Introducing Memory: Agents that learn & improve over time
We just rolled out memory for your agents. Now an agent can learn from relevant work across all its threads, take on more work simultaneously from your whole team, and become more useful over time.
Existing agents won’t use memory until you turn it on. New agents will use it automatically.
Memory covers everything an agent works with in a thread: the conversation itself, plus files, images, and attachments. Whatever it picks up in one thread, it can now remember and use in every other.
One agent, many focused threads
Each Tasklet agent is a persistent AI personality with its own instructions, knowledge, connections, automations – and now, memory. Each thread is a focused conversation with that agent.
Until now, every thread started fresh: your agent brought its instructions and knowledge, but it had no memory of what happened in any other thread. Memory changes that. Your threads stay separate and focused, but your agent learns from all of them, and it can even search its earlier conversations when it needs more context.
One way to think of it: starting a new thread with your agent is like starting a new email with an employee. If you’re working on an ongoing project where the past conversation has useful context, you’d keep replying in the same thread. If you have a new question or a new project, you’d naturally start a fresh one. Either way, that employee would have context from all your other conversations. Same idea here.
More work can happen at the same time
Focused threads also make it easier for teammates to put the same agent to work simultaneously. Each task can move forward in its own thread, without teammates waiting on one another or manually relaying context between isolated conversations.
Imagine your team shares a customer operations agent. One teammate asks it to turn a new signed agreement into an onboarding plan. Another asks it to prepare a kickoff brief. A third asks it to add the customer to Stripe and send their first invoice. The agent handles all of it at once, while maintaining its automations and existing work in other threads.
+9
5.6 Sol Medium
No more funneling every request through one long conversation, or splitting related work across separate agents. It’s now easier for multiple streams of work to move through one shared agent at once.
Each thread makes the agent more useful
Every task gives your agent more experience to draw on. As your team works with an agent, memory helps it carry preferences, decisions, customer context, and recurring ways of working into future threads. A new conversation does not start from zero. It starts with an agent that already understands how your team operates.
That improvement compounds. One thread helps the next, and every teammate benefits from work they didn’t do themselves. The agent gets better at the work it owns and more valuable to the team the longer you work together.
Getting started with Memory
New agents use memory automatically. Existing agents can turn it on.
Memory works asynchronously in the background, so there’s nothing to manage. As your team uses an agent, it updates what it remembers on its own.
To turn on memory for one of your existing agents, open the agent’s sidebar and click More > Advanced, then find Agent-wide memory. Click Turn on and then Turn on memory. Memory is a one-way door for that agent: once it’s on, it stays on.
What can I do for you this afternoon?
New thread or new agent?
Start a new thread for a new task. Most of the time, this is the right choice. If the work is close to what your agent already handles, like another customer, another report, or another request from a teammate, just start a fresh thread. The agent brings everything it has learned, plus all of its connections and knowledge. The thread keeps the task focused.
Create a new agent for a new area of work. If the work calls for a different role, different knowledge, and access to different tools, it may be time to onboard a new agent.
With memory, one agent can take on more work, support more teammates, and become more useful over time.
Have questions? Drop me a line at tyler@tasklet.ai