New — Introducing Yaju Software Factory — Try the tool for free
Explore nowPlatform
Functionalities
Use Cases
Deployment
By Industry
Explorations
Initiatives
Versions
Every version of the Yaju Agent Orchestration System, from the first runtime to the platform teams run today.
Keter
The first generation. Where Yaju Agent Systems began, and everything it could not do yet.
Keter — Version 1.0
The first build of Yaju Agent Systems. A bare runtime that could execute one agent, and very little else.
Single agent per workspace
No cost visibility
No external integrations
Raw logs, no real monitoring
Keter — Version 1.1
The origin of Yaju Agent Systems as a product. Our first released version still lacked most of what we offer today.
No efficient agent management
No cost visibility
No external integrations
No advanced agent monitoring
Keter — Version 1.2
Introduced the agent catalogue, so teams could finally see what was actually running across the organization.
One catalogue for every agent
Still no owner per agent
No cost visibility
No external integrations
Keter — Version 1.3
Gave every agent a unique identity and a named owner, ending the first wave of shadow agents.
Unique ID for every agent
A named owner per agent
No credential handling yet
No policy at runtime
Keter — Version 1.4
The first credential store, so keys stopped living in repositories, config files and shell history.
Central credential store
Secrets out of agent code
Manual rotation only
No access rules yet
Keter — Version 1.5
Added run history, the first step towards understanding what an agent had actually done.
Persistent run history
Inputs and outputs captured
No tool-level detail
No exportable record
Keter — Version 1.6
Brought the command line, so engineers could deploy and inspect without ever opening a dashboard.
CLI for deploy and inspect
Runs inside CI/CD pipelines
Partial parity with the dashboard
Still no policy enforcement
Keter — Version 1.7
The last Keter release. Stable and genuinely useful, but clearly limited by having no governance layer.
Stable multi-agent execution
Credentials and run history
No policy enforced at runtime
No cost attribution
Nahar
The second generation. Governance and observability arrive: policy at runtime, and a trace for every run.
Nahar — Version 2.0
The second generation opens with a policy engine that runs at the moment of action, not in a document.
Policy enforced at runtime
Access rules per agent and user
Credentials resolved per call
No spend attribution yet
Nahar — Version 2.1
Full session tracing arrives: every model call, tool invocation, input and output captured as the run happens.
Complete session traces
Naharable in real time
Tool-level detail
Traces not yet exportable
Nahar — Version 2.2
Made the record immutable and exportable, so evidence could leave the platform intact for an auditor.
Immutable audit trail
Timestamp and acting user
Exportable on demand
No automatic quality checks
Nahar — Version 2.3
Added human approval gates, pausing the runtime before anything irreversible could happen.
Approval gates on risky actions
Bounded loops and retry limits
Outside agents inherit policy
Still no cost controls
Oran Models
The current generation. Cost control, optimization and the capabilities organizations run on today.
Oran — Version 3.1
The third generation begins with the question nobody could answer before: what does each agent actually cost?
Spend attributed per agent
Breakdown by user and model
Reported in dollars, not tokens
Reporting only, no limits
Oran — Version 3.2
Turned visibility into control with circuit-breaker budgets that stop an overrun instead of reporting it later.
Hard limits per workspace and team
Warning at 80 percent
Automatic block at 100 percent
Forecast for the rest of the cycle
Oran — Version 3.3
Evals arrive, turning passive logging into active quality enforcement on every single run.
Behavioural and structural checks
Latency and cost criteria
Halts an agent on a failed check
Scores every run automatically
Oran — Version 3.4
Optimizer closes the cost loop, cutting spend without letting quality slip on your own benchmarks.
Model and prompt optimization
Tool trimming
Hybrid agent conversion
Measured against your own evals
Oran — Version 3.5
Made agents reusable across the organization instead of being rebuilt from scratch inside every team.
Agent Hub and internal catalogue
Browse, fork, publish, improve
Versioning and one-command rollback
Skills shared between teams
Oran — Version 3.6
Our latest and most capable version, giving organizations what they need to automate work with AI agents efficiently.
Multiple integrations
Agent situation awareness
Advanced agent management
Cost visibility and control
