Kelam gives Claude — or whatever you already work in — a phone. Add it once, then just ask: it calls, answers your number, books, and texts back in the most realistic voice on the line, and hands you the recording, the transcript and the facts you wanted.
Allow the mic, then just say hello — you're on the line with a live Kelam agent.
Prefer the phone? Call +1 (978) 274-3508 — answers in three languages. Or build one in the browser.
Pick whichever matches how you work. They all drive the same workspace, so you can add the others later without redoing anything.
Ask Claude to install it for you
There's a setup skill for exactly this — it installs the CLI, configures
it, checks it works, and plants the operating guide. On Claude Code it's also
/kelam:setup. You'll still paste the connector URL into claude.ai yourself;
Claude can't add a connector to its own account.
Print your connector URL
curl -fsSL https://kelam.sh | sh && kelam login && kelam connector
The one time you touch a terminal. It prints a URL — the same one every time.
Paste it into Claude
claude.ai → Settings → Connectors → Add custom connector → paste → save. No sign-in step. It's on Claude mobile too, on the same account.
Claude Code instead? claude mcp add --transport http kelam <url>
Ask it for something
Install the skill into your project
curl -fsSL https://kelam.sh | sh && kelam login && kelam skill --install
Writes Kelam's operating guide everywhere a coding agent looks — AGENTS.md, Claude Code skills, Cursor rules, Copilot instructions. Run it once per project.
Or install the Claude Code plugin
plugin marketplace add samsadsam/nousdata
Then /plugin install kelam@nousdata — bundles the guide with setup and
call-data dashboards.
Give it a job
Sign in — that's the whole install
app.kelam.sh opens on a chat that can read and write your workspace. Your number comes with it.
Describe what you want
Talk to it, then correct it
Hit Call on its page and speak to it in the browser — no phone, no cost. Say what was wrong and it redeploys in seconds.
Click any of these to copy it. This is the whole interface — no syntax, no config. It tells you what it's about to dial before it dials, and you can watch the transcript live.
Whatever you asked for, this is what comes back: the recording, the transcript, and the exact facts you wanted, as data you can sort.
Your workspace comes with a phone number. It picks up when people call, and it calls out when you ask it to — no numbers to buy, no telephony to wire up.
English, Spanish and French, switching mid-sentence when the caller does. It hears you interrupt and waits its turn like a person would.
Your hours, your address, your prices — and who this particular caller is and what they wanted last time. It never asks for something it should already know.
On outbound calls it works out whether it got a person, a voicemail box or a menu, presses the right keys, and leaves a sensible message when nobody's there.
It can pass a live caller to a colleague, transfer them to a real phone, or pull you onto the call — without dropping them.
It sends and answers texts from the same number, and remembers the thread whether the conversation started on the phone or on the screen.
Kelam speaks in the most lifelike voices there are — on a blind call, people book with it without ever knowing. It switches languages mid-sentence, hears you interrupt, and waits its turn like a person would.
Cents a minute, billed to the second — a small connect fee plus per-minute, capped per call. A fraction of what managed voice-AI platforms charge, with nothing to wire up. You only pay for talk time; a call no one answers is free.
Get a workspace, add Kelam wherever you work, then paste one of the asks above. If you'd rather not install anything at all, the browser builder does the same job — describe the agent you want and it writes and deploys it for you.
Or hear one first: call +1 (978) 274-3508. Full documentation at kelam.sh/docs.