Skip to content
Logrite

AI Agents Roadmap

Overview

Section titled “Overview”

The agentic AI system is built in 5 phases, progressing from foundational infrastructure to advanced multi-agent orchestration.

Phase 1: Foundation (Backend + AI Service Extensions)

Section titled “Phase 1: Foundation (Backend + AI Service Extensions)”

Goal: Establish the data model and NLP creation flow.

Deliverables

Section titled “Deliverables”
  • agents DB table with full schema (Backend)
  • agent_execution_logs DB table (Backend)
  • Agent CRUD API endpoints (/api/v1/agents/*) (Backend)
  • POST /api/v1/agent-builder/parse endpoint (AI Service)
  • Tool Registry (src/tools/registry.ts) (AI Service)
  • toolsAllowed filter in AgentLoop (AI Service)
  • AgentBuilder UI — NLP input + review screen (Frontend)
  • Agent List page /agents (Frontend)
  • Tool Browser modal (Frontend)

Success Criteria

Section titled “Success Criteria”
  • User can type “Monitor payment errors hourly” and see a parsed agent config review screen
  • Agent is created via Backend API with correct schema
  • AgentLoop respects toolsAllowed filter

Phase 2: Agent Runner

Section titled “Phase 2: Agent Runner”

Goal: Agents execute automatically on schedule.

Deliverables

Section titled “Deliverables”
  • New logrite-agent-runner repo under Deployd-io
  • Node.js/TypeScript Express service on port 9400
  • Cron scheduler using node-cron
  • Backend polling (60s interval)
  • Execution context builder
  • AI Service /chat/sync integration
  • Execution log writer
  • Basic alert condition evaluation
  • Discord alert posting
  • Ship it as a compose service and add it to the deploy catalogue
  • Master CLI support (logrite-manager.js start agent-runner main)
  • Execution history tab in AgentDetail (Frontend)
  • Agent metrics row (success rate, last run, avg duration) (Frontend)

Success Criteria

Section titled “Success Criteria”
  • Active agents with cron schedules run automatically
  • Execution logs visible in the UI
  • Discord alerts fire when thresholds are breached

Phase 3: Write Tools + Alert Engine

Section titled “Phase 3: Write Tools + Alert Engine”

Goal: Agents can take action, not just read.

Deliverables

Section titled “Deliverables”
  • write category tools: create_scan, dismiss_anomaly
  • external category tools: send_email, call_webhook
  • Alert engine: condition expression language (beyond simple gt/lt)
  • Alert history page (Frontend)
  • Alert silence/snooze functionality
  • Rate limit enforcement per workspace
  • Agent templates library (pre-built templates for common monitoring tasks)

Success Criteria

Section titled “Success Criteria”
  • Agent can trigger a compliance scan when a pattern is detected
  • Users can silence alerts for configurable periods
  • Template-based agent creation in < 30 seconds

Phase 4: Event-Triggered Agents

Section titled “Phase 4: Event-Triggered Agents”

Goal: Agents fire on events, not just time.

Deliverables

Section titled “Deliverables”
  • Event bus in Agent Runner (subscribe to Backend events)
  • Trigger types: anomaly detected, error rate threshold, scan complete, log pattern
  • Event-based agent config in agents table
  • Frontend: event trigger selector in AgentBuilder
  • Event history in AgentDetail
  • Webhook trigger support (external systems can trigger agents)

Success Criteria

Section titled “Success Criteria”
  • Agent fires within 60 seconds of an anomaly being detected
  • External systems can trigger agents via signed webhook

Phase 5: Multi-Agent Orchestration

Section titled “Phase 5: Multi-Agent Orchestration”

Goal: Agents can spawn sub-agents and coordinate on complex tasks.

Deliverables

Section titled “Deliverables”
  • Agent-to-agent calling protocol
  • Orchestrator agent type (coordinates multiple sub-agents)
  • Parallel execution support in Agent Runner
  • Shared context/memory between coordinating agents
  • Agent DAG visualization (Frontend)
  • Multi-agent execution logs with call tree

Success Criteria

Section titled “Success Criteria”
  • A high-level “Incident Responder” agent can delegate to specialized agents for different log types
  • Full call tree visible in execution logs

Current Status Summary

Section titled “Current Status Summary”
PhaseStatusETA
Phase 1: FoundationShipped 2026-07-19—
Phase 2: Agent RunnerPartially shipped — manual runs live, scheduling not—
Phase 3: Write ToolsFutureQ2 2026
Phase 4: Event TriggersFutureQ3 2026
Phase 5: Multi-AgentFutureQ4 2026