Olyvr TeamMind

Olyvr TeamMind is a memory your team shares. One person’s Olyvr hosts it in a memory server and chooses which data sources go into it. Everyone else adds it as a data source with an address and a token. They can read it but not write to it. Nothing is imported up front. Each question you ask goes to the TeamMind as well as to your own sources. Olyvr joins the memories it gets back with what your workspace holds, and your model writes one answer from both, with citations into each.

You can connect as many TeamMinds as you use (“Sales”, “Platform team”), each under its own name.

How a question is answered

Searches run together. Olyvr asks your workspace index (mail, files, chats, everything you connected) and every TeamMind source you have switched on, all at the same time.
One ranking. The result lists are joined by rank rather than by score, because a keyword score and a similarity score can’t be compared. Your own data and the team’s each keep a minimum share of the list. So when most of the answer is in one of them, you still see what the other knows.
Memories become documents. Each memory that comes back is kept in its source as a document, with its title, where it came from and a link. A TeamMind memory’s title starts with the server’s name (Sales · Jira: …), so a citation shows which team memory it came from. The link goes to the original only when an account you have connected confirms the original exists: the GitHub pull request, issue or file through your GitHub account, the Google Drive file through your Google account, the Jira issue through your Atlassian site. Without that, there is no link, and the memory opens as its cached copy. Olyvr never guesses a link or points to a generic page. The GitHub sign-in reads only your profile, so it can confirm public repositories only. For pull requests, issues and files in a private repository, connect GitHub with a personal access token that can see that repository. These documents are what the answer cites and what you can open under Your data (and from there with Open original). They still answer when the server can’t be reached.
The answer is written here. Your configured model gets both kinds of evidence, marked personal or company. It is told to say where each fact comes from and to point out where your data and the team’s disagree.

If a server doesn’t answer within the source’s wait time, Olyvr answers from your workspace and the memories already cached, and the source’s row shows why.

Join a team’s memory

You need two things from the person who hosts the memory: the server’s address (such as https://203.0.113.9:8013) and a token.

Go to Olyvr Mind → Memory Data Sources → Add data source and choose Olyvr TeamMind.
Enter a Name for this memory (it appears before each of its memories), the Server address, and the Token. Then choose Connect.
Pick a collection and choose Add Olyvr TeamMind. The Also use it for Snapshot cards and other background refreshes setting is on by default. See Snapshot and background work.

To add another team’s memory, choose connect another server in the source’s settings. Each server is a separate connection with its own name and token, and you can add one source for each of its collections.

The certificate. The connection always uses HTTPS. A memory server usually makes its own certificate, so Olyvr keeps that certificate the first time you connect and accepts only that one from then on, the way ssh trusts a host key. If the host has changed the certificate on purpose, choose forget it in the source’s settings and connect again. If they haven’t, don’t reconnect: something may be sitting between you and the server. A server that uses a certificate from a real certificate authority is checked like any website, and Olyvr keeps nothing.

Renaming a server also renames the memories already cached from it. use another token replaces the token when the host gives you a new one.

Host a team’s memory

The memory server runs on your computer. You choose which of your data sources go into it, and other people’s Olyvr reads from it. Only your computer can add documents to it.

Install and start it. In Settings → Memory Server, choose Install and start. The install is a one-time download of several GB into Olyvr’s data folder. The server uses the model you set in Models & Account → Model access to find memories, so set one first. Without a model, people can search its documents only by keyword. To have the server start with Olyvr, select Start the memory server when Olyvr opens.
Choose what it holds. Under What the memory holds, select the data sources to share and choose Save and sync. Every 15 minutes, Olyvr adds new and changed documents and removes deleted ones. If you clear a data source, its documents are removed from the memory. Sync now runs a sync right away. Anyone with a token can read every document in the memory, so put in only what your team may see. A TeamMind source can’t be selected.
Share it. Under Share with others, select Let others read this memory over the network and choose a port (8001 by default). People on your network can now connect. To accept connections from outside it, select Let people connect from the internet: Olyvr asks your router to open the port, and olyvr.ai then tests whether it can reach your computer. If your router can’t open the port, or there is another network between you and the internet (common with mobile and office connections), the pane says so. In that case, use a VPN or ask whoever runs that network.
Give people access. Under Access tokens, create a token and copy it. One token can serve the whole team. Send each person the address shown under How others connect and the token. They follow Join a team’s memory.

To use the memory yourself, choose Use it in this Olyvr.

Tokens. There are two slots, Primary and Secondary. Each token is valid for the number of days you set (365 by default, up to 3,650). To replace a token without cutting anyone off, share the secondary token first, then replace the primary one. Replacing or deleting a token stops it working immediately. The pane shows when each token expires and when it was last used.

Your own address and certificate. If people reach your computer at an address that isn’t listed, such as a company domain name or a cloud server, enter it under Your own address and certificate (optional). If you have a certificate for that address (from your IT team or from Let’s Encrypt), also give its two PEM files. Their Olyvr then checks the connection the way it checks a website.

Limits. Advanced: limits sets how much work other people can give your computer: how many requests it answers at once (4 by default), how many can wait in line (16), and how many seconds to wait for the memory server (120). The defaults suit a laptop. Changes apply immediately.

What is sent

When you ask a question, only the question (or the agent’s search query) leaves your machine, together with the collection name and the source’s settings. Nothing you imported is sent to a TeamMind, and nothing is written there: TeamMind tokens can only read.

As the host, you choose the data sources that go into your memory. Everyone with a valid token can read their documents. See Data and privacy.

Snapshot and background work

Snapshot cards use TeamMind sources the same way a chat does, so a card’s answer draws on team memory too. Cards refresh on a timer. If that sends more traffic to the server than you want, turn off Also use it for Snapshot cards and other background refreshes in the source’s settings. The source then stays out of cards entirely, including the memories cached from it.

Disconnect

forget it, in the source’s settings, deletes the token from your machine, along with the certificate Olyvr kept for that server. Sources stay, with the memories cached in them, until you connect again.
Remove on a source deletes the memories cached in it.
A host who replaces or deletes a token cuts off everyone using it.