patchllm 1.0.0

An interactive agent for codebase modification using LLMs

About

PatchLLM is an interactive command-line agent that helps you modify your codebase. It uses an LLM to plan and execute changes, allowing you to review and approve every step.

Key Features

• Interactive Planning: The agent proposes a step-by-step plan before writing any code. You stay in control.
• Dynamic Context: Build and modify the code context on-the-fly using powerful scope definitions (@git:staged, @dir:src, etc.).
• Mobile-First TUI: A clean, command-driven interface with autocompletion makes it easy to use on any device.
• Resilient Sessions: Automatically saves your progress so you can resume if you get disconnected.

Getting Started

1. Initialize a configuration file (optional): This creates a scopes.py file to define reusable file collections.

patchllm --init

2. Start the Agent: Running patchllm with no arguments drops you into the interactive agentic TUI.

patchllm

3. Follow the Agent Workflow: Inside the TUI, you direct the agent with simple slash commands.

# 1. Set the goal
>>> /task Add a health check endpoint to the API

# 2. Build the context
>>> /context @dir:src/api

# 3. Ask the agent to generate a plan
>>> /plan
1. Add a new route `/health` to `src/api/routes.py`.
2. Implement the health check logic to return a 200 OK status.

# 4. Execute the first step and review the proposed changes
>>> /run

# 5. If the changes look good, approve them
>>> /approve

Agent Commands (TUI)

Command	Description
/task <goal>	Sets the high-level goal for the agent.
/plan [management]	Generates a plan, or opens an interactive TUI to edit/add/remove steps.
/run [all]	Executes the next step, or all remaining steps with /run all.
/approve	Interactively select and apply changes from the last run.
/diff [all | file]	Shows the full diff for the proposed changes.
/retry <feedback>	Retries the last step with new feedback.
/skip	Skips the current step and moves to the next.
/revert	Reverts the changes from the last /approve.
/context <scope>	Replaces the context with files from a scope.
/scopes	Opens an interactive menu to manage your saved scopes.
/ask <question>	Ask a question about the plan or code context.
/refine <feedback>	Refine the plan based on new feedback or ideas.
/show [state]	Shows the current state (goal, plan, context, history, step).
/settings	Configure the model and API keys.
/help	Shows the detailed help message.
/exit	Exits the agent session.

Headless Mode Flags

For scripting or single-shot edits, you can still use the original flags.

Flag	Alias	Description
-p, --patch	Main action: Query the LLM and apply file changes.
-t, --task	Provide a specific instruction to the LLM.
-s, --scope	Use a static scope from scopes.py or a dynamic one.
-r, --recipe	Use a predefined task from recipes.py.
-in, --interactive	Interactively build the context by selecting files.
-i, --init	Create a new scopes.py file.
-sl, --list-scopes	List all available scopes.
-ff, --from-file	Apply patches from a local file.
-fc, --from-clipboard	Apply patches from the system clipboard.
-m, --model	Specify a different model (default: gemini/gemini-1.5-flash).
-v, --voice	Enable voice interaction (requires voice dependencies).

Setup

PatchLLM uses LiteLLM. Set up your API keys (e.g., OPENAI_API_KEY, GEMINI_API_KEY) in a .env file.

The interactive TUI requires prompt_toolkit and InquirerPy. You can install all core dependencies with:```bash pip install -r requirements.txt

Optional features require extra dependencies:
```bash
# For URL support in scopes
pip install "patchllm[url]"

# For voice commands (in headless mode)
pip install "patchllm[voice]"

License

This project is licensed under the MIT License.