Skip to main content

Cortex Overview

Cortex is a high-performance runtime for running autonomous agents. It handles process isolation, resource enforcement, capability permissions, and lifecycle management — so your agent code doesn't have to.

Built with a Rust-native core (v2.5.10), Cortex is designed to handle thousands of concurrent agents with sub-millisecond orchestration overhead and cold starts under 5 seconds.

Key Features (v2.5.10)

  • ⚡ Turbo Mode: A high-concurrency task scheduler that enables 3-5x parallel speedup for agent sequences and tool execution.
  • 🧠 Shared KV Cache: Dramatically reduces memory footprint for multi-agent swarms by sharing common context tokens across agent instances.
  • 🛡️ Secure Sandboxing: Hard CPU/RAM ceilings and capability-based permissions ensure agents remain isolated and safe.
  • 📦 Portable Bundles: .cortex archives package logic, prompts, and dependencies for identical execution across dev, staging, and production.
  • 📊 Real-time Telemetry: Native support for p99 latency tracking and structured JSON logging.

What it does

Agents defined in a cortex.yaml manifest run inside isolated sandboxes. Each agent gets:

  • Resource Limits: A hard CPU and RAM ceiling — it cannot starve other agents or the host system.
  • Capability Isolation: Network access, filesystem access, and tool use must be explicitly granted in the manifest.
  • Lifecycle Control: Start, stop, suspend, resume, and health check without writing orchestration code.
  • Global KV Store: Access to a high-speed, shared key-value cache for inter-agent state sharing.

Architecture

Your Application (SDK/CLI)


┌───────────────────────────────────────────┐
│ Cortex Runtime │
│ (Rust Engine + Tokio Scheduler) │
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │ Agent A │ │ Agent B │ │ Agent C │ │
│ └──────────┘ └──────────┘ └──────────┘ │
│ │ │ │ │
│ └──────┬─────┴─────────────┘ │
│ ▼ │
│ Shared KV Cache (Turbo) │
└───────────────────────────────────────────┘


External APIs & Infrastructure

Cortex + ICE

Cortex and ICE are independent — you can use either without the other. But they compose directly:

  • Cortex manages agent execution — isolation, resources, lifecycle.
  • ICE manages agent memory — session context, retrieval, multi-tenant history.

When used together, each Cortex agent leverages ICE for infinite memory horizons while Cortex ensures the agent's execution is efficient, parallelized, and secure.

Source

Cortex is open source. The repository, issue tracker, and contribution guide are at github.com/dopove/cortex.