Software Architect · Agentic AI Infrastructure · GovTech · Open Source
I architect GovTech platforms at publicplan. In parallel I build an open-source ecosystem that lets AI agents plan, build, validate, deploy, and monitor software alongside humans, focused on trustworthy autonomy through claim gates, grounding, and human-in-the-loop patterns.
New here? The quickest thing to grasp is depsight: a CVE, license, and dependency-health dashboard with a Dev Login (no setup, no credentials) at depsight.opentriologue.ai. The most mature module is agent-tasks: enforced workflows for humans and AI agents. The map below shows how it all fits together.
Most repos here are modules of one Project OS: a single pipeline that carries software from idea to production with AI agents in the loop; project-pilot puts its create, develop, and deploy stages behind one dashboard. Each module also runs standalone. Triologue, the human-and-agent chat workspace, is a separate product line; opentriologue.ai is the shared hosted demo.
flowchart TD
subgraph Policy["harness · Policy Plane"]
H["harness<br/>control plane"]
end
subgraph DevLifecycle["Project OS · Human-Agent Dev Lifecycle"]
direction LR
PLAN["agent-planforge<br/>plan + backlog"]
SCAFFOLD["project-forge / scaffoldkit<br/>scaffold"]
COORD["agent-tasks<br/>coordinate"]
VERIFY["agent-grounding<br/>verify"]
GATE["agent-preflight<br/>gate"]
DEPLOY["deploy-panel / agent-relay<br/>deploy"]
MONITOR["agent-ops-dashboard<br/>monitor"]
PLAN --> SCAFFOLD --> COORD --> VERIFY --> GATE --> DEPLOY --> MONITOR
end
subgraph CrossCutting["Cross-cutting layers"]
direction LR
ORACLE["codebase-oracle<br/>semantic search"]
MEMORY["agent-memory<br/>persistent memory"]
TRI["triologue<br/>human+agent chat"]
end
H -.->|governs| DevLifecycle
CrossCutting -.->|serves all stages| DevLifecycle
The full ecosystem: every repo, grouped
Project OS · Human-Agent Dev Lifecycle, the pipeline above, module by module:
| Module | What it does | Live |
|---|---|---|
| project-pilot | Unified dashboard over the create / develop / deploy stages | |
| agent-planforge | Architecture planning + backlog generation | via project-forge |
| project-forge | Project scaffolding from a description | project-forge.opentriologue.ai |
| scaffoldkit | Declarative blueprint engine behind project-forge | via project-forge |
| agent-tasks | Task workflow for humans + agents, with claim gates and audit | agent-tasks.opentriologue.ai |
| agent-grounding | Stops agents acting on assumptions: claim gates, evidence ledger, runtime checks | |
| agent-preflight | Local pre-push validation with an agent-readable confidence score | |
| deploy-panel | Deployment control with an API, an MCP server, and a GitHub Action | |
| agent-relay | Controlled execution on VPS targets | |
| agent-ops-dashboard | Agent-fleet + repo health monitoring | ops.opentriologue.ai |
| harness | Declarative control plane: one YAML for grounding/tools/memory/hooks/policies |
Supporting libraries and tools
| Library | What it does |
|---|---|
| codebase-oracle | Semantic search across all your local repos, via MCP or CLI |
| agent-memory | Sync, route, and digest agent memory across sessions and machines |
| agent-dx | Workshop of CLIs and playbooks for building AI-agent coding workflows: slop-detector, orchestrator-workflow, okf-kit |
| repo-intelligence | Repository hygiene, PR quality, CI health, and performance-drift scoring; sibling to depsight and agent-ops-dashboard |
| opencode-sandbox | Bash wrapper for running OpenCode in Docker against the current workspace |
- depsight: dependency health, CVE tracking, license + security scoring · live
- triologue: chat workspace where humans and AI agents collaborate as a team · opentriologue.ai; external agents connect through triologue-agent-gateway
- telerithm: AI-powered log analytics for self-hosted teams
- clawd-monitor: monitoring dashboard for OpenClaw, fed by the push-based clawd-monitor-agent
TypeScript · Next.js · Hono · Node.js · PostgreSQL · Prisma · Docker · Traefik · Symfony
Architecting trustworthy AI agents, in public.