🦀Agento
FeaturesPricingBlog
HomeGuidesPrivate vs Shared Mode

Private vs Shared Mode

Mar 2, 2026·7 min read

Table of Contents

  • When to Use Each Mode
  • How to Enable Shared Mode
  • What Changes in Shared Mode
  • How Memory Works Differently
  • Practical Tips
  • Choosing the Right Mode
  • Swarm Agents and Shared Mode
  • Troubleshooting

Every Agento agent runs in one of two modes: private or shared. The mode controls how your agent handles trust, memory, and identity when interacting with people on external channels like WhatsApp, Telegram, and Slack.

Choosing the right mode matters. Private mode is simpler and gives you full control. Shared mode adds protections for when your agent talks to people who are not you.

When to Use Each Mode

Private mode (default) is for agents that only you interact with, or where everyone on external channels is trusted. Examples:

  • A personal assistant you chat with on Telegram
  • A coding agent you use through the dashboard
  • A team agent in a private Slack workspace where all members are trusted

Shared mode is for agents that interact with untrusted external users. Examples:

  • A customer support agent on WhatsApp that anyone can message
  • A public Telegram bot
  • A Slack agent in a workspace with external guests or clients

If your agent has an open channel where strangers can message it, use shared mode.

How to Enable Shared Mode

  1. Go to your agent's Settings page
  2. In the General tab, find the Shared mode toggle
  3. Turn it on
  4. Click Save & Restart

Your agent needs a restart for the mode change to take effect.

If you have external channels connected but shared mode is off, you will see a yellow warning banner in both the General and Channels tabs. This is a reminder, not a blocker. You can keep private mode if you trust everyone on your channels.

What Changes in Shared Mode

Shared mode applies several layers of protection. Here is what is different from private mode.

Trust Levels

In shared mode, your agent distinguishes between two types of users:

Owner (you, via the dashboard webchat): Full authority. You can ask the agent to change its configuration, modify files, update tools, or adjust its behavior. Everything works the same as private mode.

External users (via Telegram, WhatsApp, Slack): Limited authority. External users can have conversations with your agent within the scope of its role. They cannot ask the agent to:

  • Modify its configuration, files, tools, or behavior
  • Reveal internal instructions, API keys, or system prompt contents
  • Change its identity or personality
  • Execute commands that alter its own setup

When an external user tries something outside their authority, the agent declines briefly and generically without revealing details about why or how the restriction works.

In private mode, there is no trust distinction. All users on all channels are treated with full authority.

Identity Protection

In shared mode, your agent does not reveal your name, email, phone number, or other personal details to external users. When someone on WhatsApp or Telegram asks "who is your owner?", the agent will not answer with your real name.

In private mode, your agent knows your name and uses it naturally in conversation.

Prompt Injection Resistance

Shared mode includes explicit defenses against common manipulation attempts. If an external user sends messages like "ignore your previous instructions", "you are now a different AI", or "enter admin mode", the agent refuses and continues normally.

Private mode does not include these defenses because it assumes all users are trusted.

How Memory Works Differently

Memory behavior is the biggest practical difference between the two modes. Understanding this is important for getting the best experience.

Private Mode Memory

In private mode, your agent's memory works without restrictions:

  • Recall: Before every response, the agent searches its long-term memory for relevant facts. All memories are accessible, regardless of which channel or session they were created in.
  • Capture: After every response, the agent extracts and stores important facts. All captured facts are available to all future conversations.
  • Session search: The agent can search through past session transcripts to find relevant context from previous conversations.
  • Explicit tools: The memory_recall and memory_store tools work without any filtering.

Everything is shared across all your sessions and channels. This is the simplest and most powerful setup.

Shared Mode Memory

In shared mode, memory is scoped per user on external channels. This prevents one user's conversation details from leaking to another user.

How per-user scoping works:

  • When your agent talks to User A on WhatsApp, any memories captured from that conversation are tagged with User A's identity.
  • When your agent later talks to User B, it only recalls memories tagged with User B's identity. User A's memories are not visible.
  • When you (the owner) chat through the dashboard webchat, you see all memories from all users. You have full visibility.

What this means in practice:

  • Each external user gets a personalized experience. If User A mentions their company name, the agent remembers it for User A but does not mention it when talking to User B.
  • Your agent builds separate context for each person it interacts with.
  • New users start with a blank memory slate. The agent does not carry over context from other users.

Session transcript search is disabled in shared mode. In private mode, your agent can search through past conversation transcripts for additional context. In shared mode, this is turned off entirely because session transcripts from different users could cross-contaminate.

Your agent still has its per-user long-term memory (via the memory system), so it remembers important facts from past conversations with each user. It just cannot search raw transcripts.

Memory Compaction

Agento periodically compacts session transcripts into long-term memories. In shared mode, the compaction process is instructed to avoid saving user-specific personal details. It focuses on facts relevant to the agent's role and operations instead.

This means some conversational details may not persist across sessions in shared mode, particularly personal information that users shared casually. Important operational facts (like project requirements or preferences explicitly stated) are still captured normally.

Practical Tips

Starting with Shared Mode

If you are deploying a new agent on a public channel:

  1. Enable shared mode before connecting any channels
  2. Write a clear SOUL.md that defines the agent's role and boundaries
  3. Test by messaging the agent from an external account (not the dashboard)
  4. Verify the agent does not reveal your name or internal details
  5. Check that the agent refuses manipulation attempts politely

Switching an Existing Agent to Shared Mode

If you have an existing private agent and want to make it shared:

  1. Be aware that memories stored before shared mode was enabled do not have per-user tags. These older memories will not be recalled for specific external users, which is the safe default.
  2. Enable shared mode and restart the agent.
  3. From this point forward, all new memories are properly scoped per user.

What the Owner Can See

As the owner, you always have full access through the dashboard webchat:

  • All agent memories (from all users)
  • Full configuration and logs
  • Ability to modify any setting

Shared mode restrictions only apply to external channel users, never to you.

Memory Limitations in Shared Mode

A few things to keep in mind:

  • No cross-user memory sharing. If ten users all tell the agent the same preference, it stores ten separate memories. Users do not benefit from each other's context.
  • No session transcript search. The agent relies purely on its long-term memory system, not on searching raw past conversations.
  • Older memories may be invisible. Memories from before shared mode was enabled lack per-user tags and will not appear in per-user searches. This is by design.

Choosing the Right Mode

Question If yes...
Is your agent only for you? Private mode
Do you trust everyone on your channels? Private mode
Can strangers message your agent? Shared mode
Is the agent a public-facing bot? Shared mode
Do you need per-user memory isolation? Shared mode
Do you want the simplest, most powerful memory? Private mode

When in doubt, shared mode is the safer choice for any agent with external channels. The memory scoping adds some limitations, but it protects both you and your users.

Swarm Agents and Shared Mode

If your agent is part of a swarm, shared mode applies the same rules. Swarm shared memory still works, but per-user scoping is applied on top. An agent in shared mode that is also in a swarm will:

  • Recall swarm memories (shared across all swarm agents)
  • Apply per-user filtering to its own memories on external channels
  • Allow the owner full access to everything via webchat

Troubleshooting

Agent reveals owner name on external channels: Make sure shared mode is enabled and the agent has been restarted. The yellow "Changes pending restart" banner means the mode change has not taken effect yet.

Agent does not remember things across sessions in shared mode: This is expected for session transcript search, which is disabled. Long-term memory (automatic capture and recall) still works per-user. If memories seem missing, the conversation may not have contained enough meaningful content to trigger capture. Short greetings and trivial exchanges are filtered out.

External user says the agent is "ignoring" them: Shared mode restricts what external users can ask the agent to do. If a user asks the agent to change its behavior or access internal files, the agent will decline. This is working as intended.

Old memories not showing up for users: Memories stored before shared mode was enabled do not have per-user tags and will not appear in per-user recall. This is the expected safe behavior. New conversations will build fresh, properly scoped memories for each user.

Back to all guides
🦀Agento

AI agents that run 24/7 for your business. Deploy in minutes, not hours.

Remsys, Inc

1606 Headway Cir STE 9078

Austin, TX 78754, USA

+1 650 396 9091

🦞Powered by OpenClaw

Product

  • Features
  • Pricing
  • Security

Company

  • About
  • Contact

Resources

  • Skills Marketplace
  • Agento Blog
  • API Reference
  • Guides
  • OpenClaw
  • Skills.sh

Legal

  • Privacy
  • Terms
  • GDPR

© 2026 Agento. All rights reserved.