The global agent is the heart of Cobrief 2. It's a single conversation partner, your digital bid manager – that can work across all of Cobrief: search your knowledge base, look up tenders, query market insight, and jump into a specific bid to write or edit.
How to open the agent
You'll find the agent in the menu, or you can press ⌘J (Mac) or Ctrl+J (Windows) anywhere in Cobrief. The same shortcut closes the panel again.
In the panel you can start a new session or jump back into an earlier thread. Conversation history is stored per workspace, so you'll find your previous threads the next time you log in.
You can also give a thread a clear name: hover over the conversation in the history and choose "Rename conversation".
What the agent can do
The agent has access to all your content in Cobrief and picks the tools it needs on its own:
Search your knowledge base: find text, references, and company information across everything you've uploaded.
Look up tenders: search announced competitions and your own, and summarise the most important parts.
Suggest search profile changes: help administrators adjust CPV codes, keywords, buyer countries, and other filters.
Query market insight: answer questions about who wins what, historical agreements, and competitors.
Fetch CVs and reference projects from Flowcase, if you're connected.
Search SharePoint and OneDrive when you have connected Microsoft 365, using your own permissions. See Connect Cobrief to SharePoint and OneDrive.
Start a new bid: set up the bid room with a draft based on the competition basis.
Jump into an existing bid: read, edit, and create documents directly in the bid room.
In practice this means you can ask questions ranging from "what did we win last time we delivered to Helse Sør-Øst?" to "start a new bid on the competition I opened yesterday and write a draft cover letter".
Examples of what you can ask for
"Find all open tenders for IT consulting services with a deadline this month."
"What does the knowledge base say about our GDPR work?"
"Start a new bid on the competition Operation of core system 2026 and draft a cover letter."
"Open the bid Operation of core system 2026. Read the award criteria and improve the cover letter so the competence aspect (40%) comes through more clearly. Use the CVs of Kari and Ola."
"Check whether the cover letter, solution description, and competence description are consistent – flag conflicting numbers and claims."
"Create a project plan with mobilisation, start-up, and operations phases based on the contract period in the competition basis."
"Fill out the self-declaration form with information from the knowledge base."
"Find out why we're getting too many irrelevant cleaning tenders, and suggest an adjustment to the search profile."
"Make a copy of the price schedule template, so we can try an alternative answer without changing the original."
"Find the latest version of the quality manual in SharePoint and summarise the requirements that matter for this bid."
Get smarter about the market
The agent has access to a large database of public procurements, buyers, suppliers, and awards – and can combine the numbers into answers you previously had to gather from several sources. Some things you can ask:
"Who has won framework agreements for cleaning at municipalities in Innlandet over the past three years, and what are the contracts worth?"
"Which award criteria does Helse Sør-Øst weight most heavily on IT tenders – price, quality, or competence?"
"Which competitors have grown fastest in the public sector over the past year?"
"How many tenders have been announced for environmental consulting services per year over the past five years?"
"Which buyers have the most awards within our CPV area, and what is the average contract value?"
"Look up the key figures for Nordic Tech AS before I add them as a subcontractor."
You don't need to know what a CPV code is or how the data is organised – describe what you're wondering about, and the agent picks the sources and numbers itself.
Smart things the agent does for you
Knows the page you're on. If you open the agent while you're in a bid or a competition, it already knows what you're working on. That way you don't have to explain which bid you're talking about.
Reads files you attach. Upload PDF, Word, and Excel files directly in the conversation – requirement specifications, emails from the buyer, or previous responses – and the agent will use them in its work, or save them to the knowledge base for you.
Queues your next message. You can write the next instruction while the agent is working. The message is queued and sent automatically when the agent finishes the current step, unless the agent is waiting for your answer to a question.
Asks for approval before changing the search profile. If the agent suggests changes to the search profile, an administrator gets a card saying "Approval required: Search profile". The whole change must be approved with "Approve" or rejected with "Reject" before it is saved.
Takes you to the right place. Need to adjust tender alerts, fill out a price form, or upload the competition basis? The agent replies with a link or button straight to the right place in the app.
Keeps documents consistent. Ask the agent to compare the cover letter, solution description, and competence description, and it will flag contradictions – differing numbers, conflicting claims about references, inconsistent use of terminology.
Shows structured content clearly. Replies can include tables, code blocks, inline field names, and Mermaid diagrams. Wide tables scroll inside the chat, and code blocks can be copied.
Replies in your language. Norwegian, Danish, Swedish, Finnish, German, and English – the agent matches the language you write in.
Follow along while the agent works
While the agent works, you see what it's doing in real time – for example "Searching all content", "Reading Cover letter", or "Editing Solution description". The status line also shows how long it has been working and how many tool calls it has made. Click "Show thinking" to see more detail; repeated tool calls are grouped, and long reasoning traces can be expanded when needed.
The agent can ask you questions along the way if something is unclear. Reply directly in the conversation, or minimize the question if you need to read the context behind it before answering. You can also stop it with the stop button or Esc at any time.
On replies, use copy to copy formatted content, or regenerate on the latest reply if you want the agent to try the same instruction again. If the workspace is out of AI capacity, a clear notice appears by the composer with information about when capacity resets or who to contact.
Tips
Be specific. "Write the solution description focused on sustainability and ISO 14001" gives better results than "write something nice".
Use one conversation per task. That makes it easier to go back and adjust.
Verify the result. AI can make mistakes, especially with numbers, dates, and references.
Ask for the reasoning. Click "Show thinking" in a response to see how the agent reasoned.
Frequently asked questions
What's the difference from the old chat in Cobrief 1?
What's the difference from the old chat in Cobrief 1?
The old AI chat (Cobrief AI) was in practice a writing assistant with predefined actions like "Improve", "Make longer", and "Make shorter". It ran on an older AI model, had access to little context at a time, and couldn't take actions on its own – it could only rewrite text you selected.
The global agent runs on the latest AI models, has access to your entire workspace – the knowledge base, all bids, market insight, and Flowcase – and can carry out multiple steps in sequence. Among other things, it can create new bids, edit documents, and combine searches across sources without you having to switch between screens.
Is there still a separate agent inside the bid?
Is there still a separate agent inside the bid?
No. All new conversations go through the global agent (⌘J / Ctrl+J). When you're in a tender or bid room, the agent loads the right context automatically.
Do I need a paid plan to use the agent?
Do I need a paid plan to use the agent?
Yes. The global agent is available on paid plans and during an active trial. On the free plan, you can still use Pipeline, Search, and tender alerts.
Does the agent change my documents permanently?
Does the agent change my documents permanently?
When the agent edits inside a bid, the changes are saved directly. You can always ask the agent to revert, or edit the documents manually afterwards. When the agent wants to change the workspace's search profile, an Administrator must first approve the whole change in the chat.
Who can see my conversations?
Who can see my conversations?
Conversations are tied to your workspace and are only visible to you and colleagues with access to the same conversation.
What happens if the agent stops mid-work?
What happens if the agent stops mid-work?
Sometimes the agent reaches the limit on how many steps it can take in one turn. You'll then get a yellow message saying "The reply was interrupted because the AI reached the max number of steps". Send a new message – for example "continue", and the agent will pick up where it left off.