Agent Platforms: 16 solutions compared
Which Agent Platforms tools can you self-host, run on a Mac or connect to? FrontierStack's catalogue covers 16 Agent Platforms options: Browserbase, ChatGPT Work Sites, Devin, E2B, GitHub Copilot, Herdr and others. For 11 of them FrontierStack installs the service on a Mac or server, or connects to one you already run, and checks status, ports and certificates where they apply; 5 are hosted services FrontierStack connects to with an API token and monitors.
| Name | Type | What FrontierStack does |
|---|---|---|
| Browserbase | Self-hosted or connected | Install or connect; status, ports and certificate checks where they apply |
| ChatGPT Work Sites | Self-hosted or connected | Install or connect; status, ports and certificate checks where they apply |
| Devin | Hosted service | Connect with an API token; reachability and status alerts |
| E2B | Self-hosted or connected | Install or connect; status, ports and certificate checks where they apply |
| GitHub Copilot | Self-hosted or connected | Install or connect; status, ports and certificate checks where they apply |
| Herdr | Self-hosted or connected | Install or connect; status, ports and certificate checks where they apply |
| Hyperagent | Hosted service | Connect with an API token; reachability and status alerts |
| Instinct | Hosted service | Connect with an API token; reachability and status alerts |
| Lakebed | Self-hosted or connected | Install or connect; status, ports and certificate checks where they apply |
| Lumina | Self-hosted or connected | Install or connect; status, ports and certificate checks where they apply |
| Manus | Hosted service | Connect with an API token; reachability and status alerts |
| OpenAI Operator | Hosted service | Connect with an API token; reachability and status alerts |
| OpenHands (OpenDevin) | Self-hosted or connected | Install or connect; status, ports and certificate checks where they apply |
| Rex | Self-hosted or connected | Install or connect; status, ports and certificate checks where they apply |
| Sandcastle | Self-hosted or connected | Install or connect; status, ports and certificate checks where they apply |
| SuperAGI | Self-hosted or connected | Install or connect; status, ports and certificate checks where they apply |
Browserbase
Headless browser infrastructure for AI agents (cloud sessions)
Browserbase runs managed, stealth headless browsers in the cloud for AI agents and web automation; start a session, drive it with Playwright/Puppeteer/Selenium (CDP) or the Stagehand SDK, with live view, recordings, proxies and captcha handling. Manage with the REST API (sessions/projects) and SDKs; keep the API key + project ID in Keychain. FrontierStack monitors it live; shows running/total sessions and alerts when concurrent sessions exceed your threshold (a cost/leak guard).
Install or connect Browserbase; then check its status, ports and certificate from FrontierStack. The same screen links to firewall checks and backups where they apply.
ChatGPT Work Sites
OpenAI's work agent + published Sites/web apps (alpha)
ChatGPT Work is OpenAI's agent for long-running work tasks (GPT-5.6); it gathers information across your apps and produces finished docs, sheets, slides and web apps, with Scheduled Tasks for recurring/monitored work. Its Sites feature (public beta) publishes your work as an interactive site or web app shared by URL; live dashboards, project trackers, launch calendars, prototypes and internal portals. FrontierStack support is alpha. For an Enterprise/Work workspace the pane connects live to OpenAI's admin controls: members and groups via the SCIM API, plus a Compliance API (audit feed) health check with a workspace ID; alerting when members exceed your licensed seats (over-provisioning guard), when the audit feed fails, or when the admin APIs are unreachable (keys in Keychain). Published Sites keeps a live list of your *.chatgpt.site apps with reachability checks and one-click Alerts watching. Local Runtime Policy governs the admin-enforced requirements.toml that constrains the ChatGPT desktop app / Codex CLI / IDE extension (approvals, sandbox, network, Computer Use); inspect every layer on This Mac, deploy a reference policy via the privileged helper, and check/deploy it across fleet servers. Track OpenAI key spend in Model Costs.
Install or connect ChatGPT Work Sites; then check its status, ports and certificate from FrontierStack. The same screen links to firewall checks and backups where they apply. Full page →
Devin
Cognition's AI software engineer (cloud API)
Devin (Cognition) is a cloud autonomous software-engineer agent; give it a task and it plans, codes, runs and tests in its own cloud workspace. Cloud/enterprise-hosted; integrate via the Devin API and Slack/IDE, manage sessions in its dashboard. API key in Keychain. Cloud sessions, not local instances.
Devin is hosted, so FrontierStack does not try to install it. Connect an API token to check reachability and published status beside the services you host yourself; FrontierStack alerts you if it stops answering. Credentials stay in Keychain.
E2B
Secure cloud sandboxes for AI agents (spin up / clone / kill)
E2B runs secure, isolated cloud sandboxes for AI-agent code execution; spin up a sandbox, run code/commands, fork (clone) it, pause and resume, then kill it. Manage with the e2b CLI (npm i -g @e2b/cli; e2b sandbox spawn|list|kill) and the SDKs (Python/JS), or self-host the infra. Keep the API key in Keychain. Ideal as the execution backend for your agents.
Install or connect E2B; then check its status, ports and certificate from FrontierStack. The same screen links to firewall checks and backups where they apply.
GitHub Copilot
AI pair programmer; live org seat/usage monitor
GitHub Copilot is GitHub's AI pair programmer (completions, chat and the Copilot coding agent). For a Copilot Business/Enterprise org you can watch seats live here; total, active and inactive (paid-but-unused) seats and pending invitations; via the org billing API. Add a classic or fine-grained PAT (org owner, manage_billing:copilot or read:org). Pin it in Alerts to be warned when inactive seats (wasted spend) cross a threshold or the API can't be reached.
Install or connect GitHub Copilot; then check its status, ports and certificate from FrontierStack. The same screen links to firewall checks and backups where they apply.
Herdr
Agent multiplexer; tmux for AI coding agents (terminal)
Herdr (herdr.dev) is an open-source agent multiplexer: a single ~10 MB Rust binary that runs many terminal AI coding agents in parallel with panes, tabs and workspaces; tmux, but agent-aware. It tracks each agent's semantic state in a sidebar (blocked / working / done / idle) so you can see which one needs input, and sessions persist across detach/reattach; including over SSH from another machine or a phone. A CLI and a local JSON socket API let the agents themselves open workspaces, split panes and spawn helpers. Works with Claude Code, Codex, Gemini CLI, Droid, OpenCode, Aider, Copilot CLI and any terminal agent. brew install herdr (macOS/Linux; AGPL-3.0). A terminal tool; no port or web UI of its own; complements FrontierStack's Agent Sessions.
Install or connect Herdr; then check its status, ports and certificate from FrontierStack. The same screen links to firewall checks and backups where they apply. Full page →
Hyperagent
Cloud agent platform; FrontierStack as its on-site hands (cloud)
Hyperagent (hyperagent.com; not Hyperbrowser's open-source HyperAgent browser tool) builds specialist agents with their own instructions, tools, knowledge and memories, triggered from Slack, Telegram, email, a webhook, a schedule or by hand. It is hosted only: agents run on Hyperagent's own cloud computers, so FrontierStack is the on-site half of a Hyperagent location. Hyperagent → this Mac: add the FrontierStack connector (frontierstack.app/plugin/mcp) as a Custom MCP server and an agent can check health, read logs, run diagnostics and, with your approval, restart services, under this Mac's permissions and audit log. This Mac → Hyperagent: sign in to Hyperagent's MCP server to list agents, start runs and follow them, and answer pending approvals (or get alerted when they wait). Per location: an alert raised while this Mac is at a location can start that location's agent (rate-limited, because every run spends Hyperagent credit), and FrontierStack writes the agent's standing instructions. Webhook triggers send Hyperagent's required secret header.
Hyperagent is hosted, so FrontierStack does not try to install it. Connect an API token to check reachability and published status beside the services you host yourself; FrontierStack alerts you if it stops answering. Credentials stay in Keychain.
Instinct
Personal AI assistant you text or call (cloud, no API yet)
Instinct is a personal assistant that learns what you're working on and what matters to you, and gets things done the way a person would; by phone and on a computer: disputing a bill, sending a gift, ordering groceries, booking appointments, planning a trip. You talk to it by text message or phone call; the web app is at app.instinct.com. Instinct has no public API yet, so FrontierStack can't read usage or billing from it. Record what you pay in Plan & Cost below and it counts toward your monthly AI spend in Spend & Billing.
Instinct is hosted, so FrontierStack does not try to install it. Connect an API token to check reachability and published status beside the services you host yourself; FrontierStack alerts you if it stops answering. Credentials stay in Keychain.
Lakebed
Agent-native runtime for full-stack TypeScript capsules (alpha)
Lakebed (alpha) is an agent-native CLI and runtime for small full-stack TypeScript apps called capsules; server (schema/queries/mutations/webhooks), Preact client and shared code in one structured shape agents can build unattended. npx lakebed new <name> scaffolds one, npx lakebed dev runs the local runtime on :3000, npx lakebed deploy publishes it (anonymous, or claimed for env vars + outbound fetch). Inspect what agents built with npx lakebed db list|dump, logs and inspect <deploy>. Lakebed is free and there is nothing to install beyond Node: the CLI runs through npx. FrontierStack support is alpha: the pane checks npx can run the CLI, creates capsules in a folder you choose, lists them with their local dev servers (start, stop, open, deploy, logs) and your hosted deploys, and the Agent Platforms board shows running runtimes.
Install or connect Lakebed; then check its status, ports and certificate from FrontierStack. The same screen links to firewall checks and backups where they apply. Full page →
Lumina
Local-first desktop AI agent with persistent memory & voice
Lumina is a local-first agentic harness; a native desktop assistant (Python + PySide6/Qt, no Electron) built for running on your own hardware rather than someone's cloud. It keeps multi-tier persistent memory with temporal decay, so it remembers across sessions, and ships around 70 tools covering the filesystem, sandboxed Python and shell execution, and browser automation, plus subagents for delegated work and scheduled background tasks. It can also write its own tools, clone a voice, and swap personas. It is model-agnostic: llama.cpp is the recommended backend, and it also drives Ollama, LM Studio and vLLM; all of which FrontierStack already manages; or any cloud provider (Claude, Gemini, OpenAI, DeepSeek, Groq, OpenRouter and more) via an OpenAI-compatible endpoint. Install by cloning the repo, pip install -r requirements.txt and playwright install chromium; copy config.example.py to config.py and point it at your backend. Two caveats on a Mac: macOS support is still in testing (Linux is the primary target), and it needs brew install portaudio first for the voice stack. The project targets an NVIDIA GPU with CUDA for local inference, and documents no Apple-silicon/Metal path; so on this Mac, point it at a model served by Ollama, LM Studio or a cloud provider rather than expecting local GPU inference. Apache-2.0.
Install or connect Lumina; then check its status, ports and certificate from FrontierStack. The same screen links to firewall checks and backups where they apply. Full page →
Manus
Autonomous general AI agent (cloud API)
Manus is a general autonomous AI agent that plans and executes multi-step tasks (research, browsing, coding) in the cloud. Cloud-hosted; drive it through its API and manage runs in the Manus dashboard. Keep the API key in Keychain. These are cloud agent sessions, not locally spun-up instances.
Manus is hosted, so FrontierStack does not try to install it. Connect an API token to check reachability and published status beside the services you host yourself; FrontierStack alerts you if it stops answering. Credentials stay in Keychain.
OpenAI Operator
OpenAI's browser-using agent (cloud)
OpenAI Operator is OpenAI's agent that uses a web browser to carry out tasks on your behalf, built on the Computer-Using Agent model. Cloud-hosted in ChatGPT; programmatic access is via OpenAI's Responses/computer-use tooling. Keep the OpenAI key in Keychain (see the Model Costs pane). Cloud sessions, not local instances.
OpenAI Operator is hosted, so FrontierStack does not try to install it. Connect an API token to check reachability and published status beside the services you host yourself; FrontierStack alerts you if it stops answering. Credentials stay in Keychain.
OpenHands (OpenDevin)
Open-source autonomous AI software engineer (self-hosted)
OpenHands (formerly OpenDevin) is an open-source autonomous coding agent; it writes code, runs commands and browses, each task in an isolated sandbox container. Self-host via Docker (it drives Docker to spin up per-session runtime sandboxes); web UI on :3000. Start/stop it from its Docker pane here. Point it at your local LLM servers or a cloud key.
Install or connect OpenHands (OpenDevin); then check its status, ports and certificate from FrontierStack. The same screen links to firewall checks and backups where they apply.
Rex
Terminal multiplexer for agents and long-running work (beta)
Rex (superlogical.com/rex) is a terminal multiplexer built on libghostty, in beta. One rex binary is the server and the CLI: the Mac app runs a local server, and Linux and FreeBSD hosts run servers it can connect to, so sessions keep running with nothing attached. Programs report whether they are working, blocked or finished with its status protocol (OSC 7501), and its agent skill teaches Claude Code, Codex, OpenCode and Pi to open tabs, split panes and read terminals. FrontierStack shows each server's sessions, the programs waiting for you and what they run, here and across your servers, without starting a server or reading what's on screen.
Install or connect Rex; then check its status, ports and certificate from FrontierStack. The same screen links to firewall checks and backups where they apply. Full page →
Sandcastle
Orchestrate sandboxed coding agents (isolated containers)
Sandcastle runs AI coding agents inside isolated sandboxes; each agent gets its own Docker/Podman (or Vercel) container with full isolation, and commits made in the sandbox are patched back to the host automatically. Provider-agnostic (Claude Code, Codex, OpenCode…), local/offline, MIT-licensed; orchestrate runs in TypeScript with a single sandcastle.run(). A safe way to let agents execute untrusted code without touching your machine.
Install or connect Sandcastle; then check its status, ports and certificate from FrontierStack. The same screen links to firewall checks and backups where they apply.
SuperAGI
Open-source autonomous-agent framework (self-hosted)
SuperAGI is a dev-first framework for building, running and managing autonomous AI agents; a GUI, toolkits, concurrent agents, telemetry and a tool marketplace. Self-host with Docker Compose; the web GUI serves on :3000 (backend API on :8001). Build agents, connect tools and watch runs from its dashboard.
Install or connect SuperAGI; then check its status, ports and certificate from FrontierStack. The same screen links to firewall checks and backups where they apply.
Run it from your Mac.
FrontierStack installs, monitors and secures services on this Mac and on linked servers; and keeps your passwords and keys out of any AI's sight.
Download FrontierStackApple notarized · Safe & secure · macOS 13 Ventura+
