FIND-20260325-024 · 2026-03-25 · Innovation Veille
Chorus — AI-DLC Agent Orchestration Harness (AGPL-3.0)
adhoc
HIGH
Chorus is an open-source AI-Driven Development Lifecycle (AI-DLC) orchestration framework shared by @tom_doerr on March 25, 2026 (tweet ID 2036890176367009982). It manages multiple AI agents (PM, Developer, Admin) through a complete task lifecycle — from requirements to delivery — with a real-time Kanban board, task DAG visualization, sub-agent coordination, and failure recovery. Built with TypeScript/Next.js 15/React 19/PostgreSQL/Prisma/Redis and integrated with Claude Code via the MCP SDK. The architecture mirrors the ODS ADLC pipeline almost exactly: it implements a dispatcher pattern, circuit breaker logic, human-approval workflows, and role-based agent specialization. Its 'Reversed Conversation' philosophy (AI proposes, humans verify) matches the ODS approach of human-gated promotions. Relevance is HIGH for ODS because Chorus could serve as a reference implementation or complement the ods-dashboard for visualizing ADLC state. HOWEVER: licensed under AGPL-3.0 which requires legal review before any code reuse or SaaS deployment.
Source
https://github.com/Chorus-AIDLC/Chorus
ODS Impact
Directly relevant to the ADLC orchestration layer. The Kanban + DAG task visualization and real-time activity stream could inform the ods-dashboard roadmap. The MCP SDK integration pattern is reusable for Claude Code agent coordination. The multi-agent session lifecycle management (including recovery) is the same problem ADLC solves — comparing implementations could surface improvements. Do NOT incorporate Chorus code directly due to AGPL-3.0 license — use as architectural reference only. Affected services: ods-dashboard (UI patterns), ADLC pipeline (orchestration patterns).
Security Review
License: AGPL-3.0 — FLAG: requires legal review before any code reuse or SaaS exposure | Maintenance: ACTIVE | Risk: MEDIUM | Recommendation: USE_WITH_CAUTION
Tags
ai-agents
orchestration
typescript
nextjs
postgresql
adlc
mcp
claude-code
agpl
kanban