Self-hosted AI agents, in production · installs in fifteen minutes
Rikoo
Open coreSelf-hosted

Put your AI agents in production. On your infrastructure.

Rikoo is not another assistant or another model. It is the layer that was missing between an agent prototype and a service you can hand to customers, employees and auditors: organizations, immutable versions, a secret vault, per-run isolation, traces and costs, installed in your own environment, with the language model of your choice.

34
Model providers
0
Data sent to the vendor
9
Services to operate
Air-gap
Supported deployment
Five promises

Things you can say in front of a security officer.

  • Your agents run at your place

    One command starts the whole platform on your servers or your cloud. No data, no secret and no conversation transits through a vendor, and nothing is reported back to us.

  • No model is ever imposed

    OpenAI, Anthropic, Mistral, OVHcloud, Scaleway, DeepSeek, Groq, or your own model served by vLLM or Ollama. Switching provider does not mean rewriting an agent, and an automatic fallback takes over when one goes down.

  • Governed from day one

    Organizations, projects, roles, immutable agent versions, a secret vault, budgets and an audit log. What an agent did, on which version and at what cost, is verifiable after the fact.

  • Safe by construction, not by instruction

    Every run is confined to its own sandbox. All egress passes an anti-SSRF guard. API keys are never readable again. Documents, web pages and tool output are framed as untrusted content before they reach the model.

  • Open core, no lock-in

    The core is AGPL-3.0. Read it, fork it, audit it. The enterprise modules are separable by design, and the platform keeps running without them.

  • Observable in production

    Every run emits OpenTelemetry traces to your own collector. Costs, latencies and failures land in the tooling your team already operates.

One brain, many surfaces

One brain, many surfaces.

An agent is written once, then exposed where your users already are: a web widget, a Slack app, or a public REST API. The version they talk to is pinned, so a change in the Studio never silently changes what a customer sees.

  • Web widget and JavaScript SDK
  • Slack app with signed ingress
  • Public REST API, OpenAPI described
  • A pinned version per environment, test and live

Module by module

Companies have agent prototypes, not agents in production. The gap is not the model. It is everything around it.
RikooWhy we built this
Get started

Install, publish a version, expose a surface.

Nine services, one command, fifteen minutes. No account, no key, no call with us.