Keiki

Keiki is infrastructure for putting AI agents into the conversations customers and colleagues already have, rather than behind a chat widget nobody opens. You declare an agent once — a system prompt, a model, custom tools and feature toggles — and Keiki delivers it on SMS, iMessage, WhatsApp, Telegram, Slack, Microsoft Teams and email. Adding a channel later does not mean rebuilding the agent.

What the platform handles

The agent config stays small because the platform owns everything around it: inbound routing and outbound delivery per channel, conversation memory, steering when several messages arrive at once, browser automation, sandboxed code execution, billing and observability. Agents can retrieve live information, call business systems, take payment, remember context across a thread, and escalate to a person when policy requires it.

Developer surface

• SDK — @onkeiki/agents, with AgentKit as the primary API. The same config runs two ways: kit.sync() pushes it and Keiki runs the agentic loop, or kit.serve() runs the loop on your own infrastructure.
• Agents as models — call any agent through the OpenAI or Anthropic SDKs.
• Partner deployments — provision an agent per end user from your own backend, over REST or MCP.
• Capabilities — a versioned JSON manifest format that turns an ability into tools, prompt guidance, schedules and stored state. Runtimes include render-template, openapi, codemode and mcp. Every built-in Keiki feature is a capability in the same format, and developers can publish their own to a marketplace that currently lists 10 addons.
• Testing and operations — Eval Studio for ranking candidate models before a change ships, Replay for testing a prompt change against turns that actually happened, external-channel evals that grade what a real contact received, agent swarms, agent wallets that put a spending policy in front of your payment account, and organization-scoped SAML 2.0 SSO.

A catalogue of 13 reusable agent templates, an OpenAPI document and a hosted MCP endpoint are all published publicly, and the platform advertises SOC 2 and HIPAA compliance.

Who it is for

Product and engineering teams shipping customer-facing or internal agents that must work over real messaging channels and do more than answer a static FAQ. Keiki does not publish a public price list; sign-up is self-serve and a guided onboarding call is offered.