claude-jr

A coding agent for the terminal, in a 2 MB binary.

claude-jr, or Claude Junior, is a small alternative to Claude Code: it reads your code, edits files and runs commands, asking before anything changes. It is written in Rust with no runtime, so it starts in 5 ms and sits in 0.7 MB of memory: 160 times less disk and 190 times less memory than Claude Code. It runs alone, or as the workers Claude Code, Cursor or Codex hand tasks to: a thousand of them fit in 500 MB.

curl -fsSL https://claudejr.iwhitebird.com/install | sh

Linux and macOS, on x86 and ARM; on Windows, inside WSL. Then run claudejr and paste an Anthropic API key.

Small enough not to notice

There is no Node or Python to install and nothing running in the background. Memory stays under 4 MB however long a session runs, because the conversation is kept on disk. A whole working session, including the commands it runs, fits in a 16 MB container on a twentieth of a CPU core.

2 MBone static binary
0.7 MBmemory when idle
5 msto start
0 msCPU while it waits
claude-jrClaude Codepi
Install size2.0 MB324.5 MB109.0 MB
Memory, idle0.7 MB135.0 MB141.1 MB
Memory, working2.0 MB270.2 MB137.5 MB
Startup5 ms167 ms~260 ms
Its own CPU, 20 tool calls30 ms1,150 ms580 ms

All three run against the same fake API, so the model's time isn't counted; how it's measured.

And it gets the work done

SWE-bench Verified gives an agent real GitHub issues from projects like Django, sympy and scikit-learn, and checks its fix with the projects' own hidden tests.

88%

of a random sample of 100 tasks resolved with Claude Sonnet 5.5, one attempt each, at about five cents a task. With Claude Opus 5.5, 45 of 50 (90%).

What's in it

Workers for other agents
claudejr install claude-code (or cursor, or codex) lets your main agent start claude-jr workers over MCP, then read, merge or drop what each one did.
Tasks and fleets
Runs carry on in the background, each in its own git worktree and branch, with a shared budget and a log of every step. claudejr fleet runs a whole list.
Any model
Claude by default; OpenAI, Gemini, Groq, OpenRouter, Ollama and other local servers with -m provider/model.
Subagents
Side tasks go to agents that work in parallel and report back, in worktrees of their own when they change files.
Permission modes
From asking about everything to an auto mode where a small model reviews each command. shift tab switches.
A sandbox
With --sandbox on Linux it can only write to your project; keys, sockets and the rest of your home stay out of reach.
Rewind and resume
Go back to before any message with its file changes undone. Every session is saved as it goes.
Skills and MCP
Reads your .claude/skills and talks to MCP servers, local or remote.
Images and the web
Drag in a screenshot, and it can search and fetch pages when the task needs them.
A stash
ctrl s puts a half-written message aside while you send another.
Scripting
claudejr -p "…" runs one prompt, with JSON output and a spending limit for CI.

Using it

claudejrstart in this folder
claudejr -ccontinue the last conversation
claudejr -p "explain main.rs"one prompt, then exit
claudejr --sandboxonly write inside this folder
claudejr -m ollama/qwen3use another provider's model
claudejr task start -y "…"run a task in the background, on its own branch
claudejr fleet -y "…" "…"run a list of tasks side by side
claudejr install claude-codelet Claude Code hand tasks to workers
claudejr --updateget the newest release
/every command, inside it