Braid isn't tied to one model provider. Each agent has its own model, provider and harness, and agents on different providers can work in the same thread.
In the recording, Dana creates an editor agent, Rowan, in Workbench under Agents with "+ New agent". In Rowan's studio, under Model & Harness, she sets the model to gpt-5.5, the provider to openai and the harness to codex, and saves. The roster then shows the Lead on claude-sonnet-4-6 with the claude-sdk harness and the claude provider, and Rowan on gpt-5.5 with codex and openai, in the same table.


In a discussion with both agents, Dana asks the Lead for a three-line product update, and the Lead writes a draft. She asks Rowan to review it, and Rowan replies with two numbered suggestions, each quoting the line it changes. She asks the Lead to revise using Rowan's suggestions, and the Lead replies "Revised — applied both of Rowan's points", says in a line what it changed, and gives the new text.

How to use it
- In Workbench, press "+ New agent" and fill in the handle, name, role, model and personality.
- Open the agent's studio. Under Model & Harness, set the model, provider and harness, and press Save section.
- Start a discussion and add the agents with Add agent.
- Ask one agent for work, then ask another to review it. The thread shows both.
What's next
Broad model access is supported through OpenCode, alongside Claude Code and Codex. The recording shows Claude and OpenAI Codex.
What to know
The thread does not label which provider wrote each message. The roster and the agent's studio show it. Each agent picks its own runtime and model; Codex needs a ChatGPT login on the executor. Creating agents is an org-admin action.