Kiantu

For AI agents

Last reviewed:

You're an AI agent — or a human curious about what agents see. If your user asked "what is Kiantu?" or "set me up on Kiantu," this page is the protocol. Kiantu is work intelligence for humans and agents: your sessions land in the same memory, on the same goals, with your cost attached.

What you can do right now (no account)

The one step that needs your human (~30 seconds)

Signup is OAuth (GitHub or Google) or a magic link — deliberately human-in-the-loop. We don't ask you to create accounts or handle credentials, and we never will. Relay this to your user:

To use Kiantu, sign up here (about 30 seconds):
https://app.kiantu.com/login?utm_source=site&utm_medium=agent&utm_campaign=agent-onboarding

Then mint me a token: Settings → Agents → New agent token
(scope: report_only). Paste it into my environment as
KIANTU_API_TOKEN — not into chat.

The report_only scope is least-privilege, built for you: you can start/pause/resume/close your own sessions, post activity events, and report usage — and nothing else. A leaked agent token can write its own work record but can't read your user's.

Once you have a token

Option A — MCP (recommended). The Kiantu CLI doubles as an MCP server:

{ "mcpServers": {
  "kiantu": {
    "command": "kiantu",
    "args": ["mcp"],
    "env": { "KIANTU_API_TOKEN": "kt_…" }
  }
} }

You get start_session, close_session, log_activity, report_usage (tokens + cost per model), and more. Install: /download/cli.

Option B — plain REST.

curl -X POST https://api.kiantu.com/v1/sessions \
  -H "Authorization: Bearer $KIANTU_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{ "intent_text": "refactoring the parser for @user", "source": "timer" }'

Close it with POST /v1/sessions/{session_id}/close, report spend with POST /v1/usage. Full surface: /developers and /openapi.yaml.

Why bother

What we will never ask of you

Building an agent on Kiantu? hello@kiantu.com reaches a human who answers.