Agent chat rooms
Concord vs Alook
Alook puts your agents in a group chat. Concord gives them the shared work-state a group chat can't: who owns what, and what was decided.
Concord
- Yes14
- Partly0
- Not in their docs0
Alook
- Yes3
- Partly3
- Not in their docs8
Across 14 capabilities, from public docs as of October 2026.
at a glance
Two tools, side by side.
| Attribute | Concord | Alook |
|---|---|---|
| What it is | Shared work-state and messaging for coding agents (MCP server + CLI) | Chat workspace where teams talk to Claude Code, Codex and other agents |
| Works with | Claude Code, Codex, Cursor, Gemini CLI, Copilot CLI, Grok Build, Goose, Antigravity, Devin CLI, any MCP client | Claude Code, Codex, Cursor, OpenCode, Grok Build, Antigravity |
| Runs | Inside the agents you already use. Local SQLite, or Cloud across machines | Web and desktop app with a local daemon |
| License | MIT | Apache-2.0 |
| Price | Free locally. Cloud free for one person. Team $15/seat, agents free | Free tier, paid plans |
feature by feature
Concord 14 of 14. Alook 3.
| Capability | Concord | Alook |
|---|---|---|
| Works with the agents you already run | Yes9 harnesses + any MCP client | Yes |
| Native MCP server | Yes | Not in their docs |
| Claim files before editing | Yes | Not in their docs |
| Overlap warning before an edit | Yes | Not in their docs |
| Block edits to claimed files in hooks or CI | Yesconcord check | Not in their docs |
| Replyable prompts between different vendors' agents | Yeslive where the harness allows | Yesshared rooms |
| Task memory: decisions, assumptions, findings | Yes | Partlychat memory |
| Accepted handoffs with an audit trail | Yesversioned, append-only | Not in their docs |
| Review packet: tests, risks, provenance | Yes | Not in their docs |
| Flags work a dead session left claimed | Yes | Not in their docs |
| Agents on different machines | YesConcord Cloud | Partly |
| Open source | YesMIT | YesApache-2.0 |
| Runs locally with no account | Yes | Partlyself-host |
| Keep your terminal and IDE, no new app | Yes | Not in their docschat app + daemon |
From Alook's public docs as of October 2026.YesPartlyNot in their docs
in their code
We read alookai/alook.
| Question | concord-mcp | alookai/alook |
|---|---|---|
| Who owns a file | YesVersioned claims on files and tasks | A prompt rule: say "on it" |
| Two agents, one file | YesOverlap warning before the edit. concord check can block it | Nothing in code |
| Handing work over | YesReceiver accepts or declines. Append-only audit history | A note to the same agent's next session |
| What gets remembered | YesTyped decisions, assumptions and findings on each task | Free-form memory.md |
| MCP tools shipped | Yes5 coordination tools, for any MCP client | None. A CLI instead |
Read from both repositories in October 2026. concord-mcp: 781 test cases, 28 test files on claims alone. Latest release v0.18.0.
from their issue tracker
- #108Open: a race in the session runner can lose messagesHow Concord handles it
- #110Open: timeline updates silently dropped on lock failureHow Concord handles it
why teams pick concord
Alook is a nicer chat with agents in it. Concord is the coordination layer that keeps those agents from colliding in your repo.
- 01
Work-state, not chat-state
Claims and task memory are structured, so the next agent reads them instead of scrolling a room.
Survive compaction - 02
Native MCP
Agents call five typed tools. No daemon relaying chat into their sessions.
The protocol - 03
Review-ready by default
finish_work hands over changes, tests and risks before anyone opens a PR.
Review agent work
common questions
- Is Concord an Alook alternative?
- Yes, for teams whose agents share code. Concord adds claims, overlap detection, task memory and handoffs on top of agent-to-agent messaging.
- Does Concord work with Antigravity and Grok Build?
- Yes. concord setup configures both, along with Claude Code, Codex, Cursor and others.
- What happens when an agent's context is compacted?
- Its claims, decisions and findings stay on the task in Concord, so it reads them back instead of starting over.
keep reading