Jcode
The harness company
About Jcode
An agent is LLM + harness: We work purely on that second component. We do research and engineering to produce the most capable and efficient harness. Our harness has 20k stars on , and has $1m/month of inference spend
Public traction evidence
Each signal links to the public source used for attribution.
- X
I run 20 coding agents in parallel as my everyday workflow.
I run 20 coding agents in parallel as my everyday workflow. Today, I’m launching Jcode. It’s an open-source agent 20x more memory-efficient than Claude Code, so you can run 20x more agents at once. Try it today: https://t.co/Vwg0NhNA3h
- X
Launching today! http://JCode.sh The number of coding agents you can run at once is capped by the RAM needed. Jcode cuts it 20x. It's op…
Launching today! http://JCode.sh The number of coding agents you can run at once is capped by the RAM needed. Jcode cuts it 20x. It's open source, spawns 63x faster than Codex CLI, 60% on Terminal Bench 2 vs Claude Code's 48%. Try it: http://jcode.sh
- X
If pg thinks you are gmi, you are gmi.
If pg thinks you are gmi, you are gmi.
- X
Today I'm fixing benchmarks with Jcode bench! Jcode bench is the first open and uncontaminatable benchmark.
Today I'm fixing benchmarks with Jcode bench! Jcode bench is the first open and uncontaminatable benchmark. There are a few problems in the coding benchmarks of today: >Private benchmarks are hard to trust >Public benchmarks are easy to benchmaxx >Benchmark task grading can be
- X
people tell me spawn time doesn't matter, but codex and claude code being so slow is the only reason i had it in the readme.
people tell me spawn time doesn't matter, but codex and claude code being so slow is the only reason i had it in the readme. (this is 1x speed)
- X
I was talking to a fellow YC founder about this yesterday (I work on harnesses).
3 is correct. I was talking to a fellow YC founder about this yesterday (I work on harnesses). The real power in a harnesses is that it allows you to tune up and down test time compute in certain directions. For example, the todo tool requires the model to assign scores from
- X
there is a secret sauce to being better at managing coding agents.
there is a secret sauce to being better at managing coding agents. This is a real time demo of niri
- X
You don’t need a cheap llm to generate session titles.
You don’t need a cheap llm to generate session titles. You can have a name parameter for the todo tool, which means it’s generated on the fly by the main model. Then you can deterministically extract it and use as session title, and if there wasn’t a significant task done you