Approve the people
Admins and approved users may talk to the agent. A channel can also admit explicitly granted guests. Tool permissions remain a separate decision.
ChannelGate AI OS is a company-governed AI operating system that coordinates your team’s AI agents, shared work and access.
People set the operating boundaries
Give agents the access the work needs.
Useful agents need the ability to act. ChannelGate gives operators explicit controls over who may use them, which tools a channel receives and where its agent process runs.
User authorizationChannel policyContainer isolation
Governance & security
Make authority explicit.
Choose the operating mode
Read-only, Worker and Admin define base tool access. Auto is a separate option that approves tool requests; an admin author in an Admin channel can use bypass permissions.
Grant the tools and knowledge
Select channel connections, skills and instructions for the task. Changing access changes what future runs receive.
Enforced organization instructions · Enterprise
Apply centrally required instruction policies across the organization’s channels. Ordinary channel and shared instructions remain core.
Mandatory organization skills · Enterprise
Require organization skill policies that channels cannot opt out of. Existing organization-wide skill grants, reusable skills and reviewed publishing remain core.
Review what happened
Conversation history, run usage records and administrative events help operators review work and configuration changes. Check evidence appropriate to the action; these are not a compliance certification.
Read the ChannelGate security modelIllustrative workflow
A policy decision before execution
The configured mode and tool permissions determine whether an action proceeds or needs approval.
Authorize the request
Check that the author is approved or explicitly admitted to this channel.
Apply channel policy
Provide the selected tools and runtime mounts under the configured operating mode.
Execute and review
Request approval where the policy requires it, then return progress, results or a clear refusal.
Frequently asked questions
Governance and security questions
Understand who can use an agent, which actions it can take and how to review its work.
Who can use an agent in a channel?
An administrator sets the channel's access policy, including whether it admits approved members, administrators only or explicitly granted guests. Direct messages require an approved user or administrator. Access to the conversation does not automatically grant every tool or connected account. See the access and permissions overview.
Does every action need a human approval?
Approval behavior depends on the channel's operating mode and approval settings. Read-only, Worker and Admin determine the base tool access; Auto is a separate approval setting. Configure those controls for the work you want to allow instead of assuming every action will pause for approval. The permissions overview documents the available controls.
How does ChannelGate enforce network access?
In the default proxy mode, conversation containers have no direct network access. A per-channel proxy checks each request. With Allow network off, AI engine endpoints and the channel's selected remote MCP services remain reachable. With it on, HTTP(S) requests can reach public destinations, while private and loopback addresses remain blocked. Legacy bridge mode and an administrator-enabled raw-network exception make the switch advisory. Admin-only Slack /sudo threads run on the host, outside the container network boundary. See the network controls and exceptions before choosing your deployment settings.
Can agents use personal and shared app accounts?
ChannelGate supports separate personal and agent connection identities. Which services an agent can reach depends on the connections and permissions granted to the conversation. A personal request should use the requester's connection, and a write needs the intended account. Read about credentials and account ownership and the data sent to connected services.
How can we review an agent's work?
Conversation history, usage records and administrative event history help your team inspect work and configuration changes. Review the resulting file or record in the destination system as part of the workflow, too. The operations overview describes the review tools, and memory and history explains how work stays available in a conversation.
Put it to work
Start with one useful workflow.
Explore the setup yourself, or build the foundation alongside Makeitfuture.