Skip to main content
Owners and admins can open Settings, then Access control for the selected team. Members cannot view this page. The page has three tabs: The Confirmations tab shows a badge with the number of pending, unexpired requests when that count is greater than zero.

Team roles

Every team member has one of three fixed team roles: owner, admin, or member. You cannot rename, customize, or add team roles. The role controls what a person can do directly in Felan, and it caps what agents, automations, and API keys can do on that person’s behalf. Change a member’s role from Settings. Each role includes everything the roles below it can do. Admin and owner can also do the following: Members can still connect Microsoft Teams and select the Asana projects or GitLab project for an integration that an admin has connected.

How team roles cap agents

When an agent, automation, or API key acts, Felan identifies a requester and applies that person’s team role as a ceiling. An agent role can only restrict access further. It cannot grant an action the requester’s role does not allow.
  • For a session someone starts, the requester is that person.
  • For an automation, the requester is the person who last edited it, or its creator if no edit is recorded.
  • For an API key, the requester is the person who created the key.
If Felan cannot identify the requester, or that person has left the team, delegated calls are capped at member permissions. For example, an automation last edited by an admin can create a subagent through an agent. After that admin is removed from the team, the same automation can no longer do so.

Audit trail

Review access decisions and their outcomes. Filter by principal, action, resource, session, provider, decision, or time range. Text filters require an exact match. Dates use your local time. Open an event’s details for its principal chain, reason, and correlation data. Use the page controls to browse older events.

Confirmations

Review pending requests across the team and the 50 most recent resolved requests. Read the stored action summary before choosing Confirm or Decline. Requests expire after 5 minutes. Expired requests cannot be confirmed. Refresh to check for new requests or execution results. For provider actions that require confirmation, the gateway holds the request until an authorized person approves it. Approval lets the gateway send that request with the provider credential. Declining or expiry prevents the request from reaching the provider. Approval checks current policy again. A new deny prevents execution. An approved request may still fail at the provider; check its execution result. You may approve your own request if your role permits it. Confirm is supported for platform writes under /api/v1, gateway provider requests, and MCP tool calls. Platform reads cannot wait for confirmation. A Confirm rule on a platform read returns 403 with reason confirmation_unsupported and records a deny audit event. Confirmation approval, decline, and cancellation do not start another confirmation.

Who can confirm

Only a person who started the session, in the web UI or from Slack or Microsoft Teams, can confirm their own request. Requests from automations and API keys have no personal confirmer, so only an owner or admin can confirm them, even though the automation’s last editor or the key’s creator still caps what the request may do.

Where requests appear

Every pending request appears in the Confirmations tab. If the root session is bound to a Slack or Microsoft Teams conversation, Felan also posts a confirmation card in that thread, and updates it when the request is confirmed, declined, expired, or cancelled. The person who clicks the card must match a team member, as described in Slack and Teams senders. Teams replies “You can’t confirm this” to a sender it cannot match.

Roles

The built-in roles are: You can also create custom roles with their own rules.

Default confirmations

Default access asks for confirmation before an agent, automation, or API key does any of these: Confirmation applies only to delegated actions. When you do one of these in the web UI yourself, Felan checks your team role and does not ask you to confirm. Confirmations are also subject to the team role cap: an action the requester’s role does not allow is denied rather than held for confirmation. Custom role rules can add Confirm for other operations. An explicit Allow rule can remove a default confirmation.

Default role for root sessions

In the Roles tab, choose a built-in or custom role under Default role for root sessions. The initial setting is Default access. The selection saves immediately. This default applies to sessions people start and to automations and API keys without their own role. An automation or API key role replaces the root default for its sessions, even if you have changed the root default. For example, with a Read-only root default, an automation set to Default access can start a root session that permits writes, subject to the requester’s permissions. For automations and API keys, Team default means they have no role of their own. Their sessions and direct API requests use the current root default unless another automation or API key in the chain has its own role.

How roles combine

Among the roles that apply, the most restrictive wins. Automation and API key roles in the chain combine with each other and with subagent roles. A subagent cannot grant access that its starting session denies. The requester’s role still caps the result, so an agent role cannot grant the requester additional access. Unknown or removed requesters cap delegated calls at member permissions. The gateway caches decisions for up to 60 seconds, so policy changes may not apply immediately.

Custom roles and rules

In Roles, create a role with a name, optional description, and built-in base role. Each rule is a scope, an operation, and an Allow, Deny, or Confirm effect. Choose the scope first. It decides which operations the rule can use: The operation list then shows only what the scope can perform. Jira Cloud offers issue creation, updates, deletion, and comments, because Felan classifies its API in detail. A provider without that detail, such as Asana, offers only its reads and writes. Every scope offers its reads and writes as a group. Felan platform, Any provider, and named providers also offer prefixes such as external.tcms.*. Everything offers only the read and write groups. Any provider includes MCP servers. MCP tool calls are writes, so an Any provider rule that denies all writes also blocks every MCP tool. An MCP rule with a tool pattern takes priority over every rule without one. Otherwise, a rule for a named provider takes priority over one for Felan platform or Any provider, which take priority over Everything. Within the same scope, exact operations take priority over prefixes, longer prefixes take priority over shorter ones, and prefixes take priority over read or write groups. Equally specific rules use the most restrictive effect: Deny, then Confirm, then Allow. If no rule matches, the base role applies. An explicit Allow rule can remove a destructive operation’s default confirmation requirement. It does not override a stricter role elsewhere in the chain or the requester’s permissions. Edit a role to change its rules. Deleting a custom role removes its bindings, so affected principals return to their inherited or default role. This may allow more access.

MCP server and tool rules

Choose a connected MCP server as the rule’s scope. An MCP server has one operation, calling a tool, so the rule asks only for a tool name or * pattern, such as delete_*. Tool matching is case-insensitive. Leave the pattern blank to cover every tool on that server. In the API, the rule’s provider ID is mcp-<connectorId> and its operation is external.mcp.tool.call. All MCP tool calls count as writes, even tools that only read data. Read-only roles block them. Confirm holds the call until approval; denial, decline, or expiry prevents forwarding. For a batch, every tool must pass authorization and any required confirmations before the batch is sent. If one fails, none of the batch is sent.

Automation roles

Open Automations and create or edit an automation. Owners and admins can select Team default, a built-in role, or a custom role in the editor. Members cannot view or change the role binding. The role binding saves after the automation itself saves. If saving the binding fails, the editor reports the error so you can retry. An automation’s own role applies even when a person starts it with Run now. It replaces the root default, but other applicable roles and the requester’s permissions still limit the run. Anyone who can edit or run an automation, or send a message in a session it started, can use the automation’s role. Members can do all three. Give an automation a less restrictive role than the root default only if every member may use that access through it. The same applies to sessions started by an API key.

Subagent roles

Open Agent, choose the Subagents tab, and create or edit a subagent. Owners and admins can select Inherit, a built-in role, or a custom role in the Role field. Choose Save subagent to save the selection. Built-in subagents start with these platform defaults: Owners and admins can change these roles, including setting Explore or Review to Inherit. New custom subagents start with Inherit. Read-only blocks every MCP tool call and most POST requests, so Explore and Review cannot call MCP tools by default. To allow specific tools, bind them to a custom role based on Read-only with Allow rules for those tools, or set them to Inherit. Inherit uses the role of the session that started the subagent. For a nested subagent, that is its parent session, not necessarily the root-session default. An explicit subagent role can add restrictions but cannot remove restrictions that already apply to the starting session. Subagent roles apply to platform API calls and to gateway requests that carry the subagent’s credential. A subagent shares the workspace VM with its root session, so a workspace command that omits its credential is checked against the root session’s roles instead. Deleting a custom subagent also removes its role binding. Reset to default restores a built-in subagent’s platform definition and default role.

API key roles

Open Settings, then API keys. Owners and admins can choose Team default, a built-in role, or a custom role when creating a key. To change an existing key’s role, select a role beside that key. Changes to an existing key save immediately. A key’s own role replaces the root default for direct API requests and for sessions it starts. With Team default, the key uses the root default.

Slack and Teams senders

Felan accepts requests from Slack and Microsoft Teams only when it can match the sender to a member of the team. Felan acts for that member, so their team role applies to what the agent can do. When Felan cannot match a sender, it replies that it only accepts requests from team members and asks them to get invited or to use a matching email. For Slack, Felan reads the sender’s profile from the installed workspace or Enterprise Grid organization. It accepts the sender only if all of these are true:
  • The account belongs to that workspace or organization.
  • The account is not deleted and is not a bot.
  • The Slack email address is confirmed.
  • The account is not a guest or an external Slack Connect user.
  • The Slack email matches the email of a Felan team member.
For Microsoft Teams, Felan accepts the sender only if the sender belongs to the tenant where you installed the app. When Teams provides the sender’s Microsoft Entra ID (Azure AD) identity, Felan matches it to a team member who signed in to Felan with Microsoft in that tenant. This match is refused for teams with more than 50 members. Otherwise Felan matches the sender’s email to a team member’s email. Non-members also cannot answer agent questions in a channel. Felan replies that only team members can answer and suggests asking a team admin for an invitation. If a member is not recognized, check that their Slack or Teams email matches their Felan account email, or invite them to the team under Settings, then Members. Felan caches the sender’s Slack or Teams profile lookup for 5 minutes, so a change to a Slack or Teams email may take that long to take effect.

Repository and provider credentials

Gateway enforcement requires v10 or later workspace VMs with the interceptor. In these workspaces, provider requests and repository operations made by git, gh, and glab go through the gateway. The workspace receives credential placeholders. The gateway checks access and replaces placeholders with real credentials before sending allowed requests. Repository reads, pushes, and ref deletions have separate operations. There is no fallback to real credentials for older snapshots. Replace older workspace VMs before relying on gateway enforcement. Provider credentials stay in the agent runtime and never enter the workspace VM. The runtime opens a channel into the workspace for provider requests. The workspace stores no runtime address or token that grants access to the runtime. Background processes can reach providers only while that channel is open. Felan closes it after two minutes with no running agent command and no open provider request. It closes at the latest 12 minutes after the last agent command ends, even if a background request is still open. Felan cuts those requests at that deadline. Later provider calls receive 503 until an agent command can reopen the channel. Background traffic cannot keep the workspace running indefinitely. If the channel fails during a running command, Felan retries automatically. The delay doubles from 5 seconds up to 60 seconds after consecutive failures. After a failure, a command waits for a reconnect only when the delay has elapsed and no other attempt is running. During the delay or an active retry, commands run without waiting. A failed reconnect does not prevent the command from running. Interrupted provider requests are not replayed. Each workspace supports at most 100 concurrent provider requests through the channel. Manage human membership and API keys in the other Settings screens.