The Autonomous, Local-First AI Terminal Assistant.
Unlike passive AI chat widgets, KURO has hands. It plans, executes shell commands, inspects live telemetry, validates outcomes with deterministic AST compilation, heals its own errors, and evolutes directly on your workstation.
Engineered for Developer Autonomy
Features verified against the KuroCodeAgent 2.0 master specification and implementation.
Autonomous ReAct Loop
Decomposes high-level instructions into multi-step actions. Never assumes success—validates each step with AST checks (py_compile) and exit code scrutiny.
Self-Healing Engine
Failure classifier detects errors across 10 structured categories. Triggers automated pip install, path creation, or git-stash rollback checkpoints.
Windows DPAPI Vault
Hardware-backed encryption for API keys tied to your Windows OS user profile (CryptProtectData) with AES-256 fallback on Linux/macOS.
Modular Puzzle Store
Synchronize, install, and hot-load tools, plugins, procedural skills, and external bridges from KuroHub without restarting the running agent session.
Desktop Vision & Telemetry
Inspects screen grabs to diagnose visual layout bugs, runs Playwright browser automation, and reads live CPU, RAM, and disk utilization statistics.
External Bridges
Connect terminal notifications to Discord webhooks, Telegram bots, GitHub PRs/issues, GitLab pipelines, or Google Drive file storage.
Deterministic ReAct Architecture
How KuroCodeAgent processes user instructions from terminal input to verified disk modification.
Terminal CLI input, slash command, or headless single-shot string.
Context engine synthesizes prompt; ModelRouter selects active LLM provider.
Safe Mode verifies AST security; blocks destructive calls unless approved.
Invokes local tools or modular puzzle pieces loaded from Kuro Hub.
Validates AST & exit codes; triggers auto-pip or rollback on failure.
Ready to run KuroCode on your machine?
Install the verified release directly from PyPI or clone the source code for local contribution.