OR is an AI-powered control plane for product-driven engineering teams.
AI coding tools (Copilot, Cursor, agent IDEs) have dramatically increased code output — more PRs, more lines changed — but overall product cycle time has not improved proportionally.
The bottleneck has shifted from writing code to workflow orchestration and context management.
The root issue is structural: AI accelerates the middle of the pipeline but leaves the upstream (spec clarity, acceptance criteria) and downstream (validation, measurement) stages unchanged. You get more code, but not more outcomes.
OR introduces a structured workflow layer that sits between product intent and execution:
Intent → Execution → Evidence → Outcome → Feedback
It connects and coordinates the tools teams already use — Jira, Linear, Confluence, Notion, GitHub, GitLab, CI/CD, IDEs — rather than replacing them.
Control Plane (OR): context synthesis, execution planning, evidence validation, outcome tracking. Execution Plane (your tools): code editors, Git, CI/CD, test frameworks — OR coordinates and validates this layer, not replaces it.
| Concept | Description |
|---|---|
| Execution Bundle | A structured, machine-readable execution plan derived from a ticket + spec. Contains tasks, dependencies, acceptance criteria, and relevant context. Input for AI-assisted development in the IDE. |
| Context Synthesis | OR pulls context from tickets, documents (Confluence/Notion), and code to produce a unified context for each bundle — reducing the manual reconstruction that happens at every handoff. |
| Acceptance Criteria Traceability | Criteria are linked from requirement through to code diffs and test coverage, making them programmatically verifiable rather than free-text checklists. |
| Evidence | Structured payload capturing test results, coverage, CI status, and validation signals — ingested from CI/CD pipelines and submitted from the IDE. |
| Evidence Validation | Before human review, OR validates whether the evidence satisfies the bundle's acceptance criteria, reducing review bottlenecks and clarification loops. |
| Outcome Feedback Loop | Post-release KPI signals are tied back to the bundle and the original spec, closing the loop from idea to measurable outcome and feeding insights into the next cycle. |
Idea → Spec → Ticket
↓
[OR: Context Synthesis]
↓
Execution Bundle ──→ Dev (IDE + AI)
↓
PR → CI
↓
[OR: Evidence Validation]
↓
Release → Measure
↓
[OR: Outcome Tracking]
↓
Feedback ──→ (next Idea)
For the full architecture and RFC index, see docs/orca.md and docs/rfc/.
Control Plane for the product flow: Intent → Execution → Evidence → Outcome → Feedback. Orchestrates workspaces, tickets, execution bundles, evidence, and AI-assisted chat.
- Node.js 20+
- pnpm 9+
- PostgreSQL (with pgvector for vector search)
pnpm installCopy .env.example to .env in apps/api and fill in the required values (at minimum DATABASE_URL and JWT_SECRET).
pnpm dev # build packages + run API with hot reload
pnpm dev:web # run Next.js web app
pnpm dev:worker # run background workerAPI runs at http://localhost:3000.
pnpm build
pnpm start # starts @or/api| Variable | Description |
|---|---|
DATABASE_URL |
PostgreSQL connection URL |
JWT_SECRET |
JWT signing secret (min 16 chars) |
| Variable | Default | Description |
|---|---|---|
VECTOR_DATABASE_URL |
DATABASE_URL |
Separate DB for pgvector (falls back to DATABASE_URL) |
AI_DECOMPOSER_PROVIDER |
stub |
stub, openai, or anthropic — AI bundle decomposition |
AI_DECOMPOSER_MODEL |
provider default | Override decomposer model |
AI_CHAT_PROVIDER |
stub |
stub, openai, or anthropic — AI chat agent |
AI_CHAT_MODEL |
provider default | Override chat model (gpt-4o-mini / claude-sonnet-4-20250514) |
OPENAI_API_KEY |
— | Required for OpenAI providers |
ANTHROPIC_API_KEY |
— | Required for Anthropic providers |
CONTEXT_CACHE_TTL_MINUTES |
5 |
Context synthesis cache TTL |
| Package | Description |
|---|---|
@or/agent |
Standalone AI chat agent (provider-agnostic, streamable) |
@or/domain |
Shared Zod schemas and types |
@or/sdk |
Typed API client |
@or/code-chunker |
Code chunking for vector indexing |
@or/ui |
Shared UI components (used by @or/web) |
| App | Description |
|---|---|
@or/api |
Hono API server |
@or/web |
Next.js web dashboard |
@or/worker |
Background worker |
See LICENSE.