Models
Use supported OpenAI, Anthropic and Google models with your provider credentials, or connect a model service you operate.
Connect your agents and services.
Bring agents, models, tools and messaging under your policies through supported connections or Atellagent Client.
- ChatGPT / CodexAvailable
- Claude / Claude CodeAvailable
- GeminiPlanned
- Microsoft CopilotPlanned
Models, workflows and channels.
Connect the services your agents depend on. Policy coverage follows each connection's supported actions.
For hosted GitHub coding, explore Atellagent Code. Free records policy decisions; paid plans enable supported enforcement. View pricing
Connect Codex or Claude Code.
Finish setup in the desktop app. No platform configuration file or enrollment token to copy.
Discovery & reporting shows supported agent settings and declared MCP servers on your machine. Review them before connecting; detection does not approve or start a connection.
- Install and sign in
Install the desktop app and sign in to your organization.
- Request device approval
An organization administrator reviews your device request.
- Choose the agent
Select Codex or Claude Code in Connections. This connection needs separate approval and your consent.
- Activate and verify
Review hook settings, restart the agent when instructed, and start the service. Test a supported action and inspect its decision.
Download the desktop app
Check the installer information for system requirements and signing status.
Checking desktop downloads…
Check what your connection covers.
Both agents support checks on submitted prompts and supported tool calls. These checks cover enabled host hooks, not every action the agent can take.
| Agent | Reported outcomes |
|---|---|
| Codex | Completion reports when available; success is unknown. |
| Claude Code | Tool success or failure when the host reports it. |
Coverage limits
In Enforce mode, client control failures return a denial at supported hooks. The host must deliver and honor it; host errors or timeouts can leave actions unchecked.
With Enforce + Strict enabled, unrecognized actions are denied by default. Missing result records do not prove success.
Claude Cowork hook coverage is not currently supported.
Govern MCP tool calls.
MCP connects agents to tools. Check tool requests and apply data-sharing controls to responses on governed connections.
Operate your own MCP server? Enterprise MCP protects its tool execution, including direct clients you enable.
MCP permissions and coverage
Provider permissions still apply. Connecting a provider account does not install protection on that provider's server. Protection covers calls routed through a governed connection or your server's Enterprise MCP handler.
Integrate your own agents and services.
Connect your own agent, model, workflow runtime, channel or detector through Atellagent Client's Python SDK or bridge.