A customer, on a project page, asks:
Passar det här projektet oss?
Does this project suit us?
Your next project, found by asking.
Rune is Hubexo's AI assistant, built into the products customers already use — Rise on the web, Reach on mobile. They ask in their own language. The answer comes from Hubexo's own project database: the projects, the companies on them, and the people to call.
- looked upthe project on the page
- looked upthe customer's own company
- judgedfit and timing
Pursue — and position now. Nya vårdcentralen is a 4,200 m² care centre in the design stage, start planned Q2 2027. You install ventilation for healthcare buildings, and the installation contractor is not yet appointed. The consulting engineer is.
One conversation, instead of an afternoon of searching.
Finding a project, judging it and reaching the right person are three separate steps in the product today. Rune does them in one exchange. The questions are real ones from customers.
Find the work
Projects by place, trade, stage, value and start date — in the words customers already use, in ten languages.
Visa kontorsprojekt i Lund kommun med byggstart 2028.Show office projects in Lund starting 2028.
Judge it
Is this project worth pursuing for us? A fit and timing verdict for the customer's own company, with the roles still open and the next step.
Passar det här projektet oss?Does this project suit us?
Act on it
Names and numbers for the companies on the project, and Hubexo's own sales playbook for that kind of business — how companies like theirs win this work.
Vem ska jag ringa, och hur pitchar vi det?Who should I call, and how do we pitch it?
And it keeps going. Rune recommends projects from the ones a customer has already saved, and it remembers the last answer — so "just 2029 instead" or "what was number 157 on the list?" simply work.
It reads our database. It has no opinions.
Every sentence Rune writes comes from a lookup in Hubexo's own data, made in that conversation. It has no internet access and is not allowed to answer from general knowledge — when it did not find something, it says so.
- When it does not know, it says so. A rule in its instructions, not a fallback.
- A failed lookup is reported as a fault — never as "no projects found".
- Customers see only the markets they license. The licence is checked on every query.
- You can see what it looked up before the answer arrives.
"You read a project database. You are not a source of market knowledge."
From Rune's own instructions
Everything stays in the EU. Rune adds no new access.
Rune runs in AWS Stockholm, on the same governed Databricks data and the same customer login as the rest of the platform.
Where does the data go?
- Nowhere outside the EU. The AI model — Claude, on Amazon Bedrock — is reached only over an EU route. No other route is configured.
- Project data is read from Hubexo's own Databricks gold layer, in the EU workspace.
- Conversations and records are stored on Databricks, under the same governance as every other Hubexo dataset.
- No second data processor.
Who can see what?
- Every request carries the user's own login (a JWT from Hubexo ID) — never a shared account.
- The licence is enforced on every query, in three independent places.
- One user can never reach another's conversations. Rune can only ever profile the customer's own company.
- Contact details come through one capped path, never through free-form queries.
- Credentials are never written to a log, a record or the database — verified, not assumed.
Two cents an answer, with a ceiling on every axis.
Cost and quality are measured per answer, not estimated per month.
What does it cost?
- The median answer costs
$0.02, measured across real answers in test. Every answer's cost is recorded, so spend is visible per customer and per day. - Usage limits per customer and per user.
- Hard caps on the work a single question can trigger.
How do we know it's good?
- Every answer is recorded in Databricks and reviewable: the question, the answer, the cost.
- Changes are graded before release against a set of customer questions — by automatic checks and by an AI judge reading the answer the way a customer would.
- A regression points at the exact change that caused it.
Nothing new to procure.
Claude on Amazon Bedrock for the AI model. The Databricks platform Hubexo already governs for data, records and evaluation. The customer's existing login for identity. And one Rune service in between that both web and mobile use — so the next capability reaches customers without a second build.
For the customer
One place to ask. An answer they can act on today.
That is the point of Rune: a salesperson opens the product they already use, asks the question they actually have — in Swedish, Finnish, Danish, Norwegian or English — and gets the projects, the fit, the people and the pitch, from Hubexo's data, in one conversation.