🤖 bse-code

The AI coding assistant that lives in your terminal — works with ANY LLM, zero lock-in, zero compromise.

Chat with your codebase, read and write files, run shell commands, connect MCP servers, build reusable skills, persist project memory, and pick up right where you left off — all from a gorgeous interactive REPL.

✨ Why bse-code?

🚀 Install

Requires Node.js 18+. No .NET SDK needed — the binary is bundled.

npm install -g bse-code

Also available as a .NET global tool: dotnet tool install --global bse-code

🌐 Supported Providers

🧙 First-run Setup

On first run, an interactive wizard walks you through everything:

1. 🎯 Pick a provider
2. 🔗 Set the base URL (pre-filled for known providers)
3. 🔑 Enter your API key (skipped for local providers)
4. 🤖 Browse available models and pick one
5. 💾 Everything saved to ~/.bse-code/config.json

Re-run the wizard any time:

bse-code --config

⚡ Quick-start by provider

🔀 OpenRouter — free models, no credit card

bse-code --config
# Select [1] OpenRouter → get a free key at https://openrouter.ai/keys
# Pick Gemini 2.5 Pro, Llama 4, DeepSeek R1 — all free!

🦙 Ollama — fully local, zero cost

ollama pull llama3.2
bse-code --config
# Select [5] Ollama → accept default URL → pick your model

💻 Usage

🔁 Interactive REPL (recommended)

bse-code

  ╭──────────────────────────────────────────╮
  │   ██████╗ ███████╗███████╗                │
  │   ██╔══██╗██╔════╝██╔════╝                │
  │   ██████╔╝███████╗█████╗   ─ code         │
  │   ██╔══██╗╚════██║██╔══╝                  │
  │   ██████╔╝███████║███████╗                │
  │   ╚═════╝ ╚══════╝╚══════╝                │
  ╰──────────────────────────────────────────╯
  provider: OpenRouter
  model   : google/gemini-2.5-pro-exp-03-25:free
  theme   : default
  cwd     : my-project
  🧠 skills : 2 loaded
  🔌 mcp    : 5 tools from 1 server(s)
  💾 memory : 1 BSE.md file(s) loaded
  type /help for commands · /exit to quit 🚀

 my-project (main) ❯

⚡ One-shot mode

bse-code -p "explain the auth flow in src/auth/"
bse-code -p "list all TODO comments" --output-format json

🏳️ All CLI flags

bse-code                              # 🔁 Interactive REPL
bse-code -p "<prompt>"                # ⚡ One-shot prompt
bse-code --model <model-id>           # 🤖 Override model for this session
bse-code --theme <name>               # 🎨 Set color theme for this session
bse-code --output-format json|text    # 📄 Output format (one-shot only)
bse-code --config                     # ⚙️  Re-run the setup wizard
bse-code --version, -v                # 🔢 Show version
bse-code --help, -h                   # ❓ Show help

🪄 Special Input Prefixes

@ — File & directory injection

Drop any file or folder straight into your prompt. Tab-completes paths as you type.

@src/auth.ts explain this file
@src/auth/ summarize all files in this folder
@package.json what dependencies are outdated?

Directories inject up to 20 files automatically — perfect for asking about a whole module at once.

! — Shell passthrough

Run any shell command instantly, no AI involved, output right in your terminal.

!git status
!npm run build
!ls -la src/

⌨️ REPL Slash Commands

🔧 Core

🎨 Appearance

🧠 Skills

🔌 MCP

💾 Memory

📁 Sessions

🎮 Interactive Input — Feels Like a Real Shell

/ — Slash command picker

Type / and an inline menu pops up instantly:

  /  ↑↓ navigate · Enter select · Esc cancel
  ▶ /clear                🧹 clear conversation history
    /model                🤖 show or switch model
    /compact              🗜️  summarize history to save tokens
    /theme                🎨 list or set color theme
    /skills               🧠 list loaded skills
    /mcp                  🔌 list MCP servers and tools
    /memory               💾 show loaded BSE.md files
    /save                 💾 save conversation
    /resume               ▶️  list or resume a saved session
    …

• ⬆️⬇️ Arrow keys navigate the list
• ⌨️ Type more characters to filter live — /th narrows to /theme
• ↩️ Enter selects, Esc cancels
• ⇥ Tab completes the top match
• 🧠 Your skills appear right alongside built-in commands

📜 History & cursor editing

• ⬆️⬇️ arrows cycle through previous inputs — just like your shell
• ⬅️➡️ move the cursor anywhere in the line
• Home / End jump to start or end
• Backspace / Delete work at any cursor position
• ⇥ Tab on @<path> completes file and directory paths

🧠 Skills — Reusable AI Workflows

Skills are markdown files that give the AI reusable instructions or workflows. Write once, invoke from any project.

📂 Locations (both loaded and merged):

• ~/.bse-code/skills/ — user-level, available in every project
• .bse-code/skills/ — project-level, scoped to this repo

Example skill (.bse-code/skills/review.md):

# Code Review

Review the provided code for:
- Correctness and logic errors
- Performance issues
- Security vulnerabilities
- Code style and readability

Invoke it:

/review
/review @src/PaymentService.ts

Skills are also injected into the system prompt automatically — the AI always knows what's available. 🚀

💾 Project Memory (BSE.md)

BSE.md files are loaded at startup and injected into every session's system prompt. Teach the AI about your project once — it remembers forever.

🏗️ Hierarchy — all three are merged:

Scaffold one instantly:

bse-code
/init

Add notes on the fly:

/memory add always use async/await, never .then() chains
/memory add run `npm test` before committing

🔌 MCP (Model Context Protocol)

Connect any external tool or service via MCP servers. GitHub, databases, Slack, custom APIs — if it speaks MCP, it works here.

Config file: ~/.bse-code/mcp.json

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/dir"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "your-token" }
    }
  }
}

• Tools available to the AI as mcp__serverName__toolName
• Hot-reload without restarting: /mcp reload 🔄
• Inspect what's connected: /mcp
• Disable without removing: "disabled": true

🔨 Built-in AI Tools

Tool calls are shown inline as the AI works — you see exactly what it's doing in real time. ✓ or ✗ per call.

💾 Session Management

Never lose a good conversation. Save any session with a tag and resume it later.

/save auth-refactor

/resume
#   auth-refactor   2025-04-24 14:32   18 messages   [gpt-4o]
#   bug-hunt        2025-04-23 09:15   31 messages   [claude-3-5-sonnet]

/resume auth-refactor
# ▶️  Resumed session 'auth-refactor' (18 messages) — welcome back!

Sessions are stored per-project in ~/.bse-code/sessions/ — isolated, no collisions.

📊 Session Statistics

/stats

  Session stats 📊
    ⏱  duration   : 00:23:41
    💬 turns      : 12
    🔧 tool calls : 34
    📨 messages   : 47
    🤖 model      : google/gemini-2.5-pro-exp-03-25:free
    🌐 provider   : OpenRouter
    🎨 theme      : dracula
    🧠 skills     : 3
    🔌 mcp tools  : 8

🗜️ Conversation Compaction

Running low on context? Compact the conversation into a tight summary without losing the important bits.

/compact
/compact focus on the auth changes we made

The AI summarizes, history is trimmed, you keep going — same context, way fewer tokens. 🎯

🎨 Themes

bse-code --theme dracula    # one session
# or inside the REPL:
/theme monokai              # persisted

⚙️ Configuration

Config file: ~/.bse-code/config.json

{
  "provider": "OpenRouter",
  "api_key": "sk-or-...",
  "model": "google/gemini-2.5-pro-exp-03-25:free",
  "base_url": "https://openrouter.ai/api/v1",
  "theme": "default"
}

🌍 Environment variables (always override config)

🔄 Legacy OPENROUTER_API_KEY, OPENROUTER_MODEL, OPENROUTER_BASE_URL still work as fallbacks.

📁 File Structure

~/.bse-code/
├── config.json          # ⚙️  Provider, API key, model, base URL, theme
├── mcp.json             # 🔌 MCP server definitions
├── BSE.md               # 🌍 Global memory
├── skills/
│   └── *.md             # 🧠 User-level skills
└── sessions/
    └── <project-hash>/  # 💾 Saved conversations per project

.bse-code/               # Project-level (commit to your repo)
├── BSE.md               # 📁 Project memory
└── skills/
    └── *.md             # 🧠 Project-level skills

./BSE.md                 # 📁 Project memory (root level)
./BSE.local.md           # 🔒 Local overrides — gitignore this

📄 License

MIT