What it means for an agent to stay
Most AI "agents" are a prompt that disappears when the chat closes. A persistent advisor keeps its role, its lessons and its watches, and comes back when something changes.
"Agent" has become one of the most stretched words in software. Often it means a prompt with a job title: "You are a financial analyst." Close the chat and the analyst is gone, along with everything it learned about your company in the last hour.
We think an agent worth the name has to stay.
What staying means
An advisor in Manwe is a persistent identity you return to. Each one keeps:
- A role and instructions. "Model dilution before every term change" stays true across every conversation, not just the one where you said it.
- A model of its own. One advisor can run on a local model and another on a frontier model. You can change it any time.
- A learning history. What it has learned, and from where, is recorded and inspectable.
- Its watches. Questions it keeps checking on your behalf.
You can talk to an advisor directly, call several into one conversation, assemble a team that divides the work, or ask them for independent second opinions.
Learning between conversations
Here's the part that changes how an advisor feels over time. With background learning enabled, advisors can research their field while Manwe is open, idle and online. A finance advisor can read up on current option pool norms. A market advisor can look into a competitor's announcement.
That learning runs within allowances you control. By default there's a shared allowance of one million tokens per day, and each advisor defaults to 100,000, with up to 20% extra to finish a task already in progress. These are usage limits, not included credit. Your provider's subscription limits and API charges still apply.
Coming back when something changes
An advisor that stays can also keep watch. A watch is a recurring question or condition, like "tell me when registration opens for this venue" or "flag any change to the term sheet redlines". A check-in is a follow-up tied to a conversation: "remind me next Friday to review the plan".
This is the difference between a tool you have to remember to use and a colleague who remembers for you.
What an advisor is not
We're careful about one thing. An advisor is an AI perspective, not a person with verified professional credentials. Its usefulness comes from its working approach, the model doing the work, and the context and lessons it has kept. It can be wrong. It can disagree with another advisor, which is often useful.
Persistence makes an advisor more useful, but it doesn't make it infallible. It means you can see what it knows, correct it once, and have the correction stick.