agentty vs jcode

agentty vs jcode: two fast native agent harnesses (2026)

Short answer: Both are fast, native, MIT-licensed harnesses; they launch in a statistical tie. jcode is built for running many agents at once through a background server. agentty is a single process that idles at 21 MB instead of 106 MB, is 7.1× smaller on disk, and sandboxes every command by default.

jcode and agentty are the two native-code harnesses in our benchmark. jcode is written in Rust and built around a background server that hosts many agents at once. agentty is written in C++26 and runs as one self-contained process.

Measured

Both installed the official way on the same machine, fresh home directory, no login. jcode's memory includes the background server it starts on first launch and leaves running. Median of 10 launches, measured 2026-11-09 against jcode v0.99.1 (7206efd0c).

launch → ready to type lower is better
  1. agentty47 ms
  2. jcode56 ms1.2×
memory while idle lower is better
  1. agentty21 MB
  2. jcode106 MB5.0×
on disk, with runtime lower is better
  1. agentty23 MB
  2. jcode163 MB7.1×

median of 10 launches · same machine · fresh install, no login · method & all numbers

Launch times overlap across runs, so treat ready to type as a tie. Memory and disk are clear differences.

At a glance

agentty jcode
Built with C++26 Rust
Architecture One process TUI client + background server
Ready to type 47 ms 56 ms
Memory while idle 21 MB 106 MB
On disk 23 MB 163 MB
Sign in with a plan Claude Pro/Max, ChatGPT, GitHub Copilot, Kimi Claude, ChatGPT/Codex, Copilot and more
Shell sandbox On by default, enforced by the OS Permission prompts
Multi-agent Subagents; Smart Mode routes work to cheaper models Swarms of agents coordinated by the server
Editors Zed and JetBrains over ACP ACP
License MIT MIT

Where jcode is ahead

  • Many agents at once. Run several agents in one repo; the server tells each when another changes files it has read, and agents can spawn their own teammates.
  • Cheap extra sessions. Each extra session on the shared server costs only a few MB.
  • Extras. Dictation, a desktop companion, browser automation, and self-development tooling.

Where agentty is ahead

  • Lean single session. 5.0× less memory while idle, 7.1× smaller on disk, and nothing left running when you quit.
  • Sandboxed by default. Every shell command runs behind an OS-enforced wall; ~/.ssh, ~/.aws and .env are masked.
  • Retrieval. A local keyword + embedding search engine picks the code that reaches the model.
  • Everywhere. Linux and macOS on x86_64 and aarch64, Windows, Termux, SSH air-gap, scripts and CI.

Which should you choose?

  • jcode if you run swarms of agents in parallel and want them to coordinate.
  • agentty if you want one fast, light, sandboxed agent that you can drop on any machine.

Install in one line

bash
curl -fsSL https://agentty.org/install.sh | sh

Ready to try it? One line, no Node or Python.

curl -fsSL https://agentty.org/install.sh | sh
Get startedGitHub