Hops
Try Hops
  • Pricing
Log inTry Hops
  • Solutions

    Teams
    • Marketing
    • Sales
    • Professional services
    • Operations
    • Finance
    Company size
    • Startups and SMB
    • Enterprise
  • Pricing
Log inTry Hops

Collaborative AI for your whole team.

Try Hops

Product

  • Solutions
  • Pricing
  • Compare
  • Download
  • Docs

Company

  • About
  • Blog
  • Careers
  • Contact
  • Trust Center

Social

  • LinkedIn
  • X
  • YouTube

© 2026 Hops AI Inc.

Manage cookies
PrivacyTermsAUP
  1. Blog
  2. /
  3. Do AI Agents Remember Team Context Between Conversations?
Technology

Do AI Agents Remember Team Context Between Conversations?

An AI agent starts every chat blank. What carries over between conversations, how team memory differs from personal memory, and a real example from Hops.

November 19, 2026

Do AI Agents Remember Team Context Between Conversations?

Not by themselves. The AI model behind an agent starts every conversation blank, so AI agent memory is whatever the product saves and brings back: chat history, saved memories, written instructions and the team's documents. Whether your teammates benefit depends on where that context is kept. A memory saved in one person's chat helps that person. Context kept in a shared place, like a project doc or a job's written instructions, is there for everyone who works there.

Why AI agents forget

A language model only reads what's in front of it for each request: your message, the conversation so far and anything the product adds. That's the context window. The model's training happened before you ever used it, so it knows nothing about your project.

Start a new chat and the history is empty. Keep one chat going for weeks and the product has to summarize older messages to fit, so details drop out. Every kind of memory an agent has is something the product built around the model to put the right text back in front of it.

Four ways an agent carries context forward

KindWhat it holdsHow long it lastsWho it helps
Conversation historyThe messages in this chatThis chat. Long chats get summarizedWhoever is in the chat
Saved memoryFacts and preferences the product chose to keepUntil someone edits or deletes itUsually one person
InstructionsRules written for an agent or a jobUntil someone edits themEveryone who uses that agent or job
Shared recordsDocs, tables and threads in a team's spaceAs long as the record existsEveryone with access

Most personal assistants lean on the first two, so they get to know you. A team needs the last two, because the context that matters is about the work, like what was decided and what the rules are.

Personal memory vs team memory

Personal memory answers "what does this person like?" Team memory answers "what did we decide, and what's the rule?" That changes what you need from it.

  • Everyone can see it. If the agent works from a fact nobody else can read, nobody else can catch that it's wrong.
  • It stays where it belongs. What one client team tells the agent shouldn't show up in another client's work.
  • Someone keeps it current. A decision changes in a meeting. Unless the record changes too, the agent keeps working from the old one.

In Hops, team context lives in Spaces, one for each team, project or client. Hops builds up shared context from a Space's conversations and docs, and each Space has a Space Context document that Hops reads before it replies, in any thread. Everyone in the Space can see that document. Spaces are scoped: Hops working in one Space doesn't automatically carry that context into another, and in a private Space it only draws context from that Space. Your team can see, edit or remove what Hops remembers. For the bigger picture of a team and an agent sharing one place, read what an AI workspace is.

A real example: how our SEO agent picks up where it left off

Our SEO work runs in a Hops space called #seo-optimization. It holds a board (a table with one row per page), a plan doc called "SEO: start here" and a few scheduled automations. None of those runs remembers the last one. Each starts fresh and picks up from what's written down.

The weekly scorecard carries numbers forward on purpose

Every Monday, Hops posts a scorecard of our SEO goals. Two of its numbers come from checks that only run on the first Monday of the month. Its instructions tell it, on the other Mondays, to carry those numbers forward with the date they were checked.

So on November 16, the line for AI answers that mention Hops read: "Incomplete: the Nov 2 check ran 5 of 20 prompts (0 named Hops). Carried forward." The same post said our Semrush data was unavailable because the API unit balance was zero, "second week". The model doesn't remember last Monday. The run has its instructions and the space, including the board, the plan doc and last week's posts.

The rules live in instructions

Our daily to-do automation keeps the board in line with what actually happened. Its rules are in its written instructions: move a row forward only on evidence, never move one backward, and never move one forward to make the board look current. When James wants a rule changed, he edits the instructions, and the next run follows the new version.

Context from records other people wrote

On November 17, James asked Hops for a brief on a new page about AI workers. The brief quoted a note on the board from November 11 that said not to make that page, and the plan doc's rule from early November that the AI worker and AI teammate searches share one page "unless the results say otherwise". Then it showed that the search results now said otherwise, and left the board row for James to change. Both notes were written days earlier, in shared places Hops could read.

What stays private

When Hops drafted our link outreach emails in James's Gmail, he approved each send in his own chat with Hops. The rest of the team doesn't see those approvals. They see what Hops wrote back to the shared outreach table: each row marked sent, with its next follow-up date.

What goes wrong with AI agent memory

  • It goes stale. The team changed its mind and the saved fact didn't change with it.
  • It ends up in the wrong place. Something said about one client comes up in work for another.
  • Nobody can see it. If you can't read what the agent saved, you can't correct it.
  • It belongs to one person. Context in someone's private chat doesn't help the teammate who covers for them.

Most of these get easier when team context is written down in shared, scoped places that people can read and edit.

Questions to ask before you rely on an agent's memory

  1. Where does it save what it learns: in one person's memory, or somewhere the team can read?
  2. Can you see what it remembers, and change or delete it?
  3. Does context stay inside one project or client?
  4. When two records disagree, does it show both and say which is newer? Our company brain guide covers this.
  5. When it doesn't know, does it say so?

How to give Hops your team's context

  1. Make one Space for each project, team or client. Mixing unrelated work in one Space gives Hops less useful context.
  2. Write the Space Context document: what the Space is for, current goals, decisions already made and the tone you want. Two or three sentences is enough to start. Our docs explain how the Space Context document works.
  3. Put standing rules where Hops reads them every time. Rules for the whole Space go in the Space Context document. Rules for one job go in a skill or in that automation's instructions.
  4. When a thread settles something, ask Hops to "save this as a doc", so the decision lives somewhere besides the chat.
  5. Every few weeks, read the Space Context document and fix what's out of date.

Chat and docs in Hops don't use credits. AI work does, and Hops is $0 per user with 20,000 free credits to start. See pricing.

Try Hops

Make a Space for one project and write three sentences in its Space Context document.

Try Hops

Do AI agents remember previous conversations?

Not on their own. The model starts each conversation blank. An agent remembers only what its product saves and adds back, like chat history, saved memories, instructions and shared documents.

What is AI agent memory?

AI agent memory is the context a product keeps outside the model and brings back in later conversations or runs, so the agent can continue work without being briefed again. It can be personal, like one user's preferences, or shared, like a team's docs and rules.

Can an AI agent share memory across a team?

Yes, if the context is kept somewhere the whole team uses. In Hops, each Space has shared context and a Space Context document that everyone in the Space can see and that Hops reads before it replies.

What's the difference between memory and a context window?

The context window is what the model can read in a single request, and it resets with each new chat. Memory is what the product saves outside the model and adds back into later requests.

Can I see and change what an AI agent remembers?

It depends on the product. In Hops, your team can see, edit or remove what Hops remembers, and the Space Context document is an ordinary doc anyone in the Space can open.

Does an AI agent carry context from one project to another?

It depends on the product. In Hops, Spaces are scoped. Hops working in one Space doesn't automatically carry that context into another, and in a private Space it only draws context from that Space.

Technology

Author

Hops Team