thecode-agent 0.1.1

A secure, provider-independent coding agent for the terminal

TheCode

AI Software Engineer for your terminal.

TheCode is a provider-independent coding agent with a full-screen terminal workbench. It can understand a repository, search and read code, propose plans, edit files with approval, run commands and tests, inspect Git changes, and preserve local sessions.

TheCode can modify files and execute approved commands. Use version control, review permission prompts, and never share API keys.

Highlights

• Premium Textual and Rich terminal interface
• Streaming assistant responses and live tool activity
• Preview, Diff, Files, Architecture, Terminal, and Logs panels
• Tool inspector with arguments, output, status, and duration
• Permission modes for safe, interactive, or more autonomous work
• OpenRouter, OpenAI, Anthropic, and local Ollama support
• Git-aware repository analysis and session history
• Multi-folder .code-workspace support
• @file context, project rules, skills, and AGENTS.md support
• Sensitive-file protection for .env, private keys, and credentials
• Windows, Linux, and macOS support

Requirements

• Python 3.12 or newer
• Git
• Ripgrep is recommended for fast search
• An API key for OpenRouter, OpenAI, or Anthropic; Ollama can run locally without one

Install

The recommended installation uses pipx, which keeps CLI applications isolated:

pipx install thecode-agent

If pipx is not installed:

python -m pip install --user pipx
python -m pipx ensurepath

Alternatively, install with pip:

python -m pip install thecode-agent

Upgrade later with:

pipx upgrade thecode-agent
# or: python -m pip install --upgrade thecode-agent

First run

Open a terminal in the project you want TheCode to work on:

cd path/to/your/project
thecode config --setup
thecode doctor
thecode

The setup wizard asks for a provider, model, API key, and permission mode. Credentials are entered with hidden input and stored in the project .env; that file should remain ignored by Git.

Inside the workbench, type a request:

Create a small REST API for managing fruits and add tests.

Type / to discover interactive commands. Use @path/to/file.py to attach a specific file to the request.

Providers

Provider	Credential	Example model
OpenRouter	OPENROUTER_API_KEY	anthropic/claude-sonnet-4
OpenAI	OPENAI_API_KEY	Provider model identifier
Anthropic	ANTHROPIC_API_KEY	Provider model identifier
Ollama	No key required	Any locally installed model

Run the setup wizard again at any time:

thecode config --setup

Never commit .env or paste credentials into chat, issues, logs, or documentation. Revoke any key that has been exposed.

Workbench controls

Key	Action
Enter	Send the request
Ctrl+P	Quick Open a file
Ctrl+K	Open the command center
Ctrl+T	Ask the agent to run relevant tests
Ctrl+R	Run the previous request again
Ctrl+L	Clear the conversation view
Esc	Cancel the current operation
Ctrl+Q	Exit TheCode

When typing /, suggestions are filtered as you type. Press Down and Enter, click an option, or press Tab to accept the first suggestion.

Useful interactive commands:

/commands
/config
/models
/architect OBJECTIVE
/theme NAME
/workspace NAME
/processes
/session
/clear
/exit

/architect performs a read-only Current/Proposed architecture analysis. TheCode only starts implementation after you select Apply architecture.

CLI commands

The full-screen workbench is the default, but commands can also be run directly:

thecode
thecode chat
thecode chat --classic
thecode ask "How does authentication work?"
thecode plan "Migrate JWT authentication to OAuth2"
thecode run "Fix the failing tests"
thecode review
thecode init
thecode config
thecode models
thecode sessions
thecode resume SESSION_ID
thecode undo SESSION_ID
thecode delete-session SESSION_ID
thecode theme

Permission modes

• safe: reads are automatic; writes and commands require confirmation.
• ask: reads are automatic; mutating actions request approval.
• auto: lower-risk operations can run automatically; high-risk actions still ask and critical actions remain blocked.

Permission prompts support Allow once, Allow for this session, and Deny. Destructive and critical operations remain restricted.

Project instructions and context

Initialize optional project configuration and an instruction file:

thecode init

This creates .nexus/config.yaml and NEXUS.md without overwriting existing files. TheCode also understands common agent instructions including AGENTS.md, AGENT.md, CLAUDE.md, .cursorrules, Cursor rules, and SKILL.md files.

Configuration is loaded from ~/.nexus/config.yaml, then overridden by the current project's .nexus/config.yaml.

Example:

llm:
  provider: openrouter
  model: anthropic/claude-sonnet-4
agent:
  max_steps: 30
permissions:
  mode: ask
context:
  max_characters: 120000

Workspaces

TheCode detects a .code-workspace file and supports multiple project folders:

thecode workspace open platform.code-workspace
thecode workspace list platform.code-workspace
thecode workspace add ../backend --path platform.code-workspace

Inside the workbench, use /workspace NAME to switch the active folder.

Sessions and recovery

Sessions are stored locally in ~/.nexus/sessions.db.

thecode sessions
thecode resume SESSION_ID
thecode undo SESSION_ID
thecode delete-session SESSION_ID

undo refuses to overwrite a file that was subsequently changed by the user.

Troubleshooting

Run diagnostics first:

thecode doctor

If the thecode command is not found, restart the terminal after installing with pipx, or run:

python -m pipx ensurepath

If an older editable installation takes precedence, uninstall the previous package and reinstall:

python -m pip uninstall thecode-agent -y
python -m pip install --upgrade thecode-agent

For a clean onboarding reset, close TheCode first and remove its local state. This does not remove your source code:

# PowerShell
Remove-Item -Recurse -Force .nexus -ErrorAction SilentlyContinue
Remove-Item -Recurse -Force "$HOME\.nexus" -ErrorAction SilentlyContinue
thecode config --setup

API keys stored in .env or operating-system environment variables are not deleted by this reset.

Security

Paths are resolved against the active project, including symlinks. Sensitive files are hidden from automatic previews, Quick Open, project trees, and @file attachments. Commands use argument arrays rather than shell=True; unsupported shell operators and destructive patterns are blocked. File changes preserve recovery metadata.

Security reports should not contain live credentials, private keys, or proprietary source code.

Development

Clone the repository and install the development dependencies:

python -m venv .venv
source .venv/bin/activate  # Windows PowerShell: .venv\Scripts\Activate.ps1
python -m pip install -e ".[dev]"

Run the quality checks:

ruff check nexus tests
mypy nexus
pytest

HTTP provider tests use mocked transports and do not require live API access.

License

TheCode is available under the MIT License.