By Shekhar — Founder, AgenticMarket. Published September 07, 2026.

Quick Links:

• VS Code Extension: Install from Marketplace (code --install-extension agenticmarket.andromity-agent)
• GitHub Repository: agenticmarket/andromity (⭐ MIT Licensed)
• Python / CLI Package: PyPI - andromity
• AgenticMarket Lander: agenticmarket.dev/andromity

---

Most AI coding assistants suffer from one of two extremes:

1. The "Black Box" Problem: They autonomously churn through your files in the background, making destructive edits without an upfront plan, leaving you scrambling through git diffs trying to figure out what broke.
2. The "Passive Autocomplete" Problem: They only generate single-file snippets inside your editor, incapable of running tests, debugging failures, managing multi-file refactors, or running unattended maintenance tasks.

We built Andromity to bridge that gap.

Andromity is a private, Bring-Your-Own-Key (BYOK) autonomous AI coding agent designed to run as both a native VS Code Extension and a standalone Terminal TUI Workspace. Loved by over 100+ early developers, it pairs autonomous execution with rigorous Trust Governance, interactive Live Waterfall Traces, built-in Cron Scheduling, and native Model Context Protocol (MCP) support.

Here is a deep look into what makes Andromity different, how its architecture works, and how you can integrate it into your daily workflow today.

---

1. The Core Architecture: Transparency Through Live Waterfall Traces

When you ask an autonomous agent to build a complex feature—say, implementing OAuth2 refresh token rotation across five files—you shouldn't have to guess what it's thinking.

In Andromity, every task executes through an observable Waterfall Trace view:

How the Execution Pipeline Works:

1. Repository Indexing & Context Mining: Andromity scans relevant files, AST symbols, and configuration dependencies without filling your context window with irrelevant files.
2. Blueprint Generation: Before writing a single line of code, Andromity compiles an interactive, step-by-step implementation plan.
3. Step Approval & Verification: You can inspect, modify, approve, or skip each milestone before execution begins.
4. Isolated Step Diffs: As each step completes, live inline diffs are rendered in your editor. You inspect changes with syntax highlighting before they are committed to disk.
5. One-Click Session Rollback (/undo): If an agent takes an unexpected path, typing /undo instantly reverts the entire turn and restores your files to their exact previous state.

---

2. Trust Governance: You Are Always in Control

Security in AI agents cannot be an afterthought. Running autonomous tools with arbitrary terminal access can result in accidental deletions, leaky API calls, or broken system configurations.

Andromity introduces a 4-tier Trust Governance model governed by workspace boundaries:

In untrusted workspaces, Andromity strictly enforces read-only mode until you explicitly designate the directory as trusted. You can switch modes instantly with a single keyboard shortcut as your confidence in the agent grows.

---

3. 396+ Model Support — Zero Lock-In, Local Ollama Ready

Why should an editor or extension dictate which LLM you use?

Andromity is built on top of LiteLLM under the hood, giving you complete freedom to plug in any foundation model using your own API keys, or run completely offline with local models:

• Cloud Frontier Models: Anthropic Claude 3.7 Sonnet & 3.5 Sonnet, OpenAI GPT-4o & o3-mini, Google Gemini 2.5 Pro & Flash, DeepSeek R1 & V3, Mistral Large.
• Ultra-Fast Inference Engines: Groq (Llama 3.3 70B at 300+ tok/sec), Cerebras, OpenRouter, Together AI.
• 100% Local & Free with Ollama: Run ollama run deepseek-r1:14b or qwen2.5-coder:32b locally. Zero telemetry, zero network calls, zero API bills.
• Mid-Session Model Swapping: Need Claude 3.7 Sonnet for architectural planning, but want DeepSeek V3 or Groq for fast code implementation? Press Ctrl+L to swap models mid-conversation without losing your conversation history.

---

4. Built-in Cron Scheduler: AI That Works While You Sleep

One of the most requested features by software engineers is autonomous background maintenance. Why spend morning standup fixing broken lint rules, bumping dependencies, or debugging nocturnal CI test failures?

Andromity features an industry-first built-in Cron Scheduler:

bashCOPY
# Example: Auto-run test suite, fix test failures, and commit every night at 2:00 AM
/cron  -->  "run pytest, fix any failing unit tests, format code, and commit"  -->  0 2 * * *

How Scheduled Tasks Operate:

• Tasks persist in your repository under .andromity/crons.json.
• When the timer fires, Andromity initializes a headless background session, clones workspace context, runs the requested task, and validates the changes against your test suite.
• If running in FULL or YOLO mode, commits are generated automatically with conventional commit messages. You wake up to clean, working pull requests.

---

5. Parallel Subagents & Multi-Session Workflows

Real software engineering is rarely sequential. You shouldn't have to wait for an agent to finish writing a 500-line database migration before you can ask questions about an authentication bug.

With Andromity, you can spawn parallel subagents at any time:

[Main Session]   --> Designing and implementing Feature A
  └── [Subagent 1] --> Researching API docs for Stripe Webhooks in background
  └── [Subagent 2] --> Writing unit tests for Feature B

• Switch between active sessions instantly with Ctrl+O.
• Each session maintains its own isolated context window, memory buffer, and file change log.
• Trigger /compact to compress older turns when working on massive repositories to preserve context limits.

---

6. Native Model Context Protocol (MCP) Integration

Andromity natively acts as an MCP Client. By pairing Andromity with AgenticMarket's 100% Free Verified MCP Registry, you give your agent immediate access to external tools without manual configuration:

jsonCOPY
// ~/.andromity/mcp.json
{
  "mcpServers": {
    "web-reader": {
      "command": "npx",
      "args": ["-y", "agenticmarket-proxy", "@agenticmarket/web-reader"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "agenticmarket-proxy", "@agenticmarket/github"]
    }
  }
}

1-Command Server Installation:

Using the official AgenticMarket CLI, installing any verified MCP server into Andromity takes a single command:

bashCOPY
npx agenticmarket install @creator/web-reader --ide andromity

The CLI automatically detects your Andromity environment, writes the JSON configuration into ~/.andromity/mcp.json, and loads tool definitions dynamically during agent initialization.

---

How Andromity Compares

---

Getting Started in Under 60 Seconds

Option 1: VS Code Extension (Recommended)

1. Open VS Code and run in terminal:

   bashCOPY
   code --install-extension agenticmarket.andromity-agent
   

   Or search for "Andromity" in the VS Code Extensions tab.
2. Click the Andromity robot icon in your activity bar.
3. Choose your provider (Anthropic, OpenAI, Gemini, Groq, or Ollama), enter your API key, and begin coding.

Option 2: Standalone Terminal CLI

For developers who live in tmux, Neovim, or standard terminal windows:

bashCOPY
# macOS & Linux
curl -fsSL https://raw.githubusercontent.com/agenticmarket/andromity/main/install.sh | bash

# Windows (PowerShell)
irm https://raw.githubusercontent.com/agenticmarket/andromity/main/install.ps1 | iex

# Or via pipx
pipx install andromity

Launch the terminal workspace in any project:

bashCOPY
cd my-project
andromity

---

Conclusion & Resources

Software development in 2026 demands tools that give you the velocity of autonomous AI without surrendering your control, privacy, or choice of models.

Andromity was built to be the agent we always wanted: completely open source, deeply transparent with live waterfall traces, respectful of your workspace trust boundaries, and pre-wired for the Model Context Protocol ecosystem.

• VS Code Extension: Download on Visual Studio Marketplace
• GitHub Repository: Star & Fork on GitHub
• Verified MCP Directory: Browse Free MCP Servers on AgenticMarket
• Andromity Platform Guide: agenticmarket.dev/andromity

If you build something with Andromity or have feedback on our live waterfall traces, join our community on GitHub or reply to us on X @agenticmktdev.