by SportERP

AI Agents

Autonomous agents that do real work — with your approval

A five-tier hierarchical agent system plans, researches, implements, and reviews work. You stay in control with human-in-the-loop approvals before anything ships.

14 specialist agents across 5 tiers

Each agent has deep expertise in its domain and collaborates with others to deliver complete features.

Tier 1 — Orchestrator
planner

Breaks complex work into validated 30min-4hr tasks

Tier 2 — Strategy
architectllms-expertcomputer-vision-expert

System design, AI/ML integration, and vision AI

Tier 3 — Research
codebase-analystlibrary-researcherux-ui-researcher

Deep analysis of your codebase, libraries, and UX patterns

Tier 4 — Implementation
ui-implementation-expertbackend-api-expertdatabase-expertintegration-expert

Frontend, backend, database, and third-party integrations

Tier 5 — Quality
testing-expertperformance-expertdocumentation-expert

Tests, performance profiling, and technical docs

Agentic workflows with crash recovery

Every work order is tracked as a project in RuNari's database. If a session drops, the agent reconnects and picks up exactly where it left off. No lost work, no duplicate actions — guaranteed by project_id-based crash recovery.

  • 5-phase workflow: plan → research → implement → test → document
  • Tasks validated to 30min-4hr to prevent scope creep
  • Crash recovery: reconnect via project_id after any interruption
  • Full audit trail of every agent action and decision
Screenshot coming soon
Screenshot coming soon

Real-time streaming chat via CopilotKit

Interact with agents through a streaming chat panel powered by CopilotKit and the AG-UI protocol. Watch agents think and act in real time. Preview dashboard widgets, approve code changes, and redirect agents — all without leaving the UI.

  • Streaming SSE responses — see agent thoughts in real time
  • Preview dashboard widgets before they're published
  • Chat history persisted per session and per project
  • Switch between co-pilot and autonomous modes

Human-in-the-loop approvals

Agents flag actions that need your sign-off — code pushes, database migrations, external API calls. You see exactly what will happen, review the diff, and click Approve or Reject. Nothing ships without your explicit consent.

  • HITL gates on all destructive or external operations
  • Full diff view before any code change is applied
  • Approval audit log with timestamp and approver
  • Configurable approval thresholds per agent type
Screenshot coming soon

Let AI agents handle the repetitive work

Start with the Free plan and unlock agent work orders on Pro.

View PricingContact Sales