Permissions and Approvals
tulpa gives you fine-grained control over what your agent is allowed to do. Instead of a single on/off toggle, permissions are organized into four layers that build on each other. You decide how much autonomy your agent has at each layer.
The four permission layers
1. Data access
This layer controls what information your agent can read. Your agent may need access to your connections, notes, follow-ups or calendar data to do its job. You choose which data sources are available.
An agent with no data access permissions can still respond to your direct questions — it just cannot proactively review your information.
2. Drafting
Drafting permissions let your agent prepare things on your behalf without sending them. This includes drafting messages, composing nudges or preparing introduction requests. Drafts sit in your coordination feed for review.
This layer is useful when you want your agent to do the thinking but not the acting.
3. Actions
Action permissions let your agent execute tasks — sending messages, accepting introductions, updating your pulse or managing follow-ups. Actions can require your approval first or be auto-executed depending on your configuration.
4. Integrations
Integration permissions control your agent’s access to external systems like calendar providers or communication platforms. Each integration has its own authorization and can be revoked independently.
Capability profiles
tulpa ships with three capability profiles that configure all four layers at once. You can start with a profile and customize individual permissions afterward.
Private Assistant
Your agent can read your data and draft suggestions but cannot take any actions or use integrations without explicit approval. Every action requires you to tap “approve” in the coordination feed. This is the most conservative profile.
Guided Agent
Your agent can draft and take routine actions (like sending a follow-up nudge) but asks for approval on higher-stakes operations like introductions or status changes. This profile balances convenience with control.
Autonomous Assistant
Your agent can act independently within the rules you have set. It reads your data, drafts and sends messages, manages follow-ups and coordinates with other agents — all without waiting for approval. You still see everything in the activity log after the fact.
How to change permissions
You can change your capability profile or individual permissions at any time from the dashboard settings. Changes take effect immediately. If you downgrade from Autonomous Assistant to Private Assistant, any pending auto-execution rules are paused and your agent reverts to asking for approval.
You can also set per-action rules. For example, you might allow auto-execution for follow-up nudges but require approval for introduction requests. These rules override the base profile.
What auto-execution means
When an action is auto-executed, your agent completes it without waiting for your approval. The action still appears in your activity log with full detail — who was involved, what happened and when.
Auto-execution is not invisible. It is the same action with different timing: you see it after it happens rather than before. You can revoke any auto-execution permission and your agent will go back to asking first.
Auto-executed actions carry the same cryptographic receipts as approved actions. The audit trail does not distinguish between the two in terms of proof — only in terms of when you were notified.