Rylvo

Build Architect AI

Say what you want.
Approve what it builds.

Describe the agent you need in one message. The Architect reads your workspace, writes a plan you can check step by step, and builds it in Test, where no customer sees it until you release.

Start freeSee how it works
180
tools it can call across your workspace
29
ready agents it can start from
3
approval modes, from every step to end to end
2:17
to build our demo clinic agent, in 17 steps

Approval

Nothing changes until you say so.

Every request becomes a plan of named steps, each saying what it does and why. You choose how much of it waits for you.

Manual

Every step waits for you, lookups included. Approve them one at a time, or all at once.

ReadAsks youCreateAsks youEditAsks youDeleteAsks you

Auto-read

Default

Lookups run on their own, so the plan you review is only the changes. Anything that writes asks first.

ReadRunsCreateAsks youEditAsks youDeleteAsks you

Autopilot

Runs end to end. Deletes, costly runs and sensitive steps — credentials, invitations, sends, activations — still ask.

ReadRunsCreateRunsEditRunsDeleteAsks you

The server checks every call against the same rules, so a plan can never do more than its mode allows. Production is read-only. The Architect builds in Test, and changes reach customers only in a release you activate.

On every page

It knows where you are.

Open the Architect from the corner of any dashboard page. It already knows the page, the bot and the tab, so “add a field here” is enough. The change shows up on the page behind it, no reload.

Edit what exists

The smallest change that works.

Most of the work is changing what already runs. The Architect reads the bot and everything attached to it, changes only what the request needs, then checks it. Here, a fix for a flagged reply is replayed in Test before it’s marked done.

Reach

One conversation, the whole workspace.

The Architect works through the same 180 tools your AI editor can use, with your permissions and your plan's limits.

  1. 01Agents

    Bots, prompts and their versions, skills, session and memory settings.

    bots.create
  2. 02Knowledge

    Websites, files and connections, synced and tested for retrieval.

    kb.add_source
  3. 03Tools

    HTTP connectors, MCP servers and the tools a bot may call.

    connectors.create
  4. 04Conversations

    Flows, workflows and agent teams with routing and handoffs.

    flows.create
  5. 05Safety

    Guardrails, test cases, red-team attacks and edge-case fixes.

    guardrails.create
  6. 06Reach

    Channels, automations, broadcasts and alert rules.

    channels.create

Say it, or show it

Speak, and the words are typed for you to fix before you send. Or attach up to 4 screenshots of a page or a form and build from those.

A checklist of what you asked

Each chat keeps your requirements as a numbered list and marks which ones show up in the changes applied so far. Nothing you asked for quietly drops.

Three builds at once

Run up to three chats side by side. Each keeps its own draft, its own subject and its own place in the work.

Your model, your key

Pick the model it plans with, and it stays picked across chats. Every call runs on your own provider key. Rylvo adds no markup.

Questions

Asked often.

What can Workspace Architect create?
Anything in your workspace, through the same 180 tools an AI editor can use: bots, prompts, guardrails, skills, flows, workflows, knowledge sources, connectors, MCP servers, channels, automations, broadcasts, test cases, datasets, agent teams, alert rules, edge cases and red-team attacks.
Can it edit an existing workspace?
Yes, and that is most of what it does. It reads the bot and everything attached to it first, then proposes the smallest change that works: two lines of a prompt, one guardrail, one field. IDs, earlier versions, history and analytics stay as they were.
Does it change resources before I approve?
You choose. Manual waits for you on every step. Auto-read, the default, runs lookups on its own and asks before any change. Autopilot runs changes too, but deletes, costly actions and sensitive ones, such as credentials, invitations, sends and activations, always ask first. The server enforces the same rules, whatever the screen shows.
Can it change my live agents?
No. Production is read-only for the Architect. It makes every change in Test, where you can try it, and the change reaches customers only in a release you activate.
Can it build multi-agent systems?
Yes. It creates specialist bots, assembles them into an agent team with routing and handoffs, and runs a test conversation through the team before you use it.
Does Architect use my model key?
Yes. Planning uses your configured provider key, and during the 14-day Pro trial the $2 of AI usage Rylvo includes once your email is verified, so you can build before adding a key. Rylvo adds no inference markup. If there is no key and no trial usage left, Architect asks you to add one in Settings before it can plan.

Describe it today.
Run it this week.

Start your 14-day Pro trialUse it from your editor