ACP (Agent Communication Protocol) agents allow integration of external coding agents and AI assistants into the LLMling-Agent pool. These agents run as separate processes and communicate via the ACP protocol.

Overview¶

LLMling-Agent supports integration with many popular coding agents:

Claude Code: Anthropic's Claude-based coding agent
Gemini CLI: Google's Gemini-based coding agent
Codex: OpenAI Codex-based agent
OpenCode: Open-source coding agent
Goose: Block's AI coding assistant
OpenHands: Open-source AI coding agent
FastAgent: Fast and lightweight coding agent
Amp: Sourcegraph's AI coding assistant
Auggie: AI pair programming assistant
Cagent: AI coding agent
Kimi: Moonshot's AI coding agent
Stakpak: AI infrastructure assistant
VTCode: Visual Studio Code AI assistant
Mistral: Mistral AI's coding agent

These agents can be configured to work alongside LLMling agents, enabling hybrid workflows that leverage specialized external tools.

Configuration Reference¶

Custom ACP Agent Configuration¶

Configuration for a custom ACP agent with explicit command.

Use this for ACP servers that don't have a preset, or when you need full control over the command and arguments.

Example:

agents:
  custom_agent:
    type: acp
    command: my-acp-server
    args: ["--mode", "coding"]
    cwd: /path/to/project

Custom ACP Agent Configuration (YAML)
agentname:
  type: acp  # Discriminator for custom ACP agent.
  command: claude-code-acp  # Command to spawn the ACP server.
  args: []  # Arguments to pass to the command.
  providers: []  # Model providers this agent can use.
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

Codex ACP Agent Configuration¶

Configuration for Zed Codex via ACP.

Provides typed settings for the codex-acp server.

Example:

acp_agents:
  coder:
    type: codex
    cwd: /path/to/project
    model: o3
    sandbox_permissions:
      - disk-full-read-access

Codex ACP Agent Configuration (YAML)
agentname:
  type: codex  # Discriminator for Codex ACP agent.
  model: null  # Model override.
  sandbox_permissions: null  # Sandbox permissions.
  shell_environment_policy_inherit: null  # Shell environment inheritance policy.
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

OpenCode ACP Agent Configuration¶

Configuration for OpenCode via ACP.

Provides typed settings for the opencode acp server.

Example:

acp_agents:
  coder:
    type: opencode
    cwd: /path/to/project
    log_level: INFO

OpenCode ACP Agent Configuration (YAML)
agentname:
  type: opencode  # Discriminator for OpenCode ACP agent.
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

Goose ACP Agent Configuration¶

Configuration for Goose via ACP.

Block's open-source coding agent.

Example:

acp_agents:
  coder:
    type: goose
    cwd: /path/to/project

Goose ACP Agent Configuration (YAML)
agentname:
  type: goose  # Discriminator for Goose ACP agent.
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

OpenHands ACP Agent Configuration¶

Configuration for OpenHands via ACP.

Open-source autonomous AI agent (formerly OpenDevin).

Example:

acp_agents:
  coder:
    type: openhands
    cwd: /path/to/project

OpenHands ACP Agent Configuration (YAML)
agentname:
  type: openhands  # Discriminator for OpenHands ACP agent.
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

Amp ACP Agent Configuration¶

Configuration for Amp (AmpCode) via ACP.

ACP bridge adapter that spawns the Amp CLI internally. The amp-acp bridge itself has no CLI configuration options. It spawns amp --no-notifications and bridges the communication to ACP protocol.

Configuration is done via environment variables: - AMP_EXECUTABLE: Path to amp binary (default: 'amp' from PATH) - AMP_PREFER_SYSTEM_PATH: Set to '1' to use system amp instead of npx version - AMP_API_KEY: API key for Amp service - AMP_URL: URL for Amp service (default: https://ampcode.com/) - AMP_SETTINGS_FILE: Path to settings file

For amp CLI configuration (permissions, MCP servers, etc.), use the amp settings file at ~/.config/amp/settings.json

Example:

acp_agents:
  amp:
    type: amp
    cwd: /path/to/project
    env:
      AMP_EXECUTABLE: /usr/local/bin/amp
      AMP_PREFER_SYSTEM_PATH: "1"
      AMP_API_KEY: your-api-key

Amp ACP Agent Configuration (YAML)
agentname:
  type: amp  # Discriminator for Amp ACP agent.
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

Cagent ACP Agent Configuration¶

Configuration for Docker cagent via ACP.

Agent Builder and Runtime by Docker Engineering.

Example:

acp_agents:
  cagent:
    type: cagent
    cwd: /path/to/project
    agent_file: ./agent.yaml
    code_mode_tools: true
    working_dir: /path/to/work

Cagent ACP Agent Configuration (YAML)
agentname:
  type: cagent  # Discriminator for Docker cagent ACP agent.
  agent_file: null  # Agent configuration file or registry reference.
  code_mode_tools: false  # Provide a single tool to call other tools via Javascript.
  env_from_file: null  # Set environment variables from file.
  models_gateway: null  # Set the models gateway address.
  working_dir: null  # Set the working directory for the session.
  debug: false  # Enable debug logging.
  otel: false  # Enable OpenTelemetry tracing.
  log_file: null  # Path to debug log file.
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

Stakpak ACP Agent Configuration¶

Configuration for Stakpak Agent via ACP.

Terminal-native DevOps Agent in Rust with enterprise-grade security.

Example:

acp_agents:
  stakpak:
    type: stakpak
    cwd: /path/to/project
    model: smart
    workdir: /path/to/work
    verbose: true

Stakpak ACP Agent Configuration (YAML)
agentname:
  type: stakpak  # Discriminator for Stakpak ACP agent.
  workdir: null  # Run the agent in a specific directory.
  verbose: false  # Enable verbose output.
  debug: false  # Enable debug output.
  disable_secret_redaction: false  # Disable secret redaction (WARNING: prints secrets to console).
  privacy_mode: false  # Enable privacy mode to redact private data.
  study_mode: false  # Enable study mode to use the agent as a study assistant.
  index_big_project: false  # Allow indexing of large projects (more than 500 supported files).
  enable_slack_tools: false  # Enable Slack tools (experimental).
  disable_mcp_mtls: false  # Disable mTLS (WARNING: uses unencrypted HTTP communication).
  enable_subagents: false  # Enable subagents.
  subagent_config: null  # Subagent configuration file subagents.toml.
  allowed_tools: null  # Allow only the specified tools in the agent's context.
  system_prompt_file: null  # Read system prompt from file.
  profile: null  # Configuration profile to use.
  model: null  # Choose agent model on startup.
  config: null  # Custom path to config file.
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

Mistral ACP Agent Configuration¶

Configuration for Mistral Agent via ACP.

Example:

acp_agents:
  coder:
    type: mistral
    cwd: /path/to/project

Mistral ACP Agent Configuration (YAML)
agentname:
  type: mistral  # Discriminator for Mistral ACP agent.
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

VTCode ACP Agent Configuration¶

Configuration for VT Code via ACP.

Rust-based terminal coding agent with semantic code intelligence.

Example:

acp_agents:
  vtcode:
    type: vtcode
    cwd: /path/to/project
    model: gemini-2.5-flash-preview-05-20
    provider: gemini
    workspace: /path/to/workspace

VTCode ACP Agent Configuration (YAML)
agentname:
  type: vtcode  # Discriminator for VT Code ACP agent.
  model: null  # LLM Model ID.
  provider: null  # LLM Provider.
  api_key_env: null  # API key environment variable.
  workspace: null  # Workspace root directory for file operations.
  enable_tree_sitter: false  # Enable tree-sitter code analysis.
  performance_monitoring: false  # Enable performance monitoring.
  research_preview: false  # Enable research-preview features.
  security_level: null  # Security level for tool execution.
  show_file_diffs: false  # Show diffs for file changes in chat interface.
  max_concurrent_ops: null  # Maximum concurrent async operations.
  api_rate_limit: null  # Maximum API requests per minute.
  max_tool_calls: null  # Maximum tool calls per session.
  config: null  # Configuration file path.
  skip_confirmations: false  # Skip safety confirmations.
  full_auto: false  # Enable full-auto mode (no interaction).
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

Cursor ACP Agent Configuration¶

Configuration for Cursor via ACP.

Cursor CLI agent with filesystem and terminal capabilities. See https://github.com/blowmage/cursor-agent-acp-npm

Example:

acp_agents:
  coder:
    type: cursor
    cwd: /path/to/project
    session_dir: ~/.cursor-sessions
    timeout: 30000

Cursor ACP Agent Configuration (YAML)
agentname:
  type: cursor  # Discriminator for Cursor ACP agent.
  config: null  # Path to configuration file.
  log_level: null  # Logging level.
  log_file: null  # Log file path (logs to stderr by default).
  session_dir: null  # Session storage directory (default: ~/.cursor-sessions).
  timeout: null  # Cursor-agent timeout in milliseconds (default: 30000).
  retries: null  # Number of retries for cursor-agent commands (default: 3).
  max_sessions: null  # Maximum number of concurrent sessions (default: 100).
  session_timeout: null  # Session timeout in milliseconds (default: 3600000).
  no_filesystem: false  # Disable filesystem tools.
  no_terminal: false  # Disable terminal tools.
  max_processes: null  # Maximum number of terminal processes (default: 5).
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

Claude ACP Agent Configuration¶

Configuration for Claude Code via ACP.

Provides typed settings for the claude-code-acp server.

Example:

agents:
  coder:
    type: claude
    cwd: /path/to/project
    model: sonnet
    permission_mode: acceptEdits
    allowed_tools:
      - Read
      - Write
      - Bash(git:)

Claude ACP Agent Configuration (YAML)
agentname:
  type: claude  # Discriminator for Claude ACP agent.
  system_prompt: null  # Custom system prompt (replaces default Claude Code prompt).
  append_system_prompt: null  # Text to append to the default system prompt.
  model: null  # Model override. Use alias ('sonnet', 'opus') or full name.
  permission_mode: null  # Permission handling mode for tool execution.
  allowed_tools: null  # Whitelist of allowed tools (e.g., ['Read', 'Write', 'Bash(git:)']).
  disallowed_tools: null  # Blacklist of disallowed tools.
  strict_mcp_config: false  # Only use MCP servers from mcp_config, ignoring all other configs.
  add_dir: null  # Additional directories to allow tool access to.
  tools: null  # Available tools from built-in set. Empty list disables all tools.
  fallback_model: null  # Fallback model when default is overloaded.
  dangerously_skip_permissions: false  # Bypass all permission checks. Only for sandboxed environments.
  output_type: null  # Structured output configuration. Generates --output-format and --json-schema.
  toolsets: []  # Toolsets to expose to this ACP agent via MCP bridge.
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

Gemini ACP Agent Configuration¶

Configuration for Gemini CLI via ACP.

Provides typed settings for the gemini CLI with ACP support.

Example:

acp_agents:
  coder:
    type: gemini
    cwd: /path/to/project
    model: gemini-2.5-pro
    approval_mode: auto_edit
    allowed_tools:
      - read_file
      - write_file
      - terminal

Gemini ACP Agent Configuration (YAML)
agentname:
  type: gemini  # Discriminator for Gemini ACP agent.
  model: null  # Model override.
  approval_mode: null  # Approval mode for tool execution.
  sandbox: false  # Run in sandbox mode.
  yolo: false  # Automatically accept all actions.
  allowed_tools: null  # Tools allowed to run without confirmation.
  allowed_mcp_server_names: null  # Allowed MCP server names.
  extensions: null  # List of extensions to use. If not provided, all are used.
  include_directories: null  # Additional directories to include in the workspace.
  output_format: null  # Output format.
  toolsets: []  # Toolsets to expose to this ACP agent via MCP bridge.
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

FastAgent ACP Agent Configuration¶

Configuration for fast-agent via ACP.

Robust LLM agent with comprehensive MCP support.

Supports MCP server integration via: - Internal bridge: Use toolsets field to expose llmling-agent toolsets - External servers: Use url field to connect to external MCP servers - Skills: Use skills_dir to specify custom skills directory

Example:

acp_agents:
  coder:
    type: fast-agent
    cwd: /path/to/project
    model: claude-3.5-sonnet-20241022
    toolsets:
      - type: subagent
      - type: agent_management
    skills_dir: ./my-skills

FastAgent ACP Agent Configuration (YAML)
agentname:
  type: fast-agent  # Discriminator for fast-agent ACP agent.
  model: anthropic.claude-3-7-sonnet-latest  # Model to use.
  shell_access: false  # Enable shell and file access (-x flag).
  skills_dir: null  # Override the default skills directory for custom agent skills.
  url: null  # MCP server URL to connect to. Can also be used with internal toolsets bridge.
  auth: null  # Authentication token for MCP server.
  toolsets: []  # Toolsets to expose to this ACP agent via MCP bridge.
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

Auggie ACP Agent Configuration¶

Configuration for Auggie (Augment Code) via ACP.

AI agent that brings Augment Code's power to the terminal.

Example:

acp_agents:
  auggie:
    type: auggie
    cwd: /path/to/project
    model: auggie-sonnet
    workspace_root: /path/to/workspace
    rules: [rules.md]
    shell: bash

Auggie ACP Agent Configuration (YAML)
agentname:
  type: auggie  # Discriminator for Auggie ACP agent.
  model: null  # Model to use.
  workspace_root: null  # Workspace root (auto-detects git root if absent).
  rules: null  # Additional rules files.
  augment_cache_dir: null  # Cache directory (default: ~/.augment).
  retry_timeout: null  # Timeout for rate-limit retries (seconds).
  allow_indexing: false  # Skip the indexing confirmation screen in interactive mode.
  augment_token_file: null  # Path to file containing authentication token.
  github_api_token: null  # Path to file containing GitHub API token.
  permission: null  # Tool permissions with 'tool-name:policy' format.
  remove_tool: null  # Remove specific tools by name.
  shell: null  # Select shell.
  startup_script: null  # Inline startup script to run before each command.
  startup_script_file: null  # Load startup script from file.
  toolsets: []  # Toolsets to expose to this ACP agent via MCP bridge.
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

Kimi ACP Agent Configuration¶

Configuration for Kimi CLI via ACP.

Command-line agent from Moonshot AI with ACP support.

Example:

acp_agents:
  kimi:
    type: kimi
    cwd: /path/to/project
    model: kimi-v1
    work_dir: /path/to/work
    yolo: true

Kimi ACP Agent Configuration (YAML)
agentname:
  type: kimi  # Discriminator for Kimi CLI ACP agent.
  verbose: false  # Print verbose information.
  debug: false  # Log debug information.
  agent_file: null  # Custom agent specification file.
  model: null  # LLM model to use.
  work_dir: null  # Working directory for the agent.
  yolo: false  # Automatically approve all actions.
  thinking: null  # Enable thinking mode if supported.
  toolsets: []  # Toolsets to expose to this ACP agent via MCP bridge.
  cwd: null  # Working directory for the session.
  env: {}  # Environment variables to set.
  execution_environment: local  # Execution environment config for ACP client operations (filesystem, terminals).
  allow_file_operations: true  # Whether to allow file read/write operations.
  allow_terminal: true  # Whether to allow terminal operations.
  auto_grant_permissions: true  # Whether to automatically grant all permission requests.
  name: null  # Identifier for the node. Set from dict key, not from YAML.
  config_file_path: null  # Config file path for resolving relative paths.
  display_name: null  # Human-readable display name for the node.
  description: null  # Optional description of the agent / team.
  triggers: []  # Event sources that activate this agent / team
  connections: []  # Targets to forward results to.
  mcp_servers: []  # List of MCP server configurations:
  input_provider: null  # Provider for human-input-handling.
  event_handlers: []  # Event handlers for processing agent stream events.

Configuration Notes¶

ACP agents run as separate processes managed by LLMling-Agent
Each agent type has specific configuration options for its underlying tool
Model selection depends on what the external agent supports
MCP servers can be attached to capable agents for extended functionality
Environment variables can be used for API keys and secrets
Agents are started on-demand and terminated when the pool closes