Loading content…
← All Products
ModStageLIVE
Every Coding Agent, Side by Side in Claude Code
ModStage is a Claude Code plugin that runs Codex, Grok, Muse, Antigravity (Gemini), Kimi, Qwen Code, OpenCode and Claude subagents in one live panel. Watch every tool call and edit, steer any agent mid-run, and approve its risky actions from the panel or straight from Claude chat.
Install
In a Claude Code session
/plugin marketplace add qualixar/modstage/plugin install modstage@modstage
Or from a terminal (terminal, desktop app and VS Code)
claude plugin marketplace add qualixar/modstageclaude plugin install modstage@modstage
Needs Claude Code with mods (2.1.287 or newer in the terminal, 2.1.286 or newer in the desktop app's Code tab), Python 3.10 or newer, and at least one supported agent CLI installed and signed in. Start a new session after installing, then click the ModStage band or type /modstage.
8
Agents
Python stdlib
Bridge
127.0.0.1 only
Network
MIT
License
Features
Eight Agents, One Panel
Codex, Muse, Grok, Antigravity (Gemini), Kimi, Qwen Code, OpenCode and Claude subagents. Tool calls, edits, messages and thinking stream live, with tokens, context and cost where the CLI reports them.
Steer Mid-Run
Codex and Muse take your message into the running turn. For agents whose CLI cannot, the message is held and shown as queued, and Interrupt and send now ends the turn so it runs at once.
Needs-You Approval Inbox
Commands outside the sandbox, file changes and network calls wait for Allow, Allow for this run or Deny. The card scrolls into view on arrival and no answer in 10 minutes is a deny.
Approve From Claude Chat
Tell Claude to allow or deny a waiting action and it answers the card with stage_decide, only on your own words, never because an agent, a file or a web page asked.
Many Agents, One Folder
A coordination board per folder: file claims as leases where conflicts are denied, messages between agents framed as information, and Claude Code as a peer whose edits stop at claimed files.
Optional SLM-Mesh Mirror
If you run SuperLocalMemory, run peers, claims and board messages are mirrored to its local SLM-mesh so other Claude Code windows see them. Nothing depends on it; MODSTAGE_SLM_MESH=0 turns it off.
Per-Session Stage
Each Claude Code session sees only the runs it started, so a new session starts clean.
Model Pickers and Images
Pick a model from each CLI's own list, including Muse, Kimi and OpenCode, and hand runs screenshots on agents whose protocol takes images.
Tested Parity Matrix
A conformance harness drives every adapter through protocol fakes at zero token cost; every agent-by-control cell of the published parity matrix has a test behind it, and a known gap is an expected failure.
Use Cases
Hand tests to Codex while Claude keeps building the API
Run Grok and Gemini reviews in parallel and watch both
Approve or deny an agent's network or file actions from chat
Keep several agents in one repository from overwriting each other's files
See exactly what a long agent turn is doing instead of waiting blind
Use the coding-agent subscriptions you already pay for, from one place
Licensed under MIT