# Kernex

> Rust runtime for production AI agents. OS-level sandboxing, provider-agnostic, memory-persistent.

Kernex is an open-source Rust runtime for building AI agents in production environments. It provides OS-level process sandboxing (macOS Seatbelt, Linux Landlock), deterministic memory usage with no GC heap, 12ms cold start, SQLite-backed persistent memory, and support for 11 AI providers including Claude, OpenAI, Gemini, and Ollama. It compiles to a single binary with no runtime dependencies.

## What Kernex is

- A Rust crate (`kernex-runtime`) for building LLM-powered agent pipelines
- OS-level sandboxing via macOS Seatbelt and Linux Landlock (syscall-level enforcement, not proxy-based)
- Provider-agnostic: Claude Code CLI, Anthropic, OpenAI, Gemini, Ollama, OpenRouter, Groq, Mistral, DeepSeek, Fireworks, xAI
- Deterministic memory: no GC heap, no interpreter startup, no JIT warmup
- Cold start: 12ms (vs 2,200ms+ for Python frameworks)
- Peak memory for 10 concurrent agents: 24 MB (vs 310 MB+ for Python frameworks)
- Single binary, no containers required

## kx CLI

kx is the terminal-native coding assistant built on Kernex. It gives your AI coding agent persistent memory across sessions, stored in a local SQLite database per project.

- Install: `cargo install kernex-agent`
- Per-project SQLite memory: facts, decisions, and architectural choices persist between sessions
- SKILL.md files load project-specific context at startup
- Reward-based learning: `/reward last` marks high-quality responses, improving future context
- Fully offline with a local Ollama model
- OS-sandboxed by default: filesystem access is OS-enforced, not just configured

## Key differentiators vs Python frameworks

| Property | Kernex (Rust) | Python frameworks |
|---|---|---|
| Cold start | 12ms | 2,200ms+ |
| Peak memory (10 agents) | 24 MB | 310 MB+ |
| Sandboxing | OS-level (Seatbelt/Landlock) | Container boundary |
| Memory safety | Compile-time | Runtime |
| Runtime dependencies | None (single binary) | Python, virtualenv, packages |

## When to use Kernex

Right choice: regulated environments (fintech, defense, healthcare), constrained hardware, deployments where OS-level sandboxing must be provably true, air-gapped development.

Not the right choice: one-off automation scripts, research prototypes, or cases where Python's ecosystem is a direct asset.

## Blog

- [Why we built an AI agent framework in Rust](https://kernex.dev/blog/why-rust-for-ai-agents): The full reasoning for choosing Rust over Python for production AI agents, including tradeoffs and what cuts against us.
- [Benchmark methodology: cold start, memory, and throughput](https://kernex.dev/blog/benchmark-methodology): Exact test environment, measurement methodology, and what the benchmarks do not cover.

## Case studies

- [Secure code review pipeline for a fintech team](https://kernex.dev/case-studies/fintech-code-review): OS-level sandboxing passed SOC 2 audit evidence requirements where Python frameworks were rejected. 3-phase pipeline, under 30 seconds per PR.
- [Local AI assistant for air-gapped development](https://kernex.dev/case-studies/airgapped-assistant): Full offline operation with Ollama in a classified development environment. Approved where every cloud-based tool was blocked.
- [Documentation bot with persistent memory](https://kernex.dev/case-studies/documentation-bot): Open source maintainer reduced support response time from 15 minutes to 4 minutes using kx memory across sessions.
- [ETL triage agent](https://kernex.dev/case-studies/etl-triage): Automated triage of data pipeline failures with sandboxed tool execution.
- [Onboarding knowledge bot](https://kernex.dev/case-studies/onboarding-knowledge-bot): Per-project memory stores architectural decisions that new engineers can query directly.

## Links

- Website: https://kernex.dev
- GitHub: https://github.com/kernex-dev/kernex
- kx CLI: https://kernex.dev/kx
- crates.io (kernex-agent): https://crates.io/crates/kernex-agent
- crates.io (kernex-runtime): https://crates.io/crates/kernex-runtime
- API docs: https://docs.rs/kernex-runtime
- Blog: https://kernex.dev/blog
- Case studies: https://kernex.dev/case-studies
- About: https://kernex.dev/about