JetBrains

JetBrains ACP clients can launch Anvil as a custom agent server. Use an absolute path to the installed or locally built executable.

Manual Configuration

Add an entry to ~/.jetbrains/acp.json:

{
  "agent_servers": {
    "Anvil": {
      "command": "/absolute/path/to/anvil",
      "args": [],
      "env": {}
    }
  }
}

Preserve unrelated entries already present in the file. Use anvil.exe on Windows. Restart the agent UI or start a fresh session after changing this configuration.

Source-Checkout Helper

From an Anvil checkout:

rustup target add wasm32-wasip2
cargo xtask build-acp-for-jetbrains

The helper builds target/release/anvil and writes a Brokk Code (Rust Local) entry. Pass --config <path> when the active JetBrains ACP configuration is elsewhere. The helper is development tooling and is not packaged by cargo install.

Setup and Validation

Open a new Anvil session and run /setup if no model is ready. Use the editor’s session controls to select Permission, behavior, model, reasoning effort, and service tier.

Validate with a source question that requires managed Bifrost:

Use Bifrost search_symbols to find main and get_symbol_sources to read it. Name the tools you called and do not edit files.

A visible Bifrost tool card plus repository-specific source output proves the code-intelligence path. If the agent falls back to reading files, confirm the session is using Anvil and inspect stderr for MCP startup failures.

The Zed-led ten-minute evaluation uses the same Anvil behavior. Substitute this configuration and JetBrains’ permission controls for the editor-specific steps.