Skip to main content

Production Agent Ecosystem

Not theorizing about agents -- living with them daily. Four autonomous systems managing trading, home automation, health monitoring, and developer workflow analysis.

  • Agent Architecture
  • MCP Protocol
  • Swift + TypeScript
  • IoT + Health APIs

The Thesis

Autonomous agents are production systems, not experimental chat interfaces. The same observe-decide-act loop that powers high-frequency trading applies to every agent in this ecosystem. Each operates independently, makes hundreds of decisions daily, and degrades gracefully under uncertainty rather than failing silently.

Trading as Agency

Trading bots are autonomous agents by any definition. They observe real-time market data through Geyser gRPC streams, evaluate multi-signal conditions in the Brain engine, execute through MEV-protected Jito bundles, and adapt their configuration based on outcomes. Hundreds of autonomous cycles daily with no human in the loop.

OBSERVEMarket StateDECIDEStrategy EvalACTExecute TradeLEARNAdapt ConfigContinuous feedback loop · Hundreds of cycles dailyGeyser gRPCReal-time streamsBrain EngineMulti-signal evalJito BundlesMEV-protected submitAdaptive ConfigParameter tuning

Personal Agents

Three purpose-built agents handle distinct domains of daily life. Each is a standalone system with its own runtime, data store, and decision logic.

OPERATORHuman-in-loopISEHome Automation97 automations254 sensorsBun + Home AssistantWLED · NFC · ZigbeeAPEXHealth MonitorApple HealthKitSwift iOS AppTypeScript BackendPersonalized ActionsDEVREPORTSWorkflow AnalyticsClaude SessionsGitHub ActivityEmail + CalendarStructured ReportsAll agents: observe → decide → act autonomously · Human override always available
Ise
Home Automation
97 automations across 254 sensors via Home Assistant. WLED light effects, NFC physical tags for reminder dismissal, Zigbee/Matter device mesh. Bun runtime with 50+ CLI command scripts.
Apex
Health Monitor
Swift iOS app syncing Apple HealthKit data (10 biometric types) to a Bun/Express backend with SQLite. AI agent layer provides personalized coaching from structured health records.
DevReports
Workflow Analytics
Analyzes Claude Code AI sessions, GitHub activity, email, and calendar into structured productivity reports. Identifies patterns across development workflows.

MCP Contribution

Published perplexity-mcp on Smithery -- a Python MCP server that exposes Perplexity AI search as a tool for any MCP-compatible client. Single-file async architecture supporting multiple model backends including Sonar, Sonar Pro, and deep research modes.

Tool( name="perplexity_search_web", description="Search the web using Perplexity AI", inputSchema={ "properties": { "query": {"type": "string"}, "recency": { "type": "string", "enum": ["day", "week", "month", "year"] } }, "required": ["query"] } )

Research Foundation

IEEE-published neural network research providing the theoretical foundation for the agent architectures deployed in production. The transition from academic research to production autonomous systems informs every design decision in this ecosystem.

Outcomes

4
Production Agents
97
Home Automations
200+
Daily Trading Ops
1
MCP Tool on Smithery