Agent runtime

Configuration → Agent runtime bundles settings for how agents run in CompanyGPT.

You can take over each setting with Manage via companyADMIN. As long as it is not managed, the value from the base configuration applies; the page shows it (source librechat.yaml). Managed values (source companyADMIN) apply to everyone as soon as they reload CompanyGPT. Each section has its own save button.

Scheduled chats

Scheduled chats are recurring chats with an agent that run at set times.

Setting	Meaning
Allow scheduled chats	Switches the feature on. Off as long as it is switched on neither here nor in the base configuration
Max schedules per user	How many scheduled chats a person may have at most
Min interval (minutes)	Shortest gap between two runs
Auto-disable after failures	After how many failures a scheduled chat is switched off
Require a chat project	Whether a scheduled chat must belong to a project

Who may use or create scheduled chats is set per role: role → companyGPT tab → feature Scheduled chats (Use, Create), see Roles.

If the base configuration switches scheduled chats off, a setting here cannot switch them on. In that case contact innFactory.

How people use scheduled chats is shown in the preview New chat interface.

Tool approval

With tool approval, CompanyGPT asks the person before an agent runs a tool. Rules are evaluated in the order deny → ask → allow, and also cover MCP and subagent tools.

Ask before tool calls has three values:

Not set – the base configuration’s default applies (without one: no prompts).
On – ask before tool calls.
Off – no prompts: tools run without asking, and deny rules and code approvals no longer apply either.

The Mode decides how prompting works:

Mode	Behaviour
default	Ask for anything not explicitly allowed. A good default to start with
dontAsk	Deny anything not explicitly allowed, without asking. Only useful with explicit allow lists
bypass	Approve automatically unless explicitly denied. Equivalent to the person’s “stop asking me” toggle

CompanyGPT loads custom approval rules only from the base configuration; innFactory sets them up.

Subagent cap

Maximum subagents limits how many subagents can be assigned to an agent in the agent builder. The base configuration sets the ceiling (default 10, at most 50); the value here is capped at that ceiling. Agents that are already saved keep their subagents, even if you lower the value later.

PII and secret filters

This section is display only (librechat.yaml only). It shows which filters for personal data and secrets are set up in the base configuration, with their action (Block or Audit) and the number of patterns. innFactory makes changes.

Per role or group

Here you set four settings for a single role or group. First choose under Pick a role or group. Each value can also stay As default.

Setting	Meaning
Memories	Off switches memories off for this role or group, even if the base configuration enables them for everyone
Retention	How long chats are kept: What is deleted (Temporary chats only or All chats), Delete temporary chats after (hours), Delete all chats after (hours) and Keep agent files
Agent builder	Whether the agent builder is Visible or Hidden in the chat, and Providers for agents (nothing selected means every provider is allowed). The role’s right to create agents still applies
File uploads	Uploads: Allowed or Off, plus Files per message, Size per file (MB) and Size of all files (MB). Empty means as default. The values apply to agents and to models without their own file settings; providers with their own file settings in the base configuration keep those values

A value on the role always wins over a value on the group. All groups share one priority, so avoid different values in several groups of the same person. Providers for agents can therefore only be set per role.

You also find the same settings on the role (companyGPT tab) and on the group (Configuration tab).