Open-source agentic delivery control plane
DevLoop Agents
A portable, observable control plane for AI coding, QA and monitoring agents—turning issues and health signals into reviewable engineering work without locking teams to one tracker or model provider.
- 3 roles
- coder, QA and monitor
- 18 tests
- safety and lifecycle coverage
- Provider-neutral
- tracker and AI execution
The constraint
The production system behind PolyAgents was effective but tightly coupled to private Jira, GitLab, staging infrastructure and product-specific prompts. The challenge was to preserve the operating model while removing every private dependency and adding safe defaults for public reuse.
What I built
I rebuilt it as a provider-neutral Python toolkit with GitHub Issues and offline adapters, bounded coder, QA and monitor roles, SQLite leases, JSONL telemetry, budgets, health probes, incident deduplication and a local dashboard. Human review remains mandatory: automatic merge and deployment are rejected, agent credentials are isolated and raw model output stays local by default.
Ownership
System design, AI-agent orchestration, adapter architecture, security boundaries, Python implementation, dashboard, testing, documentation, packaging and open-source release.
- Python 3.11+
- GitHub REST API
- SQLite
- Docker
- TOML
- JSONL
- AI agent CLIs
- GitHub Actions