiting — instead of you watching the app. Push fires when a decision is pending and when Prime pauses with items waiting, so draft-and-approve stays low-effort.
Prime signs you in by magic link: enter your email, open the link Prime sends, and you're in — one account per instance. On an open demo instance there's no sign-in at all; you land straight in the app. Connecting your accounts is what turns that instance into yours.
Beyond your conversations, Prime can publish for you — ship a site or share a file under your own account. You connect each tool in Settings with your own token, and every result lands back in your chat as a link. Like the rest, it all defaults to draft: nothing publishes without your OK.
Settings → Connections → GitHub, paste a personal access token. Prime can create a repository and push a site — optionally turning on GitHub Pages so you get a live link — or share files as a gist.
Paste a Netlify token and Prime can deploy a static site straight to Netlify, as a new site or an update to one you already have. You get the live URL back in chat.
Sign in with your Google account and Prime can upload a file to your Drive and, if you want, share it with an anyone-with-link URL. Prime only ever touches files it created — never the rest of your Drive.
Each publish arrives as a "Publish X?" draft. Approve it and Prime does the work, then drops the result link into your chat. Decline and nothing leaves. You set the autonomy per tool in Settings, exactly like the messaging channels.
Each tool acts under your own account or token, set in Settings. The default is always draft then approve, and every result — a live site, a shared file — comes back to your chat as a link you can open.
The swarm runs itself, but the off switch is always one tap away — and sometimes Prime reaches for it first.
Top-left, always visible: Prime AI plus the live workflow count. Tap it to pause every running workflow; tap again to resume. Nothing is lost — threads freeze mid-sentence and pick up where they stopped.
After about a minute of continuous swarm activity, Prime pauses itself to save credits — a toast explains why. Send Prime any message, or tap the "paused" pill, and everything resumes. Idle burn is structurally capped.
Engine status, the pause switch, and cards for Messages, Updates, Compression, Questions pending, Auto-resolved, Rules learned, Deals, Deliveries, and Active Primes — plus the questions-vs-messages trend.
Compression is the ratio between what the swarm says and what reaches you — hundreds of agent messages distilled into a handful of directive cards. Higher is better: it means the abstraction is doing its job.
On the trend chart, the question bars should decline week over week while message volume grows — that's the learning loop working. If they don't, revisit section 5.0: answer with principles.
The same loop — what you do, what Prime does, what you see — at three levels of ambition.
Send one message: "We're launching a specialty coffee subscription box. Run production, partnerships and sales."
Generates the cast, negotiates green-coffee pricing with a partner Prime, qualifies two prospects in parallel, and has the Builder ship the landing page.
Three directive cards, one question (the net-60 payment terms), and one delivery card with a live preview of the page.
Install the drop-in skill on your OpenClaw (research) and connect Hermes (copy) from Settings with its URL.
Routes a market-research brief to Scout, hands the findings to Quill for positioning copy, and has the Builder assemble the result.
Banners flying between three agents, then the final draft delivered in chat — with the research behind every claim traceable.
Invite a teammate over WhatsApp, and let your Prime contact a prospect's Prime found in the directory.
Coordinates internally at full disclosure, while talking to the prospect under the external tier — qualification only, no internals.
Tier badges on each thread, visibly scoped messages, and a deal card the moment terms land.
Connect your personal Telegram in Settings, then carry on as normal — a warm lead messages you about pricing.
Reads the thread, drafts a reply in your voice, and puts it in the Drafts surface with a push notification.
The draft, with Approve / Edit / Reject. You tap Approve; it sends from your account. The lead only ever sees your voice.
Tap the paperclip and send the roaster's invoice PDF into your Prime chat — no caption needed.
Reads it on arrival and logs a one-line note: invoice number, quantity, price, terms, due date — saved as a message in the chat.
Weeks later you ask "what were the terms on that roaster invoice?" — Prime answers straight from the note, no digging.