AgentSuite Adds Agent Memory, Document Templates and Code Execution

Here’s what changed, and what it means when you’re building an agent.

1. Smarter agent memory

Agent memory is what an AI agent keeps about your preferences and instructions, so it carries them from one conversation into the next.

Share how you want things done and the agent remembers, rather than starting fresh each time you open a chat. You can set a memory for one agent, for a category such as Food or Finance, or for every agent you use, and the Memory tab is where you view, edit, organise or delete any of them.

For example, a creator could tell an agent once how a relative likes their weekly plan laid out, and have every later conversation already know it.

Agent memory is stored per person and per agent, creators cannot read it.

2. Templates from your docs

Document templates are reusable layouts that B3MO builds for your agent from a document you already have.

Give B3MO an existing document and it turns it into a template, switches document creation on for the agent, and produces a completed sample so you can check the result before anyone else sees it. From then on, the people who use the agent can ask for the same document in plain language and get it back ready-filled.

For example, a creator could hand over the form they fill in for a community group every month and have the agent produce it on request, in the same shape every time.

You decide which document the template comes from, and the sample is there to review first.

3. Work with files

A file space is a private store that each person who uses your agent gets, shared with the agent and switched on in Configure.

Chat attachments and documents the agent creates are saved there automatically, and the agent can go back to them whenever a later task needs them, so the same file never has to be uploaded twice. Files stay separate per person, are encrypted at rest, and are never written on-chain.

For example, a creator could set up an agent that keeps a parent’s paperwork in one place and pulls the right document back out when an appointment comes round.

Each file space belongs to the person using the agent.

4. Code execution built in

Code execution lets an agent run Python, JavaScript and TypeScript to calculate, analyse and work through problems that plain text answers cannot reach.

An agent can read a spreadsheet, run the numbers with familiar data libraries and create charts from the result. Execution is isolated from the internet and restricted to that person’s own folder.

For example, a creator could build an agent that takes a monthly expenses sheet from a family member and returns the totals with a chart.

5. Knows your material

Agent knowledge is the reference material you upload so an agent answers from your expertise rather than general information.

Search across that material works on meaning rather than matching keywords, so a question phrased differently still finds the right passage. Long documents are organised by their headings, which makes retrieval more accurate, and any file without readable text is flagged while you upload it.

For example, a creator could upload the notes they have built up on a subject they are often asked about, and let the agent answer from them.

You choose what goes into the agent’s knowledge, and what to leave out.

What this means for you

Together, these updates address two parts of working with an agent: how much you have to repeat, and how much of your own material the agent can draw on.

Agent memory keeps your preferences across conversations. Document templates let B3MO turn a document you already use into one the agent can produce on request. File spaces save chat attachments and generated documents for later. Code execution runs Python, JavaScript and TypeScript inside an isolated folder. Agent knowledge finds the right passage in your uploaded material by meaning.

Tell your agent once and let it work with your material, with AgentSuite at VeWorld.ai.