Fix a failing test
Run the suite, read the failure, repair the code, and prove it passes.
Read the guide →iCode Agent · Terminal
The coding agent that won’t burn your context window.
iCode Agent works directly in your repository: it reads the code, makes changes, runs your tests, and verifies its own work — while you watch and steer it in your terminal. Long tasks are the hard case, and context is a budget it defends.
npm install -g @icodehq/icode-agent cd ~/my-project icode-agent
Node.js 22+ · Apache-2.0 · One package for Windows, macOS, and Linux
An illustrative session showing the explore → change → verify loop. Walk through your own first task →
Context that lasts
Running a long task is where coding agents fall apart: the window fills with old file reads and command output, the agent starts forgetting what it did, and every request costs more. iCode Agent treats context as a budget to defend, not a bucket to fill.
- Cap at the source — oversized output is capped as it is produced and spilled to a file with a short preview.
- Reduce at request time — older results become compact markers only when a request would exceed its budget, oldest first.
- Summarize durably — when history truly overflows, older turns are summarized with a recent tail preserved.
A withheld result keeps a route back to the full output, failure evidence is never hidden, and reduction never rewrites your saved session.
What you can do
Refactor a module
Map the callers, then make a focused refactor with reviewed diffs.
Read the guide →Automate in CI
Run headless with -p, stream events with --mode json.
Bring your own model
iCode Agent speaks the openai and anthropic
protocols, so it works with the official APIs and any compatible endpoint.
The agent runs locally; your code goes only to the model endpoint you
configure.
Free and open source (Apache-2.0). You pay only your model provider.
How it fits with iCode Workbench.
Workbench (in early access) is the desktop environment for a team of agents — Claude Code, Codex, and OpenCode in isolated worktrees, with tasks, review, and Mira. Agent is the focused terminal surface: one repository, one session, no window. They are separate products with the same discipline, and either one can be used on its own.
iCode Agent · Keep the thread, keep the cost bounded.