Understanding Agent Memory

Pinchy agents have a memory system that persists knowledge across conversations. When you talk to an agent, it can remember important information and use it in future conversations — giving better, more contextual responses over time.

Personal vs. shared agent memory

Pinchy has two types of agents, and they handle memory differently:

Personal agents (Smithers)

Every user gets their own personal agent — called Smithers by default. This agent is completely private:

• Only you can talk to your Smithers.
• Memory is isolated. Nothing you discuss leaks to other users.
• Treat it like your personal notebook.

Shared agents

Shared agents are team resources. When any team member talks to a shared agent, the agent may remember that information and use it when helping other team members. This is by design — it builds team knowledge.

Think of a shared agent like a shared Slack channel: everyone with access can see what’s been discussed.

What this means for your team

• Shared agents are great for: team knowledge bases, project assistants, domain experts, onboarding new team members.
• Personal agents are the right choice for: tasks involving sensitive or private information.
• Rule of thumb: Don’t share anything with a shared agent that you wouldn’t share with your entire team.

How Pinchy communicates this to users

You don’t need to explain the memory model to your team — Pinchy handles it in the product:

• When users open a shared agent’s chat, they see a “Shared” badge in the header and a notice explaining that conversations contribute to shared team knowledge.
• When users open their personal agent, they see a “Private” badge confirming their conversations are isolated.

What’s coming next

• Per-agent memory toggle — ability to disable memory for specific agents.
• Memory inspection — tools for admins to see what agents have memorized.
• Memory deletion — right to erasure for compliance requirements.