OFL

Open Facilitation Library

Open‑source building blocks for agentic facilitation.

Specifications for facilitation methods, the evaluations that hold them to a standard, and the research behind both. Open source, and not tied to any one platform.

Premise

Facilitation is a critical capability for developing collective intelligence.

Facilitation is how a group thinks together: drawing people out, holding disagreement without flattening it, moving toward something the group can own. Done well, it turns a crowd into collective intelligence; done badly, it manufactures false consensus, or noise. It is a craft, hard to do well and harder to do consistently, and good human facilitators are scarce, which the AI & Democracy Foundation documents as an open product gap. Yet there is still no shared, inspectable definition of what good facilitation is. OFL is writing one, and measuring against it, in the open.

Method specifications

A method, written so it can be run and judged.

A method spec is a portable, versioned definition of a facilitation method that any capable runtime can execute. It has two halves.

Protocol

The method itself: its stages, the facilitator's role, the prompts, and what carries from one stage to the next. Forkable, and runnable on any capable system.

Evals

How the method is held to a standard: rubrics scored against the actual conversation, with good facilitation defined by expert practitioners rather than assumed. Because the standard is measured against real conversations, it sharpens as the method does, instead of sitting frozen in a document.

What counts as good facilitation is set in the open, by the people who practice it, not encoded privately inside one company's product. Whoever owns the measure of good facilitation steers it; keeping that measure open keeps facilitation answerable to its practitioners, not to whoever owns the model.

Around the specifications sits the research they draw on: patterns, facilitation methods described through the Why-How-Who framework; teardowns of how existing platforms facilitate; and a knowledge base of the field's research and definitions. Explore the wiki.

Interoperable

Mapped to the field, not walled off from it.

OFL does not reinvent the profession or its vocabulary. Each method spec cross-maps to work the field already shares: the process patterns named by Group Works and the competencies held by the International Association of Facilitators. A spec declares which established patterns it embodies rather than inventing private ones, so the standard builds on what facilitators already practice instead of starting over.

Reference implementation

Where the specifications are proven.

The specifications are developed against a working system. Harmonica, the AI facilitation platform, is the reference implementation: methods run there with real participants, and the evaluations are tested at scale. Because the specs and evaluations are open, the same methods run on any capable system.

Open specifications, run anywhere

Get involved

How to take part.

Contribute

Defining good facilitation is not one company's job. Specifications, patterns, and research are developed in the open and take pull requests; the evals are calibrated by expert facilitators. Add a method, sharpen an eval, or tear down a tool, and you help set the standard the field measures against. Open the org.

Follow

Research notes and project updates on Substack.