Agents on your GTM tablesYou approve every write
How supervision works
Complex runs, explicit approval
Ask
Prompt from the table
Tell Codex or Claude what to enrich, import, or research. Agents read your rows through Appenda tools — not a pasted CSV.
Approve
Nothing writes without you
Every batch update, provider run, and row change surfaces a permission card. Allow once, for the session, or deny.
Commit
Plays you can replay
Approved work becomes a named commit. Fork what worked, revert what did not, and hand the same play to your team.
Built-in agents
Official ACP adapters, staged with the app
Appenda speaks Agent Client Protocol over stdio. One pinned npm adapter per provider — the same stack we ship in the desktop build, never a random PATH install.
Codex
OpenAI Codex through the official @agentclientprotocol/codex-acp adapter.
Claude Code
Anthropic Claude Code through the official claude-agent-acp adapter.
How a run works
Step 1
Connect an agent tab
Pick Codex or Claude Code. Appenda stages the official ACP adapter next to the app — no PATH hacks.
Step 2
Review tool calls
Agents list tables, read rows, and draft provider runs. You see the plan before anything executes.
Step 3
Ship as a commit
Approved changes land in one bounded commit with scope and summary — ready to export or replay.
Prefer Cursor or another MCP client? Wire Appenda through MCP while the desktop app is running.