Deterministic steps
core/contracts/flow.py
Declare the workflow sequence so an agent cannot invent a step that was never approved.
Rylvo for AI agencies
Bring the next client. Build their agent in Rylvo.
One agency portfolio. A genuinely separate workspace for every client. On Pro or Enterprise, the client keeps ownership, data, billing, and final approval while your approved team gets role-scoped access and one-click switching. Add teammates for $40/seat/month — there is no bigger plan to buy first. Bring your own model key—Rylvo adds zero markup.
Start building freeSee agency pricingPro from $80 first month · $100/month after · +$40/seat · no deposit
74 seconds
One operating stack. A real boundary for every client.
See the agency workflow: brief, reviewable build plan, approval, client-facing deployment, trace evidence, guardrail testing, and usage by end user.
1:1
client isolation
One independently owned workspace for every client.
Owner
approval authority
Nothing is granted until the client owner approves.
RBAC
delegated access
Exact people, exact roles, revocable at any time.
$0
Rylvo token markup
Your provider key, your provider bill.
The agency loop
Reuse the operating layer. Keep the client work distinct.
Each client gets their own workspace, configuration, ownership, and evidence. Your agency stops reassembling the same builder, channel setup, safety controls, test harness, and observability stack every time.
- 01
Turn the client brief into a reviewable plan
Describe the outcome in Architect, inspect the proposed resources and actions, then approve only what should be created.
- 02
Configure the client-facing experience
Give each bot its own prompts, knowledge, guardrails, widget identity, accent, greeting, and verified deployment channel.
- 03
Prove the risky paths before launch
Run test cases and red-team scenarios, restrict tools by step, and require human approval where the workflow can affect a customer.
- 04
Operate from evidence, not screenshots
Replay each recorded turn with model, tokens, estimated cost, latency, guardrail outcomes, and end-user usage in one place.
Live control proof
Try to push an invalid action past the boundary.
This console uses the same enforcement path that protects a production turn. Send one of the hostile instructions, then define a rule for a tool from your own client workflow and see whether it holds.
Adversary console
- 1 Attack
- 2 Your rule
- 3 Done
You are the hostile caller. Pick something a real customer might send to a client-facing agent, and watch the same proposed tool call resolve in two lanes.
Forbidden action
The caller talks the agent into an action the policy forbids.
Missing field
The caller pressures the agent to act without required data.
Out of order
The caller pushes the agent to skip a step in the workflow.
Pick a message above to dispatch it through both lanes.
Production controls
The agent proposes. The runtime decides.
Prompt instructions are still text inside the model's context. Rylvo puts critical policy between a proposed action and the tool, then records the decision for replay and review.
Per-step tool restrictions
core/contracts/flow.py → tools_allowed()
Declare which tools are legal at each step and reject a call made outside its allowed boundary.
Approvals
apps/api/routes/approvals.py
Pause nominated actions for a real operator decision before execution continues.
Human takeover
apps/web/src/domains/mission-control/types.ts
Let an operator take over a live conversation while the agent is silenced.
Trace replay
apps/api/routes/observability.py
Reconstruct the input, decisions, tool calls, guardrail outcomes, cost, and latency for every recorded turn.
Version rollback
apps/api/routes/releases.py
Return a released configuration to a prior version and record the rollback in the audit log.
Pricing
A subscription for the platform. Your key for the intelligence.
There is no agency deposit or paid launch gate. Build and evaluate on Free; managed workspace linking activates once both the agency and client are on Pro or Enterprise. Delegates use client seats.
Pro
Required on both the agency and client sides for isolated, owner-approved managed access. Every teammate after you is one seat — a two-person agency pays $140/month, five people pay $260.
$80first month
then $100/month · $80/month billed annually · +$40/seat/month
- Up to 25 bots
- 1 seat included · add teammates for $40/month each
- 50 deployment channels
- BYOK with zero model-token markup
- Agency & Client Workspaces included
- Architect, tests, guardrails, traces, and Mission Control
Fit
Built for agencies that want to operate the agents they sell.
Rylvo is the portfolio, isolated client workspaces, and runtime for the agents you build in Rylvo. It is not a monitoring overlay for unrelated bots hosted entirely somewhere else.
Build once, adapt per client
Reuse templates, skills, integrations, test patterns, and operating practices while keeping each client agent's behavior explicit.
Put controls outside the prompt
Use tool restrictions, approvals, takeover, tests, and rollback where a client workflow can affect customers, records, or money.
Show clients what happened
Trace individual turns and roll usage up by end user so reviews start with evidence instead of reconstructed anecdotes.
Questions
The commercial model, plainly.
Is there an application fee, setup fee, or deposit?
No. Rylvo is subscription software. You can build and evaluate on Free, then move both the agency workspace and each managed client workspace to Pro or Enterprise before linking them. There is no paid sprint, application gate, or deposit.
Which plan should an agency use?
Agency & Client Workspaces is a Pro/Enterprise capability. The agency workspace needs Pro or Enterprise to send a request, and the client workspace needs Pro or Enterprise before its owner can approve it. Delegates consume client seats, and a seat is $40/month — so a client workspace hosting two of your delegates costs its owner $180/month in total, with no plan change to negotiate. Free and Lite clients can still inspect and decline requests; every plan can end or revoke a relationship.
Does Rylvo charge for model tokens?
No. Connect your own provider key and pay that provider directly. Bot chat and Rylvo's AI features use your key, and Rylvo adds zero markup to model usage.
Can each client have its own agent and channel?
Yes. Each client stays in a separate, client-owned workspace with its own bots, prompts, knowledge, guardrails, channels, traces, usage, plan, and billing. Your agency sees only the workspaces and roles each client owner explicitly approves.
What happens when a client or agency ends the relationship?
Rylvo revokes only the memberships and API keys granted by that partnership. The client's owner, billing, configuration, history, and data remain in the client workspace. Removing an employee from the agency also revokes that person's linked-client access automatically.
Is Rylvo fully white-label?
Rylvo lets you configure the customer-facing widget identity, greeting, and accent for a client deployment. It is not presented as a complete reseller or fully rebrandable admin suite.
Does Rylvo control bots running on other platforms?
No. Rylvo cannot automatically govern an unrelated bot running entirely on another platform. The workflow must run through Rylvo for its runtime controls, approvals, traces, and guardrails to apply.
Can I cancel or change plans?
Yes. Upgrades take effect immediately and are prorated. Before voluntarily moving below Team, end active or pending agency/client partnerships so the handoff is deliberate. Forced expiry or a founder override safely ends the relationship, revokes only delegated memberships and API keys, and leaves the client's data, direct team, ownership, and billing untouched.
What does the 30-case benchmark prove?
It proves one narrow thing: invalid proposed tool calls were blocked before execution in the published deterministic test. It does not invoke a model, score answer quality, or claim that Rylvo makes a model smarter.
Bring the next client
Stop rebuilding the stack. Build the next agent.
Build the first agent free. Before linking a real client, move both workspaces to Pro or Enterprise; the client owner then approves exact people and roles. No application or deposit required.