Agently

Core Concepts

Permissions

Agently gives you precise control over what an agent is allowed to do. Three permission modes — Explore, Ask to Edit, and Auto — cover every use case from safe exploration to hands-free automation.

The Three Modes

🔍

Explore

Explore

Read-only mode. The agent can access sources and read data, but cannot create, modify, or delete anything. Safe for exploratory conversations where you just want answers.

Can do

  • ✓Read files from local folders
  • ✓Query MCP server resources
  • ✓Call read-only REST API endpoints
  • ✓Search and retrieve data

Cannot do

  • ✗Write or delete local files
  • ✗Create/update GitHub issues
  • ✗Send API POST/PUT/DELETE requests
  • ✗Trigger webhooks
✋

Ask to Edit

Ask to Edit

The agent proposes write actions, but waits for your approval before executing each one. You see exactly what will happen and can approve, deny, or modify the action.

Can do

  • ✓Everything in Explore
  • ✓Propose file writes (with your approval)
  • ✓Propose API mutations (with your approval)
  • ✓Run multi-step workflows with checkpoints

Cannot do

  • ✗Execute write actions without approval
⚡

Auto

Auto

Full read and write access. The agent executes actions immediately without asking. Best for trusted, well-defined automation tasks where interruptions are costly.

Can do

  • ✓Everything in Explore and Ask to Edit
  • ✓Execute write actions immediately
  • ✓Run fully automated multi-step workflows

Setting Permissions

The permission mode is set per conversation. You can change it at any time during a conversation from the toolbar at the top of the chat window.

You can also set a default permission for each workspace in Workspace Settings → Permissions. New conversations inherit the workspace default.

Recommendation: Start with Ask to Edit for any agent connected to production systems. Switch to Auto only after you've verified the workflow is correct and the consequences of mistakes are low.

Source-Level Overrides

Individual sources can have a permission ceiling — a maximum permission level that applies regardless of the conversation setting. For example, you can mark your production database source as "read-only forever" so it can never be mutated even if the conversation is set to Auto.

Configure source-level overrides in Settings → Sources → [source name] → Max Permission.

Approval Dialog

In Ask to Edit mode, write actions show an approval dialog with:

  • The tool name and which source it belongs to
  • The exact parameters that will be sent (JSON)
  • A diff view for file edits (showing what will change)
  • Approve / Deny / Edit buttons

If you click Edit, you can modify the parameters before the action executes. The agent continues from your edited version.

Keyboard Shortcuts (Approval Dialog)

EnterApprove action
EscapeDeny action
EOpen edit mode
AApprove all remaining (caution)