Mattermost

• “Which teams and channels am I in?”
• “Summarize what was written this week in the Case discussion channel.”
• “What was the last IT-Support channel about?”
• “Who last wrote something in the Neubau channel?”

Channels in Mattermost always hang on a team, so the assistant's first step is the team list. After that it can list a team's channels, look up a channel, read its history and resolve a user ID into a name.

1. 01Ask your administration to enable Personal Access Tokens under System Console, Integrations, Integration Management.
2. 02For a non-admin account, creating tokens must also be allowed under System Console, User Management, Users, Manage Roles.
3. 03Sign in with the desired account, open Profile, Security, Personal Access Tokens and choose Create Token.
4. 04Give a description, save and copy the token immediately. It is shown only once.

1. 01In anymize open Apps, then Connectors, and choose Mattermost.
2. 02Enter your instance address, for example https://chat.your-company.com, and paste the personal access token from the section above.
3. 03Set which areas the assistant may use: account, channels, messages.

No. Mattermost has no read-only rights for a token: a personal access token always carries all rights of your account, including write. That anymize still does not write is our commitment, and we keep it where it can be checked.

The connector only sends read requests, and no tool name may carry a write verb. A test records that, so Mattermost sits on the list of connections with the “Read only” mark. If a write tool were ever added, it would have to leave that list.

Mattermost is open source and runs on a server you operate yourselves. For authorities and law firms that is often exactly why it is the choice against the big chat services. anymize does not change that: the messages stay on your server, they are only read on request, and what goes to an AI model goes anonymized.

• A full-text search across all messages does not exist at this stage. It works differently in Mattermost depending on the edition, and a tool that does something different for every customer is not a good commitment. Ask for a channel and a time range instead.
• Files from channels are not read.
• A history arrives in blocks of at most 50 messages.

Chat messages are the place where names drop most casually. People's names, email addresses, phone numbers, file references and customer names in the message text are replaced by placeholders before an AI model summarises the history.