Atessa

Autonomous CLI coding agent.

Atessa is a terminal-native coding assistant that analyzes your project, writes code, runs commands, and completes complex development tasks — autonomously. It operates through a modular skill system with 25+ tools, semantic code search, session persistence, and sub-agent parallelism.

Installation

npm install -g atessa

Requires Node.js 18+ and an Anthropic API key.

Quick start

# Set your API key
export ANTHROPIC_API_KEY=sk-ant-...

# Interactive mode
cd my-project
atessa

# One-shot mode
atessa "add user authentication with JWT"

Features

• Autonomous agent loop — executes multi-step tasks end-to-end
• Extended thinking — automatic complexity detection triggers deeper analysis
• Modular skills — load/unload capability packs on demand (software-engineer, researcher, security-researcher)
• 25+ built-in tools — file ops, code search, git, commands, sub-agents, and more
• Semantic search — vector-indexed codebase search via LanceDB
• Checkpoint system — automatic undo/rollback for safe experimentation
• Session persistence — resume previous conversations with atessa -c
• Sub-agents — spawn parallel workers for independent tasks
• MCP support — extend capabilities with Model Context Protocol servers
• Persistent memory — remembers preferences and project context across sessions
• Project instructions — respects CLAUDE.md / AGENTS.md files

Usage

Interactive mode

atessa

The agent starts a REPL. Type your request, watch it plan and execute, then continue the conversation.

One-shot mode

atessa "fix the failing tests in src/utils"

Resume a session

atessa -c              # resume latest session
atessa -c abc123       # resume a specific session

Common options

-m, --model <model>         Model to use (default: claude-sonnet-4-5-20250929)
--max-iterations <n>        Max agent iterations (default: 50)
-y, --yolo                  Auto-approve all tool executions
--auto-approve <tools>      Auto-approve specific tools (comma-separated)
--no-thinking               Disable extended thinking
--verbose                   Show full tool outputs

Commands

atessa                      Start interactive mode
atessa init                 Setup wizard (API key, model, config)
atessa config [key] [val]   View or edit configuration
atessa commit [-a] [-p]     AI-generated commit messages
atessa index [-s] [-r]      Manage semantic codebase index
atessa sessions             List sessions for current directory
atessa mcp <add|list|rm>    Manage MCP tool servers

Configuration

Run atessa init or set environment variables:

ANTHROPIC_API_KEY=sk-ant-...         # required
ATESSA_MODEL=claude-sonnet-4-5-20250929        # optional

Config file location: ~/.config/atessa/config.json

Supported models

Claude models:

• claude-sonnet-4-5-20250929 — Sonnet 4.5 (recommended)
• claude-opus-4-5-20251101 — Opus 4.5
• claude-haiku-4-5-20251001 — Haiku 4.5 (fast)
• claude-sonnet-4-20250514 — Sonnet 4

Third-party:

• MiniMax-M2.1, GLM-4.7, mimo-v2-flash

Tools

Tools are organized into skills that the agent loads as needed.

Software Engineer — file operations (create_file, edit_file, delete_file), code search (grep, find_files, semantic_search), shell (run_command), git (git_status, git_diff, git_commit, git_branch, git_log), checkpoints, fetch_url

Researcher — web_search, scrape_url, create_note, list_notes, get_note

Security Researcher — vulnerability scanning, network analysis, mobile security tools

Base tools (always available) — load_skill, unload_skill, list_skills, ask_user, create_tasks, update_task, spawn_agent, wait_for_agent, remember

MCP integration

Atessa supports Model Context Protocol servers for extending the agent with external tools.

atessa mcp add my-server -- npx @my/mcp-server
atessa mcp list
atessa mcp remove my-server

Programmatic API

Use Atessa as a library in your own applications:

import { createAtessaCore } from 'atessa/core';

const agent = createAtessaCore({
  apiKey: process.env.ANTHROPIC_API_KEY,
  model: 'claude-sonnet-4-5-20250929',
  workingDirectory: process.cwd(),
});

agent.on('message', (event) => console.log(event.content));
agent.on('tool_start', (event) => console.log(`Running: ${event.name}`));

await agent.run('Add error handling to the API routes');

See exports.ts for the full public API surface.

Development

git clone https://github.com/ssakone/atessa.git
cd atessa
npm install
npm run dev          # watch mode
npm start            # run locally
npm test             # run tests
npm run typecheck    # type checking
npm run lint         # linting

License

MIT