Skip to content

Assistants become Agents

The previous Assistants have become Agents in 9brains. This is more than a rename: agents are persistent digital colleagues with their own identity, their own memory and their own workspace. They can still be addressed via chat as before, but now also run autonomously (triggered on a schedule or via webhook from outside), keep learning between sessions, and be equipped with their own budget for clearly scoped tasks.

Your existing assistants have been carried over automatically, configuration, shared access and attached files remain unchanged. They are now simply called “Agent” and can do everything they did before, plus much more.

What this opens up for you

Section titled “What this opens up for you”

The agent platform makes use cases possible that previously required a separate workflow platform or were simply not feasible:

  • An inbox colleague that scans new emails every hour, separates the urgent from the unimportant and delivers a morning briefing with reply suggestions
  • An SEO watchdog that checks your website weekly for broken links, missing meta tags and performance issues and drops a report into your OneDrive
  • A meeting scribe that turns raw notes into structured minutes, extracts action items and notifies the people involved by email
  • A CRM agent that activates automatically as soon as a new lead lands in HubSpot: enrich the data, create follow-up tasks, notify you on Teams
  • A trip log helper that takes trips via chat throughout the month and automatically produces the export at the end of the month
  • A social media writer with your tone of voice, drafting LinkedIn posts and blog drafts from bullet-point input

You describe to the agent in natural language what it should do, the required triggers, integrations and learning steps it either sets up itself or you configure them via the configuration form. You do not need a flow editor or “if-this-then-that” drag and drop. The agent is the workflow.

Agents keep learning

Section titled “Agents keep learning”

Every agent has its own AGENTS.md file, its central knowledge file. This holds the role, behavior rules and everything the agent has learned about its area of work. The key point: the agent can edit this file itself. If you say in chat “Note that our standard format for trip-log exports is CSV with semicolons”, it adopts that into its own knowledge file, and from then on into every future autonomous run. You can also edit the file yourself at any time and see the full version history: who changed what when, with diff view and restore option.

This replaces the frustrating “the chatbot always forgets everything”: an agent builds up domain knowledge about its area of responsibility over weeks and months, targeted to its scope, not in a global store.

Autonomous runs, with explicit protection

Section titled “Autonomous runs, with explicit protection”

When an agent runs autonomously, it has no human counterpart. To keep this safe, autonomous runs follow a consent model: which of your personal integrations (for example your Microsoft 365 account) the agent may use in cron or webhook runs must be explicitly approved by you as the owner, in advance and individually. As soon as you share an agent with colleagues whose autonomous runs would use your personal tokens, the platform actively prompts you to confirm again.

On the cost side, every agent has its own monthly spend cap as a safety net, preset to 10 EUR per month, and you can adjust the value up or down per agent at any time. When the limit is reached, the agent pauses automatically and you are notified by email. After three consecutive failed runs the agent also pauses itself, so that a configuration error does not silently consume quota.

A homebase per agent

Section titled “A homebase per agent”

Every agent has a new three-column view: navigation on the left, the running chat or configuration in the middle, and on the right a context panel with status, budget usage, enabled integrations, the next scheduled execution (for cron agents) and the full run history in chronological order, no matter whether chat run, cron run or webhook run. You see at a glance: what has this agent done in the last 24 hours? Which tools did it call? What did it cost?

For every single tool call, administrators can trace back to the triggering run: which trigger? Which payload? Which model? A real answer to “Why did the agent send this email?” instead of an audit log of loose individual actions.

Threads instead of endless chats

Section titled “Threads instead of endless chats”

You now run multiple parallel threads with an agent, like in Teams or Slack with a colleague. One thread per project, per customer, per task. The agent knows its context per thread, but its memory (AGENTS.md) and its identity stay the same across all threads.

Availability

Section titled “Availability”

The agent platform is part of the Business and Max license. Existing assistants have already been migrated, all previous configurations are preserved. You will find your agents under the new Agents entry in the sidebar.

More details

Section titled “More details”