Open-source · AGPLv3 · Building in public

Idea to delivery,
with lineage.

An open-source platform for product managers that carries a feature from idea to delivery as a chain of connected, approved artifacts — where every AI step is grounded in the approved artifact above it.

ChatGPT can write a PRD. It can't tell you which approved requirement it came from, who signed off, and what changed since. Aurume can.
The differentiator isn't any single AI feature — it's lineage: requirement → test → live pass/fail, with versions.
How it's built

Evidence over autonomy.

A few principles, decided up front. The full rationale lives in the plan.

Agents propose, humans commit

A stateful workflow with human-in-the-loop approval gates — not an autonomous agent. An agent never passes a gate or writes to the board on its own.

Benchmarked, not demoed

Every AI step ships with groundedness scores, cost per generation, and documented failures — the numbers that show the thinking.

Bring your own key

OpenAI, Anthropic, or fully local via Ollama. Self-hostable in one command, so no data need leave your environment.

Deterministic where it counts

Scheduling and cost math are arithmetic, not LLM guesses. The model is used only where judgement genuinely helps.

Tech stack

Chosen deliberately, documented as decisions.

Every choice has a written decision record in the repo explaining the why. The short version:

Platform shell
Next.jsTypeScriptTailwind
Auth & RBAC
Better AuthMulti-org
Data
PostgreSQLNeonDrizzle ORMpgvector
Integrations
ResendSlackTeams
AI services
PythonFastAPILangGraph
Models · BYO key
AnthropicOpenAIOllama
QA & runtime
PlaywrightDockerGitHub ActionsAGPLv3
Roadmap

Shipped in the open, one repo at a time.

The platform spine shipped first — workspace, roles, projects, and a knowledge base. The AI capabilities land next as independent, documented, benchmarked repos the platform assembles. Honest status — only what exists is marked so.

aurumeMulti-org workspace, RBAC, projects, connectors, knowledge spaceLive · v0.1.0
playbook-drafterGrounded, structured product-playbook generationNext
spec-to-storiesGrounded story generation with citations (RAG)Planned
story-to-testsTest cases + traceability matrixPlanned
delivery-evalsGroundedness / faithfulness evaluation harnessPlanned
delivery-graphWorkflow orchestration, approval gates, resumePlanned
qa-agentBrowser agent that authors Playwright testsPlanned
llm-observabilityTraces, cost, latency, quality over real runsPlanned