Preloop vs AWS Bedrock AgentCore: The open-source AI agent control plane

Preloop is the open-source AI agent control plane: an MCP firewall, AI model gateway, human approval system, and audit trail that you can self-host. Amazon Bedrock AgentCore is AWS's managed "platform to build, connect, and optimize AI agents at scale", with "any framework, any model" running on AWS. The honest one-liner: if you are building and hosting agents on AWS and want a managed runtime with policy built in, AgentCore is the path of least resistance; if you want a vendor-neutral, open-source control plane that governs agents wherever they run, including coding agents on laptops, choose Preloop.

At a glance

What Amazon Bedrock AgentCore does well

AgentCore has grown into a broad platform, and it is a credible choice when your team builds on AWS:

• Breadth. The pricing page lists 13 AgentCore features, and the FAQ covers Runtime, Memory, Gateway, Identity, Policy, Code Interpreter, Browser, Observability, Evaluations, payments, optimization, the managed harness, and AWS Agent Registry.
• Policy outside the agent. AgentCore Policy "integrates with AgentCore Gateway to intercept every tool call in real time" and enforces rules "outside the agent's execution boundary" (FAQ). That is the right architecture.
• Framework and model choice. No longer only Bedrock models: build with LangChain, the OpenAI Agents SDK, the Claude Agent SDK, Strands, or your own framework, "and deploy with any model" (AgentCore).
• Managed infrastructure. You do not operate a control plane. AWS runs the services, with consumption-based pricing.
• Enterprise trust surface. AWS security, compliance programs, and procurement are already accepted at most enterprises.

If you build and host agents on AWS, AgentCore is a reasonable starting point.

Where Preloop is different

Preloop is a different shape of product: an open-source control plane in front of agents you already run, wherever they run.

• Open source and self-hostable. Preloop's core is Apache 2.0. Run it in your own VPC, on-prem, or on any cloud. There is no forced dependency on a managed service.
• Agents you did not build. Much agent activity in a company is Claude Code, Codex CLI, Cursor, and similar tools on developer machines. preloop agents discover finds their configurations and preloop agents onboard routes the supported ones through Preloop, with backups and a dry-run preview. No SDK changes.
• A human can decide. When a rule requires review, the right people get paged on mobile, Apple Watch, Slack, Mattermost, email, or webhook, with the tool arguments and agent reasoning. Async mode keeps long reviews from blocking the agent's transport.
• Native agent actions. Native action gates for Claude Code, Codex CLI, Cursor, Copilot CLI, OpenCode, and others put shell commands and file edits on the same approval path as MCP calls. These never pass through a cloud gateway.
• Cost in dollars, per agent. Budgets and spend attribution per agent, flow, key, and session.
• Policy-as-code in YAML and CEL. Ordered, version-controlled rules that inspect tool arguments.

When to choose Amazon Bedrock AgentCore

• You build your own agents and want AWS to host and scale them.
• You want a fully managed service inside one AWS billing and IAM boundary.
• Automated policy at the gateway is enough, and per-call human approval is not a requirement.

When to choose Preloop

• Platform and DevEx teams rolling out Claude Code, Codex CLI, Cursor, Gemini CLI, OpenClaw, or OpenCode who need to govern what those agents do without changing developer workflows.
• Security teams that need an on-prem or VPC-deployable control plane with policy-as-code, approvals, and searchable audit evidence.
• Multi-cloud or on-prem organizations that do not want every agent action to depend on one cloud provider.
• Teams building EU AI Act readiness evidence that want matched policy, approver, inputs, and outcomes in a self-hostable platform.
• Open-source-first organizations that need a control plane they can read, fork, and audit.

Adopting Preloop from an AgentCore world

Preloop can coexist with AgentCore. Preloop's gateway supports Amazon Bedrock as an upstream provider with Bedrock API keys or IAM credentials, so model traffic can keep flowing to AWS while Preloop enforces approvals, tool policy, and budgets. A common split: keep server-side agents on AgentCore, put developer-side coding agents under Preloop, and send both sides' events to the same SIEM.

FAQ

Is there an open-source alternative to Amazon Bedrock AgentCore? Preloop is an open-source AI agent control plane under Apache 2.0. It covers gateway, policy, approvals, identity for managed agents, observability, and audit, and it is self-hostable. It is not a general-purpose agent hosting runtime.

Can I run AgentCore outside AWS? AgentCore's services run on AWS. Its Agent Registry can list agents hosted elsewhere, but if you need the control plane itself in your own VPC, on-prem, or another cloud, an open-source option such as Preloop is a better fit.

Does Preloop work with Bedrock models? Yes. Bedrock is a supported upstream provider, so teams can keep using Claude or other Bedrock-hosted models while Preloop handles tool governance, approvals, attribution, and audit.

How do I migrate from AgentCore to Preloop? For server-side agents, point your framework at Preloop's model gateway URL and register your MCP servers in Preloop. For developer-side agents, run preloop agents discover, then preloop agents onboard. Most teams start with observability and approvals, then move policy enforcement.

Which AI coding agents does Preloop govern? Claude Code, Codex CLI, Cursor, Gemini CLI, OpenCode, Hermes, OpenClaw, Windsurf, Copilot CLI, and any MCP-compatible client.

Try Preloop

• Start your free trial
• Book a demo
• Preloop on GitHub

Last reviewed: 2026-10-08.