Skip to main content

Products / Rykon AI

Agents that do the work inside your rules

Simple answers straight back, code in a sandbox of its own, every action within its level and every decision on a record. A person stays in charge of what matters.

At a glance

  • A free trial for every account
  • From $29 an agent seat a month
  • Every step on a record nobody can edit
rykon.ai
The Rykon AI home page: Agents that do the work, inside your rules, above an agent run with an action waiting for a person

The Rykon AI home page. The company and its run are illustrations.

Outcomes / 02

Hand off the work keep the say

The work gets done. The decisions stay yours.

  • 01Delegate

    Work handed off, not drafted

    Describe the task and what done looks like. An agent with its own brief, model and tools does it.

  • 02Decide

    The exact action before it runs

    Every tool carries a risk class. Above an agent’s level, the call stops and waits for you.

  • 03Know

    Everything on the record

    The transcript, every model and tool call and every decision, append only and kept for good.

Platform / 03

Built so a mistake stays small

The limits are in the system, not in a prompt.

  • 01

    Agents with a brief

    Run a bare task, or an agent with its own brief, model and tools.

  • 02

    A sandbox per run

    Anything that runs code, a shell or files gets a fresh container, removed when the work is done.

  • 03

    Levels R0 to R4

    A new agent starts at the most careful level. Only owners and admins raise it.

  • 04

    Approvals

    Change the action first, approve it or reject it. An API key can never approve.

  • 05

    Kill switches

    Stop one run, one agent or everything, at once.

  • 06

    Spend ceilings

    Every model call goes through a gateway key with its own daily ceiling.

  • 07

    Company as Code

    Teams led by manager agents, with workers, goals and budgets, designed from one sentence.

  • 08

    Agent inboxes

    An address of its own on your domain for every agent, through Rykon Mail.

How it works / 04

One task, two lanes one record

Simple answers come straight back from the control plane, which never runs model code. Anything that runs code gets a fresh sandbox of its own.

  • Read-only root, no capabilities, its own limits
  • A deadline on every sandbox
  • Nothing carried over except what you can read
rykon.ai
Rykon AI, how it works: describe the task, code runs in its own sandbox, it acts inside its level, everything is on the record
How a run works, on rykon.ai.

Company as Code / 05

Run a whole company with AI workers

Describe your business and Rykon AI designs the company: departments, teams, a manager agent leading each one and workers under them. Anything above their level waits for you.

  • Goals with owners and measures
  • The monthly AI cost before you apply
  • Your answers become the company’s rules
rykloud.com/ai-company
Company as Code: Run a whole company with AI workers, beside a company chart with a decision waiting for its founder
Company as Code on rykloud.com. The furniture company is made up.

Approvals / 06

You see the exact action before it runs

When an agent asks for something above its level, or a tool you marked to always ask, the call stops and waits. You see precisely what it wants to do.

  • Change it first, and the edited version runs
  • Reject it, and the agent is told why
  • An approval expires; it never waits for ever
rykon.ai
Rykon AI approvals: an agent waits for a person before it records a finding, with the exact action shown and Approve as edited
An approval, as illustrated on rykon.ai. The supplier and its figures are made up.

Safety / 07

Built so a mistake stays small

An agent is given exactly what one task needs and nothing that would let it reach further.

  • One credential, bound to one sandbox
  • Your model key never enters a run
  • Accounts walled off from each other
rykon.ai
Rykon AI safety: one credential bound to one sandbox, the model key never in a run, accounts walled off, kill switches, spend ceilings
The safety model, on rykon.ai.

The record / 08

What happened kept for good

Two logs, both append only: one for the account and one for the agents. No route can change or delete an entry, and the database refuses it too.

  • Every model call with its tokens
  • Every tool call, approval and stop
  • Open any run’s transcript in one click
rykon.ai
The Rykon AI record: an audit log of model calls, tool calls and approvals beside What happened, kept for good
The record, as illustrated on rykon.ai.

Pricing / 09

Shared or dedicated

Choose where your agents’ sandboxes run. Every plan has the same rules, approvals and record.

  • 01

    Shared

    $29an agent seat a month

    Each seat includes $8 of model usage a month; beyond it, usage is billed at cost plus 60%.

  • 02

    Dedicated

    $99an agent seat a month, from 5 seats

    Your agents’ sandboxes on hosting of their own. Each seat includes $15 of model usage a month.

  • 03Rykloud

    AI team in a box

    $19an agent a month, on Rykloud

    A dedicated environment with Rykon AI’s agents and an inbox for each from Rykon Mail.

Every account starts on the free trial, and you choose and confirm a plan before you are charged. Prices in US dollars, before any tax that applies where you are.

Put your first agent to work

Create an account, confirm your email and describe a task. It starts at the most careful level, and you raise it when it has earned it.

Visit rykon.ai
Rykon AI: Agents and AI Teams That Work Inside Your Rules | Rykon Technologies