Channel settings
Each ChannelGate conversation has its own settings. A conversation can be a Slack channel, group, DM, or a supported conversation on another platform. Threads share that conversation’s workspace and permissions, while a thread can have its own engine session and runtime selection.
Where to configure
Section titled “Where to configure”In the admin interface, open Conversations, select a conversation, and use its settings tabs. On Slack, open Settings from a reply’s footer to see the settings available to you. The Slack modal distinguishes channel settings from this thread runtime controls.
| Setting or tab | Purpose |
|---|---|
| Access | Read-only, Worker, or Admin mode; Auto, Lean, network, user and manager policies |
| MCP Connections | Conversation-level shared connections and token fallback preferences |
| Cloud MCP | Additional servers or app groups selected separately for each engine |
| Environment | Write-only variables for this conversation |
| Skills | Explicit skill grants and the followed skill template |
| Runtime | Engine, model, effort, working folder, and related runtime options |
| Instructions | Standing instructions specific to the conversation |
| Memory | Conversation memory and review information |
Configure a project conversation
Section titled “Configure a project conversation”- Approve the people who will use the bot and choose the conversation’s access policy.
- Choose Worker if the agent needs to edit files or run commands. Turn Auto on only when you intend to automatically approve tool requests.
- Choose the engine and model, or leave the engine at Default (global setting) to inherit the gateway default.
- Select a relevant skill template and add project-specific skills.
- Configure the necessary app connections or environment variables in this conversation’s scope.
- Save the conversation settings, then send a small test request in a fresh thread.
For example, a project instruction can say:
This conversation maintains the example-store project.Read AGENTS.md before changing code. Use staging for deployment checks.Record durable project decisions in this conversation's memory.Do not paste credentials into instructions. Use Environment or the relevant connection control.
Workspace and save behavior
Section titled “Workspace and save behavior”Working folder accepts a custom path; blank uses the default conversation folder. Browse… helps an administrator select an existing folder. Saving a new path changes the workspace used by future runs; existing files remain where they were. Reset to default changes the selection after saving, without copying the old project’s files.
Most saved conversation changes apply on the next message. Secrets and Codex sign-in controls have their own save actions. Existing threads can retain engine sessions and thread pins, so use a fresh thread when checking inherited defaults.
Who can change what
Section titled “Who can change what”Gateway admins can manage conversation access. Delegated channel managers can use the Slack Access settings allowed by the management policy, but cannot make themselves organization admins. Runtime changes have a separate organization policy. Conversation skill assignments are available to members; organization templates remain admin-managed.
See Permissions, Models and effort, Shared workspaces, and Environment variables.
Monthly channel spending limit · Enterprise
Section titled “Monthly channel spending limit · Enterprise”Set a positive USD amount for the conversation’s monthly spending limit, or leave it unset for
no cap (monthlySpendingLimitUSD: null). The limit compares recorded spend in the current UTC
calendar month before admitting another run. Foreground, scheduled, API and background agent
launches use the same admission check.
This is not a provider invoice cap. A running job can finish above the limit, and concurrent launches can overshoot because spend is not reserved in advance. Unknown-cost records in the current month block admission until pricing is repaired or an admin clears the cap. Usage and costs outside this gateway are not controlled.
Setting a limit requires Enterprise. A saved limit refuses new runs after downgrade instead of silently removing the safety control; an admin can clear it without Enterprise. Existing recorded usage is retained and the accounting month changes at midnight UTC on the first of the month.