Open Source Initiative · March 2026

Groven makes the growth of ideas visible.

A typed idea graph for collaborative thinking and collective governance.

Initiated by Sofilab GmbH, Munich · Co-funded by the European Union (Creative Europe)
Scroll

Existing platforms cannot distinguish between a reaction and a synthesis.

01

Engagement ≠ Deliberation

Likes, feeds, and view counts reward provocation over precision. A reaction outperforms a careful synthesis. Platforms aren't broken — they're working exactly as designed. Just not for governance.

02

All contributions look equal

A comment that sharpens an idea is structurally indistinguishable from one that misses the point entirely. At scale, the reasoning disappears into noise.

03

Memory loss over time

A forum thread from six months ago is unreadable as a document. The reasoning behind governance decisions becomes irrecoverable. Accountability requires legibility.

Seed → Branch → Forest

Inspired by Git's version control model, Groven organises all content around a single structural metaphor — a living graph of ideas, not a flat thread of reactions.

🌱
Seed

A contribution that introduces an idea, question, or position. Always text — optionally combined with images, drawings, video, or links. The root node of every thread.

  • No parent — root node in the graph
  • Text-first; media as supplement
  • The starting point for any discussion
🌿
Branch

A contribution that explicitly builds on an existing Seed or Branch. Has exactly one parent and inherits its lineage. Not a reply — a continuation.

  • Requires a description: what does this build on, and where does it take the argument?
  • The LLM proposes a semantic type
  • Author confirms or overrides
🌳
Forest

The growing network of all Seeds and Branches in a Groven space. The living memory of the community — searchable, filterable, permanently archived.

  • Visualised as an interactive graph
  • Filterable by type, author, status
  • Governance decisions anchored as immutable nodes

Not all contributions are the same.

Groven makes the semantic relationship between contributions structurally visible. The LLM proposes a type — the author confirms or overrides.

Clarification

Sharpens a vague or ambiguous idea. Does not add new content — makes precise what is already there.

"You probably mean X, not Y — here is why the distinction matters."
Extension

Carries the idea into territory the Seed did not anticipate. Adds something genuinely new while remaining faithful to the original direction.

"This also applies to context Z, which the Seed did not consider."
Reframing

Interprets the same idea from a fundamentally different angle. Does not contradict — but arrives somewhere else from the same starting point.

"If we frame this differently, the same observation leads somewhere else."
Contradiction

Identifies a conflict and proposes a divergent direction. The most structurally important type — where genuine disagreement becomes visible.

"This premise seems wrong — here is an alternative direction."
Synthesis

Connects two or more existing Branches that have developed separately. Attempts to reconcile or integrate divergent lines of thought.

"Branch A and Branch C are not mutually exclusive — together they imply X."
Contested Signal

When the LLM proposes a type and the author selects a different one, Groven generates a visible "Contested" signal on that node.

This marks places where classification is uncertain — where community norms are still forming. These are exactly the nodes most worth examining.

Not a moderator. Not a recommender.
A semantic observer.

Engagement metrics fail for governance because they optimise for interaction volume, not argument quality. Groven uses an LLM to make the semantic structure of a discussion visible — not to steer it.

What it does
  • Reads every Branch with its parent and proposes a semantic type
  • Provides a brief explanation of its classification
  • Generates structured summaries of discussion threads on request
  • Signals where ideas have grown collectively in the Forest
  • Identifies threads where Synthesis may be possible
What it does not do
  • Steer conversations toward any particular outcome
  • Hide, suppress, or deprioritise any contribution
  • Decide when a topic is ready for a vote
  • Evaluate the quality or correctness of ideas
  • Act without human review or override
The LLM proposes — the author decides.

Voting is not a separate tool — it is a state transition in the same graph.

A separate voting tool breaks the chain of accountability. Groven integrates the vote into the same graph where the discussion took place — so the reasoning is permanently legible.

Open

Discussion is active. Seeds are planted, Branches grow, each Branch is semantically classified. Anyone with access to the space can contribute.

Ready

A designated body declares the topic ready for decision. The LLM generates a structured summary: positions, attempted syntheses, open forks. Discussion closes for new Branches.

Declaration of readiness is always a human decision.
Decided

Vote complete. The result — including full vote breakdown and required written justification — is anchored as an immutable node directly in the Forest, linked to the Branches it resolved.

Minority positions are always recorded alongside the majority decision.

The combination that doesn't exist yet.

Feature Kialo / Pol.is Loomio / Decidim Groven
Idea lineage graph Partial (tree only) No Full ancestry
Branch typology Pro/Con only No 5 semantic types
LLM structuring assistant No No Visible + overridable
Voting in same graph No Partial State transition
Decision anchored in lineage No No Immutable node
Minority position recording No Partial Always required
Self-hostable open source No Yes / Partial Core commitment

A governance tool must be transparent to be trusted.

MIT License

Maximum reuse, including commercial use. Any organisation can self-host Groven without licensing obligations. Source code is always open.

Self-hostable

No mandatory cloud dependency. Any community can run their own Groven instance on their own infrastructure.

Governed by itself

Roadmap decisions made through public Groven discussions. The RFC process uses the Seed→Branch→Forest model natively. We're accountable to the same standard we advocate for others.

CORPUS as reference

First and reference implementation. Validates the concept in a real governance context. Groven has zero dependency on CORPUS — fully general-purpose.

CORPUS
Royalty Protocol

CORPUS is an open licensing and royalty infrastructure for music in the age of AI — developed by Sofilab GmbH, Munich, co-funded by the European Union (Creative Europe).

It requires a governance platform that makes the reasoning behind decisions permanently legible, connects community discussion to formal voting, and scales without losing deliberation quality.

Groven is domain-agnostic. CORPUS is its first proof of concept.

Status Development / Pilot
Community <50 contributors (launch)
Selection Stratified sortition
Governance Sortition Jury model
Domain Music & AI licensing
Funding EU Creative Europe

From prototype to public release.

Phase 0
M 1–2
Foundation
Repo, data model, auth, basic Seed creation and display.
Phase 1
M 3–4
Seed→Branch→Forest
Branch creation, lineage graph, Forest visualisation (tree view).
Phase 2
M 5–6
LLM Integration
Type proposal, explanation display, author override, Contested signal.
Phase 3
M 7–8
Governance Module
Voting, Ready state declaration, Decision node anchoring in Forest.
Phase 4
M 9–10
CORPUS Pilot
Production integration. Sortition jury onboarding. Real governance load.
Phase 5
M 11–12
Public Release v1.0
Open source release. Documentation, contribution guide, RFC process activated.

Groven is where structured thinking becomes collective action.

It is open source, self-hostable, and governed by the same tool it provides.

This page is a Seed. Fork it.