We respect the open-terminal team. If you spot an inaccuracy, open an issue — we want this fair and factual.
Overview
| Feature | Open Computer Use | open-terminal | Claude.ai | OpenAI Operator |
|---|---|---|---|---|
| Self-hosted | Yes | Yes | No | No |
| Any LLM | Yes (OpenAI-compatible) | Any (via Open WebUI) | Claude only | GPT only |
| Code execution | Full Linux sandbox | Sandbox / bare metal | Sandbox | No |
| Live browser | CDP streaming (shared, interactive) | No | Screenshot-based | Screenshot-based |
| Terminal | ttyd + tmux (persistent, side panel) | PTY + WebSocket | IDE + terminal | N/A |
| Sub-agent | Claude Code CLI, interactive TTY + MCP | N/A | Built-in | N/A |
| Skills system | 13 built-in (auto-injected) + custom | Open WebUI native (text-only) | Custom instructions | N/A |
| Document creation | PPTX/DOCX/XLSX/PDF via skills | No | Via code | N/A |
| File preview | Server-side (docx/pdf/pptx/xlsx/code) | Client-side (via Open WebUI) | IDE | N/A |
| Container isolation | Docker (runc), per chat | Shared container (OS-level users) | Docker (gVisor) | N/A |
| MCP server | Streamable HTTP | FastMCP (stdio + streamable-http) | N/A | N/A |
| Image size | ~11 GB (full stack) | ~2 GB / ~200 MB / ~100 MB | N/A | N/A |
| Setup complexity | Docker Compose + reverse proxy + env config | Single docker run or pip install | N/A | N/A |
| Jupyter notebooks | No | Yes (per-session kernels via nbclient) | No | No |
| Bare metal | No (Docker required) | Yes (pip install open-terminal) | No | No |
| Port proxy | No | Yes (HTTP reverse-proxy to localhost services) | No | No |
| Ecosystem | Multi-client MCP (Open WebUI, n8n, OpenAI Agents SDK, LiteLLM) | Native Open WebUI integration + Terminals orchestrator | N/A | N/A |
Architecture and isolation
Open Computer Use creates a fresh Docker container per chat session. If the AI breaks something — installs wrong packages, corrupts files, fills disk — only that chat is affected. Next chat starts clean. Containers are GC’d after idle timeout; per-container limits (2 GB RAM, 1 CPU default) are enforced. open-terminal runs a single container (or bare-metal process) shared across sessions. Multi-user mode creates OS-level accounts with isolated home directories, file-ownership enforcement, and path validation. For container-per-user, the separate Terminals project orchestrates dedicated containers. Why it matters. Non-technical users + agent executing arbitrary code is the worst case for a shared env. Container-per-chat makes every session disposable. Trade-off. 2 GB per container is a ceiling, not an allocation — idle or light tasks use very little. Chromium and LibreOffice can push the ceiling. open-terminal is lighter but shares kernel and network between users.MCP tools — different design philosophies
Open Computer Use — 5 high-level tools:| Tool | Role |
|---|---|
bash_tool | Run commands with progress streaming, 15s heartbeats, 30K output cap |
view | Read files/dirs with line numbers, base64 images, binary detection |
create_file | Create files with auto-parent-dir |
str_replace | Find-and-replace with uniqueness validation |
sub_agent | Delegate to Claude Code (model selection, session resume, MCP auto-config, cost tracking) |
bash_tool for search, process mgmt, everything) vs. fine-grained operations that don’t require shell knowledge.
Security
| Aspect | Open Computer Use | open-terminal |
|---|---|---|
| Isolation | Docker containers | OS accounts (chmod 2770 + group membership) |
| Privilege escalation | Non-root + passwordless sudo; no-new-privileges | Non-root + passwordless sudo (full image); no sudo in slim/alpine |
| Resource limits | Per-container | OS-level only |
| Egress firewall | Docker network policies | Built-in DNS whitelist (dnsmasq + iptables + ipset) |
| API auth | Bearer token (MCP_API_KEY) | Bearer token (hmac.compare_digest) |
| Path traversal | Sanitized chat_id + safe_path | resolve_path + is_path_allowed |
What Open Computer Use offers that open-terminal doesn’t
- Document creation skills — 13 built-in with scripts and templates for pptx/docx/xlsx/pdf
- Skill auto-injection — structured instructions in the system prompt + per-user skills via Settings Wrapper
- Live shared browser — Playwright + CDP, AI automates via CDP, user watches/interacts in the same Chromium
- Claude Code sub-agent — model selection, session resume, cost tracking, auto-configured MCP servers
- Server-side file preview — renders from any MCP client, not tied to Open WebUI
- Container-per-chat isolation
- Persistent terminal via ttyd + tmux; full escape hatch
- Pre-installed stack (~180 packages: LibreOffice, Playwright, Tesseract, OpenCV, ImageMagick, GitLab CLI, fonts, ML libs)
- Vision AI skill
- Multi-client MCP tested with Open WebUI, n8n, OpenAI Agents SDK, LiteLLM
- Container resurrection — saved metadata recreates GC’d containers with same volumes/env
- Smart tool output — bash_tool streams with heartbeats, caps output, semantic exit codes
What open-terminal offers that we don’t
- Jupyter notebooks — per-session kernels
- Bare-metal mode —
pip install, no Docker - Port proxy — HTTP reverse-proxy to localhost services
- Lightweight image variants (slim ~200 MB, alpine ~100 MB)
- Document text extraction as API endpoint (11 formats)
- Process stdin — send input to running processes
- Session CWD tracking
- Runtime package install via env vars
- Docker-in-Docker — Docker CLI + Compose + Buildx pre-installed
- TOML config files
- Per-process JSONL logs with retention
- Simpler setup — single
docker run