Rylvo, Inc.

Workspace Architect · MCP-native builder

Build and edit your entire AI workspace.

Q: Can it edit an existing workspace?

Yes. The Architect first inspects existing resources and their wiring, then proposes targeted updates such as revising a prompt, adding a guardrail, attaching a connector, changing session configuration, updating a workflow, or reorganizing an agent group. It does not need to rebuild the system from scratch.

Q: Does it change resources before I approve?

The Architect separates inspection, planning, and execution. You see the proposed tool calls and rationale before execution. Destructive and costly actions require explicit approval, while server-side authorization, quotas, and audit logging apply to every tool call.

Q: Can it build multi-agent systems?

Yes. It can create specialist bots, build agent groups, add or remove bots, define flows and handoffs, inspect routing insights, export or import group blueprints, and run a test conversation before production use.

Q: Does Architect use my model key?

Yes. Planning uses your configured provider key. Rylvo adds no inference markup. If the workspace has no provider key, Architect asks you to add one in Settings before it can plan.

Tell Architect what you want to create—or what needs to change. It inspects the live workspace, turns your request into typed MCP actions, shows you the plan, and follows the approval mode you choose.

Open Workspace Architect See how it works

129

MCP tools

starter systems

risk classes

audit trail

Workspace Architect

MCP connected

Ready

Conversation

Build a support bot with grounded answers, a PII blocker, billing escalation, and Slack.

I inspected the workspace and found no existing support bot. I prepared five connected actions for review.

1 bot1 KB1 guardrail1 workflow1 channel

Describe what you want to build or change…

Proposed plan

5 typed actions · review before apply

Pending approval

Customer support system

templates.apply_bot_template

write

Help center knowledge

kb.create_connection

write

PII leakage blocker

guardrails.create

write

Billing escalation

workflows.create

write

Slack deployment

channels.create

write

Approve and apply 5

Reject all

How it works

Conversation becomes a governed change set.

Architect never treats a natural-language request as permission to mutate blindly. Inspection, planning, review, execution, and audit stay separate.

Inspect

Reads the workspace, plan limits, templates, and existing wiring.

Plan

Maps your request to typed MCP tool calls with arguments and rationale.

Review

Shows every proposed action. Approve or reject items individually.

Apply

Executes only the approved calls, one by one, against live server rules.

Audit

Reports each result while authorization, quota, and audit controls stay active.

What it can change

Create new. Edit existing. Connect everything.

This is not a one-time bot generator. Architect can inspect and evolve the resources already running in your organization.

Browse the full MCP tool surface

Agent foundation

Bots · prompts · guardrails · skills

InspectCreateEditDelete

Change a bot model, revise its system prompt, or add a policy guardrail.

Knowledge and tools

Knowledge sources · connectors · MCP servers

InspectCreateEditAttachSync

Add a source, update connector auth, sync knowledge, or attach an MCP server.

Orchestration

Workflows · flows · agent groups · handoffs

InspectCreateEditAttachTest

Build a specialist group, attach a flow, and test routing before launch.

Deployment and runtime

Channels · API keys · session config · storage sync

InspectCreateEditActivate

Create a channel, change session behavior, export a bot as MCP, or enable BYODB sync.

Quality and learning

Test cases · datasets · edge cases · evolution

InspectCreateEditRunReview

Create regression tests, preserve traces, configure optimization, or run a review.

Operations and reach

Automations · alerts · broadcasts · notifications

InspectCreateEditSchedule

Schedule a task, create an alert rule, prepare a broadcast, or wire notifications.

What you can ask

Build, repair, or expand— in plain language.

You ask

“Build a customer support system for a fintech app. Ground answers in our help center, block PII leakage, escalate billing disputes, and deploy to Slack.”

Architect turns the outcome into a connected build plan.

A connected, editable workspace—not a generated code snippet.

Example plan5 actions

templates.apply_bot_template

Start with the support system

kb.create_connection

Connect the help-center knowledge

guardrails.create

Block sensitive-data leakage

workflows.create

Route billing disputes for escalation

channels.create

Prepare the Slack deployment

One tool surface

The dashboard and your editor speak the same MCP.

Workspace Architect is not a separate automation engine. It calls the same organization-scoped MCP endpoint available to Claude Code, Cursor, Windsurf, Zed, and other compatible clients.

129 typed tools from one canonical registry
Short-lived OAuth tokens bound to one organization
The same authorization, quotas, and audit rules everywhere

Connect an external editor

Dashboard

Claude Code

Cursor / Zed

Rylvo MCP

129 typed tools

OAuth · org scoped

Bots

Prompts

Knowledge

Tools

Tests

Ops

Safety and control

Natural language in. Typed, governed actions out.

Each MCP tool has a canonical risk policy. Unknown tools fail into a cautious approval posture, and enforcement happens again on the server when the call executes.

Your model key, your provider bill

Architect planning uses your configured BYOK provider. Rylvo adds no inference markup. Workspace mutations remain subject to plan limits, authorization, and rate limits.

Read

Workspace discovery and inspection. Safe to batch without approval.

Write

Creates and updates resources through typed arguments and server rules.

Destructive

Deletes and resets require explicit per-item approval and confirmation.

Costly

Token-intensive test suites, group tests, and evolution reviews require approval.

Approval modes

Manual · default

Approve every action

Auto-read

Run inspection automatically

Autopilot

Run safe reads and writes; stop for destructive or costly work

FAQ

Before you let it build.

It can create and connect bots, prompts, guardrails, skills, workflows, flows, knowledge sources, connectors, channels, automations, broadcasts, test cases, datasets, agent groups, API keys, Mission Control alert rules, edge cases, and more through the same governed MCP tool surface.

120+ tools · one governed builder

Tell it what needs to exist next.

Start from a template, build a custom system, or ask Architect to inspect and improve what is already running.

Open Workspace Architect Explore 29 starter systems