Building an agent takes an afternoon. Running it is the real job.
nara brings what is missing between demo and production: typed tools, validated Operating Protocols, a test console, and versions with rollback. Every action lands in the log.
The demo was easy. Then reality arrived.
An agent that works in a test chat is far from one you can let loose on real tickets. For that you need defined flows instead of free prompting, clear limits on which tools an agent may call at all, and answers to the uncomfortable questions: What did the agent do, and when? How do I roll back a broken version? Who approves critical actions? nara answers these questions in the platform, not in your wiki.
Primitives that survive production
Tool Registry
Every tool has typed parameters and a result schema. Agents only run what you have approved for them, per agent. No tool, no call.
Operating Protocols
Flows are a multi-phase graph with validation: exactly one start phase, no unreachable phases. Escalation to people is a defined phase with the full history.
Test Console
You test tools and workflow agents right in the editor: JSON arguments in, execution, result out. No deployment for a test run.
Model tiers per agent
Budget, Balanced, or Pro, depending on what the agent has to deliver. You decide per agent, not once for the whole platform.
The Ask Me assistant builds with you. You keep the final say.
Describe
You describe the agent in natural language: what it should solve and where the limits are.
Proposal as a diff
nara proposes phases, tools, and metadata as a reviewable diff, not as finished magic.
Review
You review every change.
Nothing is applied until after your review, never automatically.
Test
In the Test Console, you run the agent with real arguments before it sees a single real ticket.
Your infrastructure, your tools, in TypeScript
You write your own edge tools in TypeScript, package them as a bundle, and distribute them centrally. The Edge Connector runs them on devices, servers, and behind the firewall, over an outbound-only connection with no firewall changes. Versions are pinnable, rollback included: if version 12 causes trouble overnight, you pin to version 11 and debug in the morning.
For the day something goes wrong
Operations means being able to trace, limit, and revoke.
Activity logs
Every agent action and every tool execution is logged. No guessing what happened.
Human in the loop
Teams can approve or reject agent plans per ticket before they run.
Server tokens
Stored only as a SHA-256 hash, shown exactly once, revocable immediately.
REST API
You run workflow agents via the REST API and hook them into your existing processes.
Roles and SSO
Roles control access, sign in via SAML and OIDC.
Data location
ISO 27001 certified, hosting and development in Germany, no training with customer data, GDPR compliant.
Frequently asked questions
Which language do I write my own tools in?
You write edge tools in TypeScript, packaged as a bundle and distributed centrally.
Can I trigger agents via API?
Yes, you run workflow agents through the REST API.
How do I test before anything goes live?
In the Test Console: enter JSON arguments, execute, and check the result right in the editor. That applies to tools and to workflow agents.
What happens when a new tool version causes problems?
Versions are pinnable. You pin to the last working version and get an immediate rollback.
Does the platform prevent broken protocols?
Operating Protocols are validated: exactly one start phase, no unreachable phases. Errors surface while building, not in operation.
Can I use a different model per agent?
Yes. You choose the Budget, Balanced, or Pro model tier per agent.
See the platform from the inside.
In the technical demo we walk through the Tool Registry, Operating Protocols, the Test Console, and edge tools, on a real example instead of slides.
