Central Orchestrator
Control plane for the fleet. Runs daily cognitive health audits, enforces skill integrity, and maintains strict security isolation between domains.
Production AI Agents
I build the tools, guardrails, and integrations that move AI models beyond chat and into production — where mistakes are expensive.
- 450+ production PRs merged by tech leads in under 4 months (>83% human acceptance rate) across 10 repos under a strict 2-PR limit.
- 9 AI agents in production — one central orchestrator and eight specialized profiles across engineering and operations.
- Self-hosted on Hermes — the open-source agent runtime by Nous Research.
pantera-bot — engineering agent
Real GitHub activity
The agents
Engineering & FinOps
Resolves GitHub issues, runs local test suites, and opens PRs across 10 production repositories under human review while pruning AWS cloud spend.
Business Operations
Seven dedicated profiles across sales, CS, support, marketing, projects, intelligence, and leadership, handling 3,000+ human messages monthly.
450+
Merged PRs
>83% human acceptance rate in <4mo
310
Code & Infra PRs
10 production repos under 2-PR limit
15%
AWS Cost Cut
Automated resource & disk pruning
<$140/mo
Token Spend Ceiling
Operational cost capped per profile
Workload Distribution (450 Merged PRs in <4 Months)
69% Code & Infra (310 PRs)
15% Architecture (66 PRs)
16% Governance (74 PRs)
Guardrails that make delegation real
Production repositories need more than a good model. This is what had to exist first.
01 // Human review
No direct pushes, ever
Capped at 2 open PRs per repo so the team never gets overwhelmed. The agent runs local tests, CI must pass green, and an engineer always reviews before merge.
02 // Real scope
Real problems, zero busywork
No cosmetic churn or whitespace refactoring. The agent tackles actual backlog issues: async race conditions, component migrations, and end-to-end test suites.
03 // Isolation
Each agent in its own lane
Each profile runs in an isolated sandbox with its own tools and procedures. The sales agent never touches source code, and context never bleeds across domains.
04 // Cost & safety
Scripts before LLMs
Anything that can be solved with Bash, Python, or Git never calls an LLM. This prevents hallucinations, runs quiet security checks, and keeps token spend strictly under $140/mo per profile.
How it's built
Self-hosted end to end — the agent runtime, SQLite state, and every integration run on our own infrastructure, decoupled from inference providers. Swapping the model is a config change; the guardrails and review process stay the same.
Hermes Agent Runtime
Python
Playwright
SQLite
GitHub CLI & API
Linux / Systemd
Bash
AWS