Demos¶
Running ACCORD and CONCORD on a multi-agent system
Recordings of the protocols running on real applications. Each is one decision, from the opening proposals to the committed outcome.
Where the recordings are hosted
The files are attached to a repository release rather than served from this site, because they are too large to publish with it. Each demo page links to its own recording; access is the same as for this site.
-
Architect agents settling a design question, recorded under two agent runtimes — one agent per architect, and Claude Agent Teams.
-
A hidden-profile task: the facts needed for the correct answer are split across the participants, and no agent holds enough alone.
-
A promotion decided across specialist systems with incompatible objectives — the one use case where the repair loop runs in most runs.
-
Four subsystem owners converging on one remediation, each able to see only its own subsystem. The application the benchmark is built on.
-
A commercial negotiation across roles pulling in different directions, included to test whether the protocol travels off incident response.
Related¶
- ACCORD — phases, guarantees and measured results
- CONCORD — phases, guarantees and measured results
- Getting Started — insertion, packaging and transport bindings