@drew /accord
Subject: Switch chat storage to WAL mode?
Selected accord participants: taylor.
Why AccordAgents
Stop rebuilding context across agent chats
Project work gets scattered when each agent lives in a separate terminal or chat. AccordAgents keeps the thread, rules, decisions, and permissions attached to the project while you route work to the agent you need.
Worked for 6m 22s
Done — rewrote the hero around the audience, added provider chips under the CTA, surfaced the real requirements, and cut the preview image from 1.15 MB to 17.7 KB. Committed to a new branch, build green.
@codex can you launch it for me?
Worked for 1m 20s
Launched. The dev server is running at 127.0.0.1:4321 — opened it in your browser.
Real workflow: the heavyweight model did the hard part, a cheaper agent picked up the chore — same context, zero re-briefing.
How it works
Mention an agent, approve requested actions, keep the answer in context
Work moves through explicit mentions and configurable approval cards, so you choose when agents ask and when trusted workflows can run.
01
@researcher Mention an agent
Start work by @handle. Mention one participant, or several to run in parallel from the same project context.
02
approve action Approve requested actions
Agents can request handoffs, permissions, and decisions through approval cards. Approve once, allow repeated requests, or use auto mode for trusted workflows.
03
reply in thread Get the reply in context
Responses return in the same conversation — grounded in your project context, roles, and history, not copy-pasted between apps.
What you get
What stays in one place
Shared project context
Keep conversations, rules, decisions, permissions, and history together across ongoing work.
Multiple agent perspectives
Ask several agents to look at the same problem from the same project context.
Roles and configurable control
Create agents with clear responsibilities, manage permissions, and choose when approvals are required.
Who it's for
Built for developers who run agents all day
solo builder You're the whole team now
You act as the tech lead; your agents are the implementer, the reviewer, and QA — on one repo, in one conversation. Ship like a team without hiring one.
lead who needs process Process instead of chaos
Reusable roles, behavior rules, and a decision trail that survives sessions. Get a second opinion from a different model before a risky change lands.
multi-vendor by design Never marry one vendor
Mix agents from different providers in the same context. Compare their takes side by side and keep your leverage.
Why a team
Why a team instead of a single agent?
| A single agent | A team in AccordAgents |
|---|---|
| One agent per terminal | Named teammates with persistent roles |
| One vendor's models | Mixed providers in one conversation, real second opinions |
| Context dies with the session | Project memory: rules, decisions, history |
| Full access or nothing | Per-participant permissions, read-only by default, approval cards |
FAQ
Frequently asked questions
How much does AccordAgents cost?
It is free and open source under the Apache-2.0 license. It coordinates the CLI agents you already pay for — your existing subscriptions or API plans — and adds no fees on top.
Where does my code go?
Nowhere new. Conversations, roles, and decisions are stored locally in SQLite on your Mac, and your CLI agents keep talking to their own model providers exactly as they do today — under your accounts and configuration.
What do I need to install?
macOS 13 or later and at least one supported CLI — Codex CLI, Claude Code, or Antigravity CLI — signed in to your account. Download AccordAgents, open your project, and add participants.
Is there a Windows or Linux version?
Not yet — AccordAgents is macOS-only today. Windows and Linux interest is tracked on GitHub, and builds will be announced there first. Join the Windows/Linux thread
How is this different from subagents?
Subagents are ephemeral workers inside a single vendor's session — they disappear when the run ends. AccordAgents participants are named teammates that persist across sessions, can come from different providers, and carry their own roles, permissions, and shared project history.
Currently supports Codex CLI, Claude Code, and Antigravity CLI. More CLI providers are coming soon.
Codex CLI Claude Code Antigravity CLI
Give your agents one workspace
Open source, Apache-2.0 — star it, fork it, extend it.