Tactical Automated Response System — AI-powered CLI agent that reads, edits, and executes across your codebase via natural language.
npm install -g tars-ai-cliPoint TARS at any project directory and it operates autonomously — analyzing structure, planning changes, executing them, and verifying results.
$ tars
✓ Started in AGENT mode
❯ Add error handling to the readFile function
- Autonomous loop — Plans, executes, and verifies changes end-to-end
- Full file control — Read, edit, and create files with codebase context
- Code search — Glob pattern and regex search across the project
- Shell integration — Run any command through the assistant
- Three modes —
agent(execute),ask(read-only Q&A),planning(plan without changes) - File pinning — Inject specific files into every prompt
- Guardrails — Blocks
.env, private keys, and sensitive files by default - Token management — Sliding context window across long sessions
| Language | TypeScript 5 |
| Runtime | Node.js v18+ / Bun (dev) |
| AI | Google Gemini 2.5 Flash via Vercel AI SDK |
| CLI | Commander.js, Chalk, cfonts |
| Validation | Zod v4 |
| Build | Bun → CJS |
Prerequisites: Node.js v18+, Google AI API key
npm install -g tars-ai-cliCreate a .env in your project root:
GOOGLE_GENERATIVE_AI_API_KEY=your_api_key_hereRun in any project directory:
tars| Mode | Behavior |
|---|---|
agent |
Full execution — reads, edits, creates, runs commands |
ask |
Read-only — answers questions and explains code |
planning |
Analyzes and plans without modifying any files |
| Command | Description |
|---|---|
:mode [agent|ask|planning] |
Switch mode |
:pin <file> |
Pin file into every prompt |
:unpin <file> |
Unpin file |
:clear |
Clear the screen |
:help |
List all commands |
:exit |
Quit |
Optional .tars.json in your project root:
{
"llm": {
"model": "gemini-2.5-flash",
"temperature": 0.2
},
"features": {
"maxContextTokens": 20000,
"confirmEdits": false
},
"guardrails": {
"blockReadPatterns": [".env", ".env.*", "*.pem", "*.key", ".npmrc"]
}
}MIT — see LICENSE.
