Runlayer
Also known as: Runlayer MCP Gateway, Runlayer Agents
Runlayer gives enterprise teams one place to connect AI clients to approved tools and run agents with their own identities. Its MCP gateway enforces access and security policy when an agent uses a tool. Its runtime supports scheduled and event driven work, scoped memory and tests against recorded runs. Security staff can review requests and trace which actions were allowed or blocked.
Runlayer sells an operating layer for AI, a control plane for enterprises that need to connect AI clients and agents to business systems under security and governance rules. Its Catalog lists the connectors, skills and agents an organization has approved. The MCP Gateway can register an organization’s existing servers, provide built in connectors or host the buyer’s own MCP server.
An administrator can grant access to specific tools or resources using identity, group, client and runtime conditions. Employees can use approved connections from Claude, ChatGPT, Cursor, Codex and other supported clients. Slack, Salesforce, GitHub, Box and many other business systems can be connected through the catalog.
With Runlayer Agents, an employee can describe work and assign approved connectors, model settings and skills. Agents can run on a schedule, from a webhook or when a connected system produces an event. An agent can delegate subtasks to child agents within its own runtime. Its memory can keep notes for individual users, Slack conversations or the agent as a whole. A built in browser can interact with websites. Staff can replay previous runs with a different prompt or model and compare the results with a control replay. Simulated MCP results let teams test changes without sending those tool calls to production services.
Platform owners can use run histories and traces to track which tools an agent called, its policy result, token usage and costs. Runtime Security scans requests and results for personal data, data exfiltration and drift from an agent's intended task, in alert only or blocking mode, and risky tool calls can require human approval. Runlayer Watch uses device management data to surface AI clients and connections that are operating outside approved channels.
The platform can run as a managed service or in a customer’s AWS account. Self hosted installation options include ECS Fargate with Terraform and Kubernetes. Runlayer has a SOC 2 Type II report and is HIPAA and GDPR compliant. Administrators can choose which providers or models are available by role. When a buyer links its Claude organization, Anthropic bills that organization directly for Claude usage.
Decagon, Lemonade, Gusto, Opendoor, AngelList, PagerDuty, AppLovin and Baseten use Runlayer. Gusto brought controlled AI access to more than 3,000 knowledge workers with it, and Homebase uses it to roll out AI tools without a lengthy security review for each one. Jane App moved more than 800 knowledge workers from experimentation to production workflows, using AppSec review to approve and publish new connectors.
A platform team can similarly publish a customer data connector with access limited to approved groups, and an operations manager can build a recurring agent to use that connector for a weekly summary. Runlayer quotes the deployment and workspace, while model usage follows the buyer’s configured provider or billing arrangement. The company raised $30 million from Felicis and Khosla Ventures in its Series A, bringing total funding to $42 million.
Vendor details
Canonical URL
https://www.runlayer.com/
Category
Agent infrastructure
Subcategory
Governed agent runtime and MCP gateway
Funding status
Raised $42 million by its $30 million Series A from Felicis and Khosla Ventures.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Claude, ChatGPT, Cursor, Codex, Slack and enterprise MCP connections such as Salesforce, GitHub and Box. Supports custom MCP servers, a management API and a run API.
In practice
An enterprise grants staff access to an approved connector while limiting certain operations to an authorized team.
An operations lead creates an agent that gathers approved project information every Friday and posts a summary in Slack.
A maintainer replays past runs with a different model and compares output quality and cost before changing the agent’s settings.
Sources & related URLs
Research sources
Agentic Index coverage score
12.5 / 14 capabilities · 89%
| Integrations & Tool Calling | Full |
|---|---|
|
The gateway registers approved MCP servers and can host custom connectors. Agents use these connections to perform actions in business systems under scoped credentials. A platform team can allow a person or an agent account to use specific operations on a connected service. For example, the catalog includes Outlook operations for working with drafts and folders and tools for modifying approved Okta groups. Sourcerunlayer.com/mcp-gatewayread 2026-11-20 |
|
| Workflow Orchestration | Full |
|
A Runlayer agent can delegate part of its work to a child agent with its own instructions and an optional model choice. Child agents use the parent’s authorized connections inside its sandbox. An owner can give a recurring agent procedures that combine steps into a reusable tool and control how it follows up on work that will finish later. The main run keeps child activity in its timeline. Sourcedocs.runlayer.com/platform-agentsread 2026-11-20 |
|
| Knowledge Grounding & RAG | Partial |
|
Runlayer agents can query approved MCP sources for context during a run. A workspace can connect records and files from products such as Box and Microsoft 365 or supply an internal tool for a specific data service. Access follows identity, resource and policy checks. The available material depends on the connectors that the team has attached to the agent and on the credentials in use. Sourcerunlayer.com/mcp-gatewayread 2026-11-20 |
|
| Human Oversight & Guardrails | Full |
|
When users request a connector or request access to an existing one, an administrator can review the pending request and approve or reject it in Runlayer. Platform owners can also require human approval for sensitive use. They can grant access to an approved group and apply policies to the connector’s tools to restrict what the assisted workflow can do once access is approved. Sourcedocs.runlayer.com/platform-approvalsread 2026-11-20 |
|
| Security, Identity & Governance | Full |
|
The gateway authorizes calls by user, role, group, agent account and other policy conditions. Administrators control which tools and resources an identity can reach, and connected services can use scoped OAuth access. Agents can have distinct identities so a team can manage their permissions independently of employee access. Runlayer has a SOC 2 Type II report. Sourcerunlayer.com/mcp-gatewayread 2026-11-20 |
|
| Observability & Auditability | Full |
|
Run histories and activity records identify the actor, connection, tool and outcome. Security scans can accompany an event, and audit records can be filtered for actions such as an approval or resource change. A team can open a run’s execution trace to follow its work and inspect token and cost usage. Customers can export audit events to storage they use for security monitoring. Sourcedocs.runlayer.com/platform-audit-logsread 2026-11-20 |
|
| Memory & State Persistence | Partial |
|
Agents can save facts in dedicated notes or store structured state in Workbench memory. The workspace separates agent, user and channel scope. Later runs can retrieve matching notes and reuse saved preferences or prior results. Users can edit and delete their own notes, and owners with access can manage an agent’s notes. The agent can retire a note when its underlying fact changes. Sourcedocs.runlayer.com/platform-agentsread 2026-11-20 |
|
| Deployment & Data Residency | Full |
|
The platform is available as a hosted Runlayer deployment or in the buyer’s AWS account. Self hosted options include ECS Fargate provisioned with Terraform and Kubernetes installation. The agent runtime can use managed or customer hosted infrastructure, depending on deployment. Platform owners attach the connections and credentials appropriate to that environment. Sourcedocs.runlayer.comread 2026-11-20 |
|
| Prebuilt Agents / Templates / Packs | Partial |
|
The catalog stores skills, plugins and connectors that an organization approves and distributes to teams. A buyer can package a skill with related capabilities, identify an owner and control which people or agents can use it. The registry lets users install the capabilities their organization has made available, alongside the skills configured for an individual agent. Sourcerunlayer.com/catalogread 2026-11-20 |
|
| Triggers & Channel Coverage | Full |
|
Runs can begin through a schedule, an incoming webhook or an event in a configured connector. A user can invoke an agent from a supported client or start it through the application. The agent can also arrange a later continuation when a deployment or another task is still in progress. An account running in the background uses the agent’s own authorized context. Sourcedocs.runlayer.com/platform-agentsread 2026-11-20 |
|
| Model Flexibility & Routing | Full |
|
The model picker lets an agent owner select an approved model from configured providers. A workspace can use Anthropic, OpenAI, Google, Amazon Bedrock or an OpenAI compatible gateway. An administrator can set the recommended workspace model and restrict options by role. Where a provider supports it, the owner can also change settings such as the model’s reasoning effort. Sourcedocs.runlayer.com/platform-agentsread 2026-11-20 |
|
| APIs / SDKs / MCP Extensibility | Full |
|
The management API lets outside applications manage MCP resources and retrieve activity. An authorized caller can use the agent run API to start work, check its state, cancel it or continue the conversation. Runlayer also exposes its own MCP capabilities for managing agents. API credentials have account scope, and a personal key uses the owner’s access and connected accounts. Sourcedocs.runlayer.com/platform-apiread 2026-11-20 |
|
| Testing, Debugging & Optimization | Full |
|
Offline evaluations can replay a selected run with a new prompt or model and compare the candidate to a control using the captured original setup. A maintainer can inspect the output and trace for both replays, then record a verdict. The default simulates MCP results and prevents memory writes. Scorers can measure goal completion, instruction adherence and changes in cost or runtime. Sourcedocs.runlayer.com/platform-agent-evalsread 2026-11-20 |
|
| Browser / Computer-use | Full |
|
Each agent can open the built in browser to interact with external websites. A buyer can set rules for hosts the sandbox is allowed to reach on supported deployments. Optional managed browser sessions are available on the hosted service when enabled, with Runlayer operating the provider account. An agent can use a browser during a run and pass the resulting material into its output. Sourcedocs.runlayer.com/platform-agentsread 2026-11-20 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
Quote for the Runlayer platform
Platform contract and model provider usage
What is public
Runlayer quotes a deployment. Anthropic can bill a linked Claude organization directly for Claude usage from Runlayer Agents under its existing terms.
Billing mechanics
The configured model provider charges for agent usage. Offline evaluations can add replay, scoring and tool simulation calls. A linked Claude organization keeps Anthropic billing.
Variable cost rationale
Provider usage depends on the models and number of agent runs. Offline evaluations can run an additional control and candidate for each selected case, with each model call billed by the configured provider.
Sales call required
Yes, required for paid access
Commercial notes
Organizations can use provider credentials or link their Claude Platform organization. The platform supports managed and customer hosted deployments.
Missing data
Runlayer platform rates and any included usage are set in the contract.
Related vendors
- AgentOps — Agent observability and debugging platform: open source SDKs trace…
- AgentPhone — AgentPhone gives the AI agents a team builds their own US or…
- Agno — Python agent framework and AgentOS runtime (formerly Phidata) for…
- Airbyte — Airbyte moves data from operational systems into warehouses and…
- AIsa — Resource and payment gateway for AI agents, with one key to 110+…
- Albato — Albato is a no code integration platform that sells an embedded,…
Alternatives to Runlayer
Other products that do the work Runlayer does, and what each one is.
- ZenML
- Anchor BrowserCloud hosted browser infrastructure
- Cloudflare
- LiveKit
- MastraOpen source TypeScript framework for agents and workflows
- Thread AI