TOMAS
Topicus Onderwijs Multi Agent System
One agent writes code. A team ships it. TOMAS is a virtual team — specialist agents that plan, build and review together, with the whole run visible and a human at the gate.
It plugs into the systems you already use — able to pick work up from a Jira ticket and hand it back as a pull request, through the same reviews and pipeline as everyone else. No separate path for AI.
One request becomes a coordinated run.
A single request becomes a plan, a set of steps, and a reviewable result — with every decision streaming as it happens, not summarised afterwards.
Read the request
Weighs how hard the work is, breaks it into steps, and drafts a plan you can read.
Your call, before anything runs
Read the plan, reshape it, or wave it through. On, off, or escalation-only — per session.
Specialists, one step at a time
Each step gets the agent suited to it, working in an isolated VM with real tools — never on your machine.
Checked before it reaches you
While one agent writes, another reviews — corrections happen inside the run, not in your inbox.
See the system think.
Live execution graph
A real-time DAG of the run — active node, visited path, and every decision as it streams.
Human-in-the-loop
Approve, reject, or reshape the plan before execution — supervision exactly where you want it.
Agents as configuration
Roles, prompts, tools and skills are declarative — add or change an agent without touching the graph.
Context, centrally curated
Frameworks, conventions and architecture reach the agents up front — so what comes back fits this codebase instead of needing a rewrite.