anteroom 1.24.5

Anteroom - your gateway to AI conversation

Anteroom

The room before the room — a secure, private space between you and the AI.

Self-hosted ChatGPT-style web UI and agentic CLI that connects to any OpenAI-compatible API.
Install with pip. Run locally. Own your data.

Documentation • Quick Start • Features • Security

---

Why Anteroom?

An anteroom is the private chamber just outside a larger hall --- a controlled space where you decide who enters and what leaves. That's exactly what this is: a secure layer on your machine between you and any AI, where your conversations never touch someone else's cloud.

Anteroom connects to any OpenAI-compatible endpoint --- your company's internal API, OpenAI, Azure, Ollama, LM Studio, or anything else that speaks the OpenAI protocol. Built to OWASP ASVS L1 standards because your conversations deserve real security, not security theater.

One command. No cloud. No telemetry. No compromise.

---

Quick Start

pip install anteroom
aroom init         # Interactive setup wizard

Or create ~/.anteroom/config.yaml manually:

ai:
  base_url: "https://your-ai-endpoint/v1"
  api_key: "your-api-key"
  model: "gpt-4"

aroom --test       # Verify connection
aroom              # Web UI at http://127.0.0.1:8080
aroom chat         # Terminal CLI (interactive REPL)
aroom exec "task"  # Execute a task (non-interactive, for scripting/CI)
aroom --version    # Show version

---

Features

Web UI

Full-featured ChatGPT-style interface with conversations, projects, folders, tags, file attachments, MCP tool integration, prompt queuing, canvas panels for AI-generated content (CodeMirror 6 editor), inline safety approvals, sub-agent progress cards for parallel AI execution, command palette, and four built-in themes.

CLI Chat

An agentic terminal REPL with built-in tools (read/write/edit files, bash, glob, grep, canvas, run_agent), MCP integration, sub-agent orchestration for parallel task execution, skills system, safety approval prompts for destructive operations, planning mode for multi-step tasks (explore with constraints, write a plan, then approve for execution), and Rich markdown rendering. Type while the AI works --- messages queue automatically.

aroom chat                          # Interactive REPL
aroom chat "explain main.py"        # One-shot mode
aroom chat -c                       # Continue last conversation
aroom chat --model gpt-4o "hello"   # Override model
aroom chat --plan                   # Start in planning mode
aroom chat                          # Then in REPL:
  > /plan                           # Enter planning mode
  > /plan approve                   # Exit planning mode and execute plan
  > /plan build a REST API          # Inline: planning mode + start prompt
  > /plan status                    # Show planning mode status

CLI Exec

Non-interactive exec mode for scripting, CI/CD pipelines, and automation. Execute a task and get JSON output with no approval prompts or interactive features. Perfect for embedding Anteroom in shell scripts, GitHub Actions, or other automation workflows.

aroom exec "summarize test_results.txt" --json      # JSON output for parsing
aroom exec "run tests" --timeout 300 --quiet        # CI/CD with timeout
aroom exec "generate report" --no-tools             # Disable tool use
aroom exec "refactor code" --no-conversation        # Skip persistence

Shared Core

Both interfaces share the same agent loop, storage layer, and SQLite database. Conversations created in the CLI show up in the web UI, and vice versa.

---

Security

Layer	Implementation
Auth	Session tokens, HttpOnly cookies, HMAC-SHA256
CSRF	Per-session double-submit tokens
Headers	CSP, X-Frame-Options, HSTS, Referrer-Policy
Database	Parameterized queries, column allowlists, path validation
Input	DOMPurify, UUID validation, filename sanitization
Rate Limiting	120 req/min per IP
Tool Safety	Destructive action approvals (CLI + Web UI), configurable patterns, sensitive path blocking
MCP Safety	SSRF protection, server-side error handling

Full details in SECURITY.md.

---

Documentation

For complete documentation including configuration, CLI commands, API reference, themes, MCP setup, skills, and development guides, visit anteroom.readthedocs.io.

---

Development

git clone https://github.com/troylar/anteroom.git
cd anteroom
pip install -e ".[dev]"
pytest tests/ -v

Backend	Python 3.10+, FastAPI, Uvicorn
Frontend	Vanilla JS, marked.js, highlight.js, KaTeX, CodeMirror 6
CLI	Rich, prompt-toolkit, tiktoken
Database	SQLite with FTS5, WAL journaling
AI	OpenAI Python SDK (async streaming)
MCP	Model Context Protocol SDK (stdio + SSE)

---

MIT License
Built for people who care about their conversations.
anteroom.readthedocs.io