Private builder, public manifest
Install the approved firm from Agentlas web.
Agentlas web keeps the private team-generation logic server-side. Desktop receives only the approved firm manifest, prompts, MCP install hints, and runtime requirements.
macOS desktop preview
Agentlas Desktop
The local AI team launcher for Agentlas firms. Install an approved firm, connect OpenAI, Claude, Gemini, Codex, or your local CLI, and run the CEO chat plus specialist agents from your Mac.
Check Mac download statusDownload for macOS - notarization pending
Bring your own OpenAI, Claude, Gemini, Codex, or local CLIAPI keys stay on your MacNo hosted runtime required
No Agentlas account is required for the Mac app. The DMG opens after Apple notarization, SHA256 hashes, file sizes, and release URLs are verified.
No account required for the DMGBring your own runtimeMCP installs from approved manifestsKeys stay local
Agentlas Desktop
Agentlas Desktop will be free to download without login after release. Public downloads are paused until Apple notarization, SHA256 hashes, file sizes, and release URLs are verified.
Web creates. Desktop runs.
The best way to run Agentlas firms on your Mac.
Build the private firm on Agentlas web, publish the approved manifest, then run it locally in Desktop with MCP servers, environment requirements, and your own runtime controls visible.
Local AI team launcher
Launch the CEO chat with specialist agents already attached.
A firm is not a blank chatbot. It opens as one CEO chat plus specialist agents, firm context, projects, automations, and library state on your Mac.
Bring your own runtime
Connect OpenAI, Claude, Gemini, Codex, or a local CLI.
Agentlas coordinates the team, but the model account stays yours. Runtime controls remain visible before the job leaves the chat.
MCP and keys stay local
See required keys, MCP servers, and setup before work begins.
Environment requirements are shown in the app, saved locally, and kept out of the public marketplace manifest. Desktop uses MCP data first and falls back only when offline.
From private builder to execution
Watch the screenshots change as the workflow moves down the page.
The flow starts with an approved firm from Agentlas web, installs its MCP definitions into Desktop, connects local runtimes, then runs the CEO chat and specialist agents on your Mac.
01Install the approved firm
Desktop reads the published Agentlas firm from MCP and shows the install action before anything runs locally.
Install the approved firm
Desktop reads the published Agentlas firm from MCP and shows the install action before anything runs locally.
Launch the installed team
After installation, the firm card becomes a direct launcher for the CEO chat and its specialist agents.
Open CEO chat with firm context
The selected firm stays visible beside the conversation, so users know which approved team is handling the job.
Type a real job
The first prompt goes to the installed CEO chat. Users start from a working team context, not a blank generic chatbot.
Confirm runtime and permissions
Runtime controls stay visible before the job is sent to a BYOK provider, account, or local CLI.
Run through the selected runtime
This capture shows Claude Code CLI being invoked from the Mac. OpenAI, Claude, Gemini, Codex, and local CLI paths follow the same visible execution model.
Open individual specialists
The installed firm also appears as individual agents, so users can jump into copy, support, review, pricing, SEO, or analytics work.
Fill required keys once
Library shows required environment variables and where to find them. Values stay in macOS Keychain and do not get pushed back to Agentlas web.
Release gate
Notarized DMGs first. Public no-login download second.
Public download redirects stay off until notarization, SHA256 hashes, file sizes, and GitHub Release asset URLs exist. The GitHub repository is live, but release assets stay hidden from the CTA until they are verified.
View release status JSON