velopersCase studies / agentside · Open-source product
An AI assistant that works inside your own software
agentside is the engine behind the assistant in Airbase, and we made it open source. It lets an AI assistant do real tasks in any software product through that product’s own API, as the person who is logged in.
- Freeopen source, MIT licence
- Your APIthe only way the AI gets in
- Any AIworks with any model
The problem
Most AI assistants added to software can only chat. They answer questions about the product but cannot do the task.
The ones that can act are often given direct access to the data, which no careful company wants.
What we built
It acts through your API
Every action goes through the product’s existing API as the logged-in user, so the assistant can never do more than that person could.
New tasks taught in plain writing
Each task is described in a short text file. Teaching the assistant something new means writing a page, with no new release.
Any AI model
Conversations are stored and can be replayed, and the model behind them can be swapped.
Small enough to read
The whole engine can be read in an afternoon.
How AI helps them scale
- Users ask for a task in their own words and the product does it
- Security stays where it already is, in your API and your permissions
- Your team adds new abilities by writing, not by rebuilding
- It already serves real users in production, inside Airbase
In production inside Airbase, and free for anyone to use.
agentside is not a demo. The same engine runs the assistant that Airbase’s recruiters use every day, and it also powers the team of agents inside scnz.
- FreeMIT licence
- Your APIthe only way in
- Any AIthe model can be swapped
- Liveserving real users in Airbase
More case studies
100% of applicants interviewed and scored by AI
The AI recruiter we built interviews every applicant by voice, scores them against the role’s criteria and hands recruiters a ranked shortlist in hours, not weeks.
Read the case study Dutch Exams · Paid product180+ practice exams marked by AI
A subscription product where AI marks speaking and writing the way an examiner would.
Read the case study