Harmonica began as an AI interviewer — 1:1 chats in isolated sessions. Version 0.4 turns it into a runtime: something that takes a facilitation method and runs it reliably in multiplayer mode. You write the logic once (the stages, the roles, what carries forward), and Harmonica executes it for a group rather than one person at a time.

Until now each conversation ran in its own bubble, so you could run a survey but you couldn't execute a method like a Delphi panel, a Wardley mapping session, or a compounding retrospective. 0.4 removes that constraint.

The engine. Sessions now chain into multi-step workflows. You pick a framework; each stage runs as a session with its own role assignments, its own facilitator prompts per role, and context inherited from the prior stage. When everyone in a stage finishes, the chain advances automatically — participants get emailed when their turn is ready, and an in-app notifications hub tracks progression for the host. Sixteen methods ship as ready-to-run templates: eight multi-step chains and eight single-step sessions.

Grounded summaries. Every summary now has the participant's exact words behind each claim — verbatim quotes, never paraphrased, attributed pseudonymously (Participant 3, Participant 7). Hosts control who sees results: public, participants-only, or restricted.

Layered prompts. The facilitation prompt now composes: your organizational context (HARMONICA.md, seeded at onboarding), a methodology layer for the chosen template, and per-session sources you attach at setup (a research PDF, a prior session's results, an MCP server the facilitator consults live).

What it runs. Three use cases we're highlighting: Wardley mapping that produces an actual map drawn from what the group said (rendered as portable Mermaid text, not locked into our tool); retrospectives that compound instead of getting cancelled, each run landing in the same shared project; and public sensemaking at scale, synthesized on public pages with every claim grounded. Metagov's gov/acc research used Harmonica to gather inputs from 50+ web3 governance experts, distilled into a public knowledge commons.

Agentic and open. Through the public MCP server, any AI agent can create and run Harmonica sessions. Drop the harmonica-chat skill into a coding agent and "run a retro on our API redesign" becomes a session pre-loaded with your project. A researcher (Maria Milosh) implemented a novel cross-pollination method entirely as agent orchestration on top of the MCP, with no changes to the platform. We're working with Metagov toward an open format: facilitation protocols anyone can describe, fork, and hand to Harmonica to run.

How it learns. A new Review tab gives an AI critique of how the facilitator actually behaved, each finding backed by participant quotes; where it proposes a rule, you apply it in one click. Underneath is a measurement layer that scores each facilitator prompt against a rubric judged by a second LLM — observability and evals as the precondition for pushing facilitation autonomy further.

Support us. We don't have external funding, so the best way to support us is to grab one of the 26 lifetime deals we're offering here on Open Collective — a 20% discount versus the in-app price. Pay once for every premium feature, forever, paired with bring-your-own-model: you connect your own LLM and cover the token cost yourself. Get the Founding 26 lifetime deal →