chase-code

English | 简体中文

chase-code is a local CLI agent inspired by the Codex experience. It helps you solve coding tasks inside a repository by combining LLM reasoning with safe, built-in tools.

Highlights

• Agent REPL (Bubble Tea TUI): an interactive terminal UI that drives multi-step tasks with tools.
• Local tools: shell_command, apply_patch.
• Pluggable LLM providers: OpenAI and Kimi (Moonshot, OpenAI-compatible).

Requirements

• Go 1.21+ (align with the Go version used in this repo).

Build

From the repo root:

cd chase-code
# Build the binary (outputs ./chase-code)
go build -o chase-code .

Optionally add the binary to your PATH for convenience.

Configure LLM

LLM config lives in server/llm/config.go (LLMModel, NewLLMModelsFromEnv).

Common environment variables

• CHASE_CODE_LLM_PROVIDER:
  • openai (default)
  • kimi

OpenAI

Required:

• CHASE_CODE_OPENAI_API_KEY

Optional:

• CHASE_CODE_OPENAI_MODEL (default: gpt-4.1-mini)
• CHASE_CODE_OPENAI_BASE_URL (default: https://api.openai.com/v1)

Example:

export CHASE_CODE_LLM_PROVIDER=openai
export CHASE_CODE_OPENAI_API_KEY="sk-..."
export CHASE_CODE_OPENAI_MODEL="gpt-4.1-mini"
# Optional proxy/base URL:
# export CHASE_CODE_OPENAI_BASE_URL="https://your-proxy/v1"

Kimi (Moonshot)

Required:

• CHASE_CODE_KIMI_API_KEY or MOONSHOT_API_KEY

Optional:

• CHASE_CODE_KIMI_MODEL (default: kimi-k2-turbo-preview)
• CHASE_CODE_KIMI_BASE_URL (default: https://api.moonshot.cn/v1)

Example:

export CHASE_CODE_LLM_PROVIDER=kimi
export CHASE_CODE_KIMI_API_KEY="your-kimi-key"
# or
# export MOONSHOT_API_KEY="your-kimi-key"

export CHASE_CODE_KIMI_MODEL="kimi-k2-turbo-preview"
# Optional:
# export CHASE_CODE_KIMI_BASE_URL="https://api.moonshot.cn/v1"

Usage

1) Agent REPL (recommended)

Run from any repo you want to work on:

chase-code

2) Subcommands

chase-code                 # Enter the agent REPL
chase-code shell [options] -- <shell command>
chase-code read <file>
chase-code edit -file <file> -from <old> -to <new> [-all]
chase-code repl            # Explicitly enter the REPL

Examples:

# Run a shell command in the current directory
chase-code shell -- "ls -la"

# Read a file
chase-code read ./main.go

# Replace the first occurrence of "foo" with "bar"
chase-code edit -file main.go -from "foo" -to "bar"

# Replace all occurrences of "foo" with "bar"
chase-code edit -file main.go -from "foo" -to "bar" -all

Safety and sandboxing

• The shell tool and subcommand use sandbox policies from server/tools.
• Supported policies: full, readonly, workspace.
• Default policy is workspace-safe to avoid modifying paths outside the repo.

Recommended usage:

• Run chase-code inside a controlled project directory.
• Double-check high-risk commands (delete, network operations) before approval.

Project layout

• main.go: entry point, calls cli.Run().
• cli/: command parsing and REPL UI.
• server/: core logic (LLM client, tool routing, file edits, sandbox exec).

To extend tools:

• Implement capability in server/.
• Add CLI/REPL wiring in cli/.
• Keep main.go minimal.

License

Apache-2.0. See LICENSE.