Codentis is a high-performance CLI AI agent designed for terminal-based developers. It provides intelligent code assistance, file operations, and automated workflows through a modular agentic loop.
- Sub-Agent Delegation: Automatically offloads complex analysis, debugging, and refactoring to specialized sub-agents with curated goal prompts.
- Structured Tool Registry: Features collapsible outputs with persistent IDs, allowing developers to expand only the data they need to see.
- Workspace Security: Implements a proactive trust system requiring explicit user approval before performing operations in new directories.
- Provider Agnostic: Seamlessly switches between OpenAI, Anthropic, OpenRouter, and local OpenAI-compatible endpoints.
- Native Packaging: Ships as standalone installers (EXE, PKG, DEB) for professional deployment beyond standard pip environments.
- Codebase Onboarding: Build a comprehensive knowledge graph and dependency map of unfamiliar projects in a single pass.
- Automated Code Audits: Perform deep security and performance reviews using specialized reviewer agents before committing changes.
- Legacy Migration: systematically translate logic between frameworks or language versions with correctness verification.
- Local Development: Connect to local LLM servers for private, offline-first development workflows.
Download the latest installer for your platform from the Releases page.
git clone https://github.com/sujal-GITHUB/Codentis.git
cd Codentis
pip install -e .
codentisStart an interactive session:
codentisRun a single prompt:
codentis chat "Analyze the current project structure"Common TUI commands:
/e <id>: Expand specific tool output/list: Show all tool calls in session/exit: Quit the session
Manage settings with the config command:
codentis config --show
codentis config --resetThe application stores settings in ~/.codentis/config.json. Project-specific overrides can be defined in .agent/codentis.toml.
Codentis follows a modular design:
agent/: Core agent loop and session management.client/: LLM provider integrations (OpenAI, Anthropic, OpenRouter).tools/: Extensible tool registry and implementations.ui/: Terminal rendering and user interaction.
Run tests:
pytestBuild native installers:
python scripts/build_exe.py 64 # Windows
./scripts/build_macos.sh # macOS
./scripts/build_linux.sh # LinuxThis project is licensed under the MIT License.
Inspired by modern AI-powered developer tools like Claude Code.