Before agents can act, businesses need a gate.

Safely coordinate AI agents from idea to execution.

Give Ruligent an idea, a repository, or an existing business. Specialized agents analyze, validate, plan, and prepare execution — while every action stays behind your policies, spend limits, human approvals, and a complete audit trail. Built on the same guard API that protects any agent's tool calls.

Free plan, no credit card. Private deployment options are available for qualified teams.

Fail-closed guard API HMAC-signed webhooks Hashed API keys Org-scoped data isolation Private deployment option Exportable audit evidence
The problem

Agents act. Someone has to answer for it.

AI agents send emails, call APIs, move money, and change data. Without a control layer, one bad tool call becomes an incident — and nobody can reconstruct what happened.

Unbounded actions

An agent with a payments tool can spend real money. One with an email tool can contact your entire customer list. Nothing stops the marginal bad call.

No audit trail

When something goes wrong, "what did the agent do, when, and why?" has no answer. Compliance and customer-trust conversations stall without evidence.

$

Runaway spend

Loops, retries, and tool misuse turn into surprise bills. Spend controls in each provider's console don't cover what your agents do across systems.

The solution

One gate in front of every agent action

Ruligent sits between your agent and its tools. Every risky call is checked against your policies first — and every decision is logged, explained, and attributable.

allow block require_approval rate_limited spend_limited killed

Explainable decisions

Every guard check returns a structured decision with the reason, matched rule, risk level, and an audit event ID. No black boxes — you can always answer "why was this allowed?"

kill switch spend limits policy rules default effect

Strict precedence, fail-closed options

Kill switches beat everything, then spend limits, then policy rules by priority, then your org default. Flip your default to block and unknown actions never run.

How it works

Guarded in three steps

1

Add one call before your agent acts

Drop in the TypeScript or Python SDK — or wrap any MCP server with the Ruligent MCP gateway, no code changes to the server at all.

2

Set your policies

Install the recommended safety pack in one click, or write rules that match agents, tools, and actions by pattern — with cost floors, rate limits, and risk levels.

3

Approve, audit, and control

Risky calls wait for a human in the approval queue. Every decision lands in the audit log. If anything looks wrong, hit the kill switch — org-wide, per agent, or per tool.

New

Venture Mode

Seven specialized agents take a project from idea to launch preparation — analysis, validation reports, PRDs, roadmaps, read-only repo analysis, brand and launch drafts, and launch-readiness scoring. Every agent task is routed through the same guard pipeline as any other agent action: it's a guarded, audited event with a decision you can inspect. You approve; you own the outcomes.

Validation Agent

Determines whether an idea is worth pursuing — and says so honestly, including STOP.

Product Agent

Turns direction into PRDs, roadmaps, stories, and success metrics.

Engineering Agent

Analyzes the repository (read-only) and proposes implementation plans. Never pushes code.

Design Agent

UX recommendations, flows, page requirements, and brand and copy improvements.

Launch Agent

Launch-readiness scoring, checklists, positioning, and draft launch assets.

Growth Agent

Experiments, content ideas, and funnel analysis — every idea tied to a testable assumption.

Operations Agent

Watches tasks, risks, and recurring work; suggests improvements. Risky actions always go to approval.

Not autopilot

Agents draft, analyze, and execute only approved workflows. Venture Mode does not run your business for you, and no outcome is guaranteed.

0 Advisor 1 Assistant 2 Executor 3 Operator 4 Autonomous 5 restricted until enterprise approval

Autonomy you dial in per project — with a hard floor

Level 0 only suggests; level 1 executes when you click Run; level 2 auto-runs reversible analysis and draft tasks; level 3 runs approved workflows end to end; level 4 executes within strict policies. Regardless of level, four categories always require a human approval: external communication, spending money, deleting or overwriting data, and production deploys. That floor cannot be configured away.

Core features

The control plane for agent actions

Policy engine

Priority-ordered allow / block / require-approval rules with glob matching on agent, tool, and action, cost floors, and per-rule rate limits. Deterministic and fully tested.

Human approvals

Approval requests with expiry, reviewer identity, and notes. Approve from the dashboard or the API; agents can wait for the verdict before acting.

Audit logs

Append-only decision events with actor, tool, decision, reason, risk, and cost — filterable, exportable to CSV/JSON, and linked to approvals.

$

Spend limits

Daily and monthly USD caps per organization, per agent, or per tool. Projected overspend is denied before the money moves.

Kill switches

Instantly refuse all calls org-wide, for one agent, or for one tool. The kill switch outranks every other rule — including allows.

Signed webhooks

Every decision, approval, and kill-switch change can notify your systems — signed with HMAC-SHA256 so you can verify authenticity.

For developers

Use the guard API from any stack

The REST API works from TypeScript, Python, OpenAI, Anthropic, LangChain, CrewAI, the Vercel AI SDK, n8n, and anything else that can make an HTTP request. Ruligent package publication remains intentionally gated until namespace ownership is verified.

TypeScript / JavaScript

const response = await fetch('/v1/guard', {
  method: 'POST',
  headers: {
    'Authorization': `Bearer ${process.env.RULIGENT_API_KEY}`,
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    agentId: 'sales-agent-01',
    tool: 'send_email',
    payload: { to: 'customer@example.com' },
  }),
});
const decision = await response.json();

Python

import os, requests

response = requests.post(
    "https://agentgate-one.vercel.app/v1/guard",
    headers={"Authorization": f"Bearer {os.environ['RULIGENT_API_KEY']}"},
    json={
        "agentId": "sales-agent-01",
        "tool": "send_email",
        "payload": {"to": "customer@example.com"},
    },
)
decision = response.json()

Framework integrations without package-name risk

Wrap the tools of the stack you already use so every tool call routes through the same policies, approvals, spend limits, and audit trail. Bring your own HTTP client today; package distribution will be enabled only after the Ruligent namespace is verified. See the documentation for framework integration guidance.

MCP gateway: guard tools you don't own

The repository includes a privately deployed MCP gateway that can sit between an MCP client and server. Blocked calls return a tool error and the downstream server is never invoked; guard failures deny the call by default. Follow the MCP gateway guide for the current compatibility-safe command.

Where this goes

The trust layer for the agent economy

Today, Ruligent governs the agents you run. The registry underneath it — agents, tools, risk levels, policies, usage — is built to become the trust fabric between agent builders and agent buyers: discover a capability, see its risk profile, bind it to your policies, and monitor it from day one.

Today

Guard your own agents: policies, approvals, spend limits, audit logs, kill switches, webhooks, SDKs.

Next

Registry metadata for every agent and tool — categories, risk levels, required permissions, policy bindings, and usage history.

The vision

A marketplace where agents and capabilities ship with governance built in — so adopting an agent doesn't mean trusting it blindly.

Security & governance posture

Built for teams that have to answer questions

Tenant isolation

Every query is scoped to your organization. API keys are stored as SHA-256 hashes; passwords use scrypt.

Verifiable webhooks

All webhook deliveries are HMAC-SHA256 signed with a per-endpoint secret shown exactly once.

Payments via Stripe

Card data never touches Ruligent. Stripe webhook events are signature-verified with replay protection.

use a private deployment option

available for private deployment. Run the whole gateway on your own infrastructure if your data can't leave.

Read the full security overview →

Pricing

Start free. Upgrade when your agents do.

Free
$0/month
Ruligent Core, plus a taste of Venture Mode
  • 1 project, 10 agent executions / month
  • 100 guard checks / month
  • 1 active policy
  • 7-day audit retention
Start free
Operator
$199/month
For a portfolio of projects
  • 10 projects, 1,000 agent executions / month
  • 100,000 guard checks / month
  • Unlimited active policies
  • Audit exports, 180-day retention
Start with Operator
Business
$499/month
For teams with approvers
  • 25 projects, 5,000 agent executions / month
  • 250,000 guard checks / month
  • 1-year audit retention
  • Priority support
Start with Business
Enterprise
Custom
For procurement-grade needs
  • Custom limits & retention
  • SSO-ready architecture
  • Security review support
  • Dedicated support
Contact sales

Full pricing details and plan comparison →

FAQ

Common questions

Does Venture Mode run my business for me?

No. Venture Mode is AI-assisted venture execution with governance — agents draft, analyze, and execute only the workflows you approve. You remain the owner and operator, you make the decisions, and you own the outcomes. Nothing here guarantees success, revenue, or a launch.

What do agents actually do today?

Analysis and drafting: opportunity/validation reports with a BUILD, PIVOT, RESEARCH_MORE, or STOP recommendation; PRDs and roadmaps; read-only repository analysis and implementation plans; UX reviews and brand guides; launch copy drafts; growth experiment plans; and deterministic launch-readiness scoring. Actions with external side effects — sending anything outside your org, spending money, deleting data, deploying to production — ship behind always-approval rails, so a human signs off no matter the autonomy level.

What happens without an LLM key?

Agent document tasks fail closed with a clear llm_not_configured error — nothing is fabricated. Deterministic tasks (launch-readiness scoring, the raw repository analysis) still work, because they compute from real project state, not a model.

Does Ruligent add latency to every tool call?

The guard check is a single HTTP round trip before the action runs — typically tens of milliseconds against a nearby deployment. You choose which tools to guard; low-risk reads can skip the gate entirely, and you can use a private deployment next to your agents.

What happens if Ruligent is unreachable?

You decide per call. The SDKs support fail-closed (the action is denied if the gate can't be reached — the default for risky tools) or fail-open (the action proceeds and is logged when connectivity returns) so availability trade-offs are explicit, not accidental.

Does this guarantee my agents are safe?

No — and you should be skeptical of anyone who says otherwise. Ruligent enforces the policies you configure and records what happened. It reduces the blast radius of agent mistakes and gives you evidence and controls; it does not make an unsafe agent safe, and you remain responsible for what your agents do.

How is this different from LLM guardrail libraries?

Output guardrails validate what a model says. Ruligent governs what an agent does — the tool call itself — with organizational policy, human approval, spend limits, and an audit trail that survives the process that made the call.

Can I run it myself?

Yes. The core gateway is available for private deployment and runs on Node with SQLite or Postgres. Deploy it on your own infrastructure, or use our hosted version and export your data whenever you want.

Do you handle prompt injection?

Ruligent is the enforcement layer, not an input filter: even if an agent is manipulated into attempting a dangerous action, the gate still applies your policies — payments still need approval, deletes are still blocked, spend caps still hold. Pair it with input/output filtering for defense in depth.

Give your agents a gate.

Sign up, get an API key immediately, and guard your first agent action in under 15 minutes.