AgentGrid
AgentGrid provides a visual canvas workspace that unifies coding agents, terminals, and browsers into a single environment for software development. The platform enables master-worker orchestration across multiple AI harnesses—including Claude, Codex, Cursor, and Devin—with persistent project context, source control review, and provider usage monitoring. Available for macOS, Windows, and Linux, the desktop app also offers mobile companions for remote agent monitoring.
- Developer Tools
- AI Agents
- Artificial Intelligence
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Developers working with multiple AI coding agents face fragmented workflows, forcing them to switch between separate chat windows, terminals, and source control tools. This fragmentation breaks context continuity, makes it difficult to coordinate multi-agent teams, and wastes time re-establishing project state every time a session is interrupted.
Solution
AgentGrid provides a visual canvas workspace that brings coding agents, terminals, and browsers into one unified environment. A master agent pane can spawn and coordinate worker agents across installed AI harnesses—including Claude, Codex, OpenCode, Cursor, Grok, and Devin—each running in its own visible pane with live streaming of reasoning and tool activity. The platform persists project layout, notes, agent sessions, and terminal output so users can restart and resume work with full context intact. Source control views show diffs beside the conversation, and provider usage indicators display limits and reset times. The desktop app runs on macOS, Windows, and Linux with a background daemon for always-ready agents, and a TestFlight beta iPhone app lets users check on agents remotely over Tailscale.
Target Audience
Primary users are software developers and engineering teams who work with multiple AI coding agents and need a unified workspace to orchestrate, monitor, and review multi-agent development workflows.
Features
- Master-worker orchestration model enabling one agent to spawn, coordinate, and reuse workers across multiple harnesses (Claude, Codex, OpenCode, Cursor, Grok, Devin, Kimi, Antigravity, pi-mono) with optional per-task model selection
- Persistent canvas with saved project layout, notes, agent sessions, and terminal output for seamless resume after restart or interruption
- Source control integration showing worktree diffs beside the agent conversation for focused review
- Cross-agent messaging via `send_to_pane` with nonce-fenced attribution, enabling direct collaboration between master sessions
- Provider usage dashboard displaying limits, reset times, and capacity across installed harnesses, with optional LiteLLM integration for team spending and budgets
- Voice control through the Coordinator, which accepts calls and submits wrap-up summaries automatically
- Recovery tools including `list_my_team` and `list_recoverable_sessions` for rediscovering workers and resuming sessions after restart
- Access control levels (Supervised, Auto-accept edits, Auto, Full access) for team leads, workers, and Coordinator composers