signal busAll systems operationalScrums.com x Vercel for AI engineering ↗
Platform capability · Model & agent gateway

AI Agent Gateway

One integration for every AI model and every AI agent in your delivery pipeline. The gateway routes each call to the right model provider or deployed agent, enforces policy on the way through, and meters usage and cost per agent, team and model. A gateway for agents as well as models.

1Endpoint · models + agents
5SDLC stages routed
93%Less AI integration wait
100%Calls metered

Used by the world's leading companies

PPRONedbankHuaweiVolkswagenNaspersNetwork InternationalIAGNiveaInvestecBDOProsusIMIImperative
01

What is an AI agent gateway?

An AI agent gateway is a single governed endpoint between your applications and the AI they call. Every request passes through it, whether the target is a foundation model or a deployed AI agent. The gateway decides where the call goes, checks it against policy, records what happened and meters what it cost.

Most AI gateways stop at models. They put many model providers behind one API, which solves provider sprawl for prompts and completions. Engineering teams now also run agents: coding agents, review agents, SRE agents and the agents they build themselves. An agent call is a different kind of call. The agent holds context, uses tools and acts for a principal, so it needs identity, authority boundaries and an audit trail as well as routing. The AI Agent Gateway governs both call classes on one plane.

Scrums.com launched the gateway as Scrums.AI: an enterprise integration layer for discovering, integrating and coordinating AI agents across the software delivery lifecycle. It now runs as a capability of the platform, next to the AI agent harness that sets how each agent behaves in production, and the AI Catalog of deployable models and agents.

Model gateway vs agent gateway, defined

Model gateway

One API in front of many model providers. It normalizes request formats, fails over between providers and counts tokens. The call is stateless: a prompt goes in, a completion comes out.

Routes prompts and completions

Agent gateway

One governed entry point for AI agents. Each agent is a principal with its own identity, scope and credential. The gateway enforces what the agent may do and records every run it performs.

Governs runs, tools and authority

Routing

The decision about where each call goes: which provider, which model, which agent. Rules route on capability, latency, cost, security and data residency, and change without a code change in the calling application.

Set by policy, not hard-coded

Governance

Access control, approval gates and guardrails applied on the call path. A block set for the organization cannot be weakened by a team or by a single agent.

Enforced in one place

Observability

Every model call and agent run is logged with latency, outcome, tokens and the policy verdict, on the same governed data layer as the rest of delivery.

One audit trail for both call classes

Metering

Usage and spend recorded per call and attributed to an agent, a team, a model and an engagement. It is the base for budgets, chargeback and a clear answer to what each agent costs.

Per agent, per team, per model
02

Integrate once. Route to every model and agent.

Integrate your applications with the gateway once and reach the leading AI agents across your development pipeline. Each stage below lists the agent roles the gateway routes and governs.

Specification AI

Architecture agents, policy-enforcement agents and business-requirements agents turn intent into a specification the rest of the pipeline can use.

Coding AI

Code-generation agents, secure-coding agents and standards-compliance agents write changes and check them against your rules.

Infrastructure AI

Infrastructure-optimization agents, cost-efficiency agents and auto-scaling agents tune the environments your code runs on.

Deployment AI

CI/CD agents, config-validation agents and orchestration agents move changes to production through your quality gates.

Production AI

Monitoring agents, incident-triage agents and self-healing systems observe production and respond when it degrades.

Agents you build

Agents your teams build register with the gateway like any third-party agent: the same identity, the same policy, the same metering.

03

Models and agents in the gateway

A short list of live records from the AI Catalog. Models and agents sit in the same catalog and deploy through the same gateway. Browse the full AI Catalog for every class.

04

What the gateway governs

Models and agents share one control plane. Every control below applies on the call path, to both call classes.

/01

Unified agent management

One dashboard for every model and agent you run.

Models and agents in one inventoryAdd catalog agents in one stepStatus and health per agent
/02

Model routing

Route each call on capability, latency, cost and residency.

Many model providers behind one endpointProvider failover without app changesBuilt-in provider integrations or your own keys
/03

Agent collaboration (MCP)

Agents share context through the Model Context Protocol.

Built-in MCP supportAgent-to-agent context without custom glueTools exposed to agents under policy
/04

Security & governance

Central guardrails for every model call and agent run.

Access control per agent, team and modelApproval gates on sensitive actionsAudit trails and data sovereignty
/05

Cost transparency

Usage and spend per agent, per team and per model.

Cost per agent in one placeBudgets per teamUsage attributed to each engagement
/06

Observability

Every call recorded on one audit trail.

Latency, outcome and tokens per callPolicy verdict on every requestSame data layer as delivery telemetry
05

How the gateway works

Integrate once, then govern every model and agent from the same place. The same onboarding as the rest of the platform, because it is the same platform.

  1. Integrate once

    Point your applications, internal tools and CI at one gateway endpoint. No per-provider SDKs and no per-agent integrations.

  2. Add models and agents

    Connect model providers, add agents from the AI Catalog, or register the agents your teams build.

  3. Set routing and policy

    Choose routing rules per task. Set access, approval and budget policy per agent, team and model.

  4. Observe and meter

    Watch every call land with latency, outcome, policy verdict and cost. Tune routing and policy from the data.

06

AI Agent Gateway vs the alternatives

CapabilityAI Agent GatewayModels-only AI gatewayIn-house buildMarketplace tool
AI agent managementOne dashboard for every model and agentModels only; agents run elsewhereHigh build cost and complexityFragmented; each tool manages itself
Agent collaboration (MCP)Built-in Model Context Protocol supportOut of scopeHeavy build for inter-agent contextNo native agent-to-agent coordination
Model routingRoutes across model providers and agentsRoutes across model providersCustom routing to build and runTied to the vendor's own models
Cost transparencyUsage and spend per agent, team and modelToken spend per modelVisible, at build costPer tool only
Security & governanceCentral guardrails for models and agentsKeys and rate limits per modelCustom, high effortVaries by vendor
Maintenance & scalabilityRun by the platformRun by the vendor, models onlyOngoing engineering burdenOverhead per tool

A models-only gateway solves provider sprawl. An agent gateway also governs what agents are allowed to do.

07

Proof and resources

Partnerships, delivery outcomes and guides on AI agents in software delivery. View all AI resources.

Scrums.com partners with Windsurf

Partnership · AI coding agents in the delivery pipeline. Read the announcement

93% less wait and lead time for AI integrations

Outcome · measured across AI integration delivery. See AI development

Global team orchestration for a unicorn SaaS

Case study · PPRO, certified fintech unicorn. Read the case study

AI-powered anomaly detection for software maintenance

Guide · how AI detects anomalies before they become incidents. Read the guide

The emergence of AI-generated code

Guide · the challenges and opportunities of AI-generated code. Read the guide

The rise of AI governance platforms

Blog · why governance platforms keep AI systems secure and compliant. Read the post

08

What engineering leaders say

“Our project was incredibly successful thanks to the guidance and professionalism of the Scrums.com teams. The Scrums.com team often pre-empted and identified solutions and enhancements to our project, going over and above to make it a success.”

Volkswagen
Product Engineering Team
Read case study →
09

AI Agent Gateway. Frequently asked questions

What is an AI agent gateway?

An AI agent gateway is a single governed endpoint between your applications and the AI they call. It covers AI models and AI agents: it routes each call to the right provider or agent, enforces access and approval policy, records every call and run, and meters usage and cost per agent, team and model.

How is an agent gateway different from a model gateway?

A model gateway puts many model providers behind one API and routes prompts and completions. An agent gateway also governs AI agents, which hold context, use tools and act for a principal. Each agent gets its own identity, scope and audit trail. The Scrums.com AI Agent Gateway does both, so models and agents share one policy plane and one meter.

Who should use AI Agent Gateway?

AI Agent Gateway is for engineering leaders, CTOs, CPOs, CIOs, developers and business leaders who run multiple AI models and agents, and who need rapid deployment, data privacy and one place to control them.

Can I see what each agent and model costs?

Yes. The gateway meters every call and attributes usage and spend to an agent, a team, a model and an engagement. Track usage, monitor spending and manage budgets from the Scrums.com Software Engineering Orchestration Platform dashboard.

Do I need a ChatGPT/OpenAI account?

No. You don't need a separate ChatGPT or OpenAI account. The gateway connects to your preferred models through built-in provider integrations, or you can bring your own provider keys.

Which models and agents can the gateway route to?

Models and agents listed in the Scrums.com AI Catalog, such as the OpenAI, Anthropic Claude, Google Gemini, Meta Llama, Mistral AI and DeepSeek model families, and coding agents such as Claude Code, GitHub Copilot coding agent, Cursor Agent and Devin. Agents your own teams build register with the gateway in the same way.

What is Model Context Protocol (MCP)?

Model Context Protocol (MCP) is an open standard for how AI agents share context and reach tools, data and APIs. The gateway uses MCP so agents can access real-world data and APIs inside your software applications, under the same policy as every other call.

What was Scrums.AI?

Scrums.AI is the name under which Scrums.com launched its AI Agent Gateway and enterprise integration platform. The gateway is now a capability of the Scrums.com platform.

Is AI Agent Gateway priced separately?

No. AI Agent Gateway is a capability of the Scrums.com platform, included in the platform subscription alongside every other capability. There is no separate gateway product and no per-capability pricing.

010

More platform capabilities

Full platform overview →

Every capability below ships in the same bundled subscription. One platform, one plan, nothing sold piecemeal.

DORA MetricsAnalytics & telemetryDeveloper ProductivityEngineering intelligenceEngineering Capacity PlanningCapacity & forecastingAI Agent HarnessAgent infrastructureTool IntegrationsUnify 50+ dev tools into one data layerin buildAI Agent OrchestrationGoverned agents across the SDLCin buildCode Quality AutomationAI review & coverage enforcementin buildSecurity & ComplianceSOC 2, ISO 27001 & audit trailsin buildDevOps OrchestrationUnified pipelines & quality gatesin buildTalent OrchestrationElite engineers on the same planein build

One platform. One subscription.
AI Agent Gateway included.

AI Agent Gateway isn't a product you buy separately. It's one capability of the Software Engineering Orchestration Platform, bundled with every other capability on this page.