Introduction

Kraken AI is the complete agent harness platform — the tooling, data, and decision-making controls autonomous agents need to operate on real business systems, with human oversight where it matters.

#What is Kraken AI

Kraken AI is a unified control plane for deploying, monitoring, securing, and governing autonomous AI agents. It is the production infrastructure any organization uses to build and operate domain-specific agents that take real action on real systems.

Agents are powerful when they can perceive data and act on tools. They become safe to run in production when every action they take passes through policy, every decision is auditable, and a human can be brought into the loop at exactly the right moment. Kraken AI provides that harness: the layer agents plug into so they can be useful without being unaccountable.

#The harness model

Everything in Kraken AI follows one mental model. An agent perceives through connectors, learns procedures through skills, carries context through memory, and takes outbound action through actions. The platform wraps that pipeline with governance and observability so the agent stays accountable end to end.

• Ingest — data connectors feed agents from databases, APIs, SaaS tools, and internal systems.
• Extend — Model Context Protocol (MCP) connectors give agents native, governed tool access.
• Equip — skills are composable, versioned procedures an agent can load and execute.
• Remember — managed memories persist context across runs with policy-based access.
• Act — actions are structured, typed outputs that feed directly into your tools and workflows.

#The five layers

The control plane is organized into five layers. Each is documented in depth in the sections that follow.

• Deploy — Build and run production agents: multi-agent teams, isolated sandboxed runtimes, and long-running execution.
• Protect — Runtime enforcement at every layer: human approval gates, automatic kill switches, and input and output validation.
• Observe — Step-level execution traces, real-time anomaly detection, structured metrics, and cost tracking.
• Connect — Plug agents into any data source, tool, or endpoint with open-source connectors and MCP.
• Govern — Just-in-time authorization, a runtime policy engine, role-based access control, and immutable audit trails.
• Compliance — Built for regulated environments: SOC 2 Type II, ISO 27001, HIPAA, and GDPR.

#How it fits together

You define agents, connectors, skills, and actions in a single typed project using the open SDKs. Pushing the project to a connected repository builds and deploys it to the platform. From then on, triggers start runs: each run is an isolated execution where the agent reasons, calls tools, and produces output.

Every tool call and every model output in a run passes through the governance gateway. Policies decide whether an action is allowed, denied, or escalated to a human. Every decision is written to an immutable audit trail, and every run is streamed live into the control plane for observability.

#Where to go next

• Quickstart — Scaffold a typed multi-agent project and run your first agent locally in minutes.
• Core Concepts — The vocabulary of the platform: agents, runs, connectors, skills, actions, policies, and audit.