/agent-brain-init
Output as JSON
> /plugin marketplace add SpillwaveSolutions/agent-brain > /plugin install agent-brain@agent-brain-marketplace
How it fires
How this command gets triggered: by you, by Claude, or both.
- Fires itselfClaude auto-loads it when your prompt matches the work.
- You can call itInvoke it directly when you want it.
- Slash command
/agent-brain-init
Context preview
What this command does when you run it.
Output as JSON
Command definition
agent-brain-init.mdname: agent-brain-init
description: Initialize Agent Brain for the current project
parameters:
- name: path
description: "Project path (default: auto-detect project root)"
required: false
- name: host
description: "Server bind host (default: 127.0.0.1)"
required: false
- name: port
description: "Preferred server port (default: auto-select from range)"
required: false
- name: force
description: Overwrite existing configuration
required: false
default: false
- name: state-dir
description: Custom state directory for index data
required: false
- name: json
description: Output as JSON
required: false
default: false
context: fork
agent: setup-assistant
skills:
- configuring-agent-brain
last_validated: 2026-03-16Initialize Agent Brain Project
Purpose
Initializes the current project for Agent Brain by creating the necessary configuration directory and files. This sets up per-project isolation, allowing each project to have its own Agent Brain instance with separate configuration and data.
Usage
/agent-brain:agent-brain-init [--path <path>] [--host <host>] [--port <port>] [--force] [--state-dir <dir>] [--json]
Parameters
| Parameter | Required | Default | Description | |-----------|----------|---------|-------------| | --path / -p | No | Auto-detect | Project path (auto-detects git root or project markers) | | --host | No | 127.0.0.1 | Server bind host | | --port | No | Auto-select | Preferred server port (disables auto-port if set) | | --force / -f | No | false | Overwrite existing configuration | | --state-dir / -s | No | .agent-brain | Custom state directory for index data | | --json | No | false | Output as JSON |
Examples
/agent-brain:agent-brain-init
/agent-brain:agent-brain-init --path /my/project
/agent-brain:agent-brain-init --port 8080
/agent-brain:agent-brain-init --state-dir /custom/path
/agent-brain:agent-brain-init --force
Execution
Run Initialization
agent-brain init
agent-brain init --path /my/project
agent-brain init --port 8080
agent-brain init --state-dir /custom/path
agent-brain init --force
This creates the `.agent-brain/` directory structure in the project root.
Verify Initialization
ls -la .agent-brain/
Output
Agent Brain Initialization
==========================
Initializing Agent Brain for current project...
Running: agent-brain init
Created directory structure:
.agent-brain/
config.json - Project configuration
chroma_db/ - Vector store (created on first index)
bm25_index/ - Keyword index (created on first index)
Project initialized successfully!
Configuration file: .agent-brain/config.json
{
"project_name": "my-project",
"created_at": "2025-01-31T12:00:00Z",
"mode": "project"
}
Next steps:
1. Start server: /agent-brain:agent-brain-start
2. Index documents: /agent-brain:agent-brain-index ./docs
3. Search: /agent-brain:agent-brain-search "your query"What Gets Created
The initialization creates the following structure:
.agent-brain/
config.json # Project configuration (bind_host, port, chunk settings, exclude patterns)
data/
chroma_db/ # ChromaDB vector store (created on index)
bm25_index/ # BM25 keyword index (created on index)
llamaindex/ # LlamaIndex persistence (created on index)
logs/ # Server logsconfig.json
Contains project-specific settings:
{
"bind_host": "127.0.0.1",
"port_range_start": 8000,
"port_range_end": 8100,
"auto_port": true,
"chunk_size": 512,
"chunk_overlap": 50,
"exclude_patterns": [
"**/node_modules/**",
"**/__pycache__/**",
"**/.venv/**",
"**/venv/**",
"**/.git/**",
"**/dist/**",
"**/build/**",
"**/target/**"
],
"project_root": "/path/to/project"
}config.yaml (Optional)
Create a `config.yaml` in the project's `.agent-brain/` directory for project-specific provider settings:
# .agent-brain/config.yaml
project:
state_dir: null # Use default
embedding:
provider: "openai"
api_key: "sk-proj-..." # Or use api_key_env: "OPENAI_API_KEY"
summarization:
provider: "anthropic"
api_key: "sk-ant-..."
**Note**: Agent Brain searches for config.yaml in multiple locations. Project-level config takes precedence over user-level (`~/.agent-brain/config.yaml`).
runtime.json
Created when server starts, contains:
{
"port": 49321,
"pid": 12345,
"started_at": "2025-01-31T12:00:00Z",
"state_dir": ".agent-brain"
}Error Handling
Already Initialized
Project already initialized.
Existing configuration found at: .agent-brain/config.json
Options:
- Continue using existing configuration
- Reset with: rm -rf .agent-brain && agent-brain init
- Check status: agent-brain status
Permission Denied
Error: Cannot create directory .agent-brain/
Permission denied.
Solutions:
1. Check directory permissions: ls -la .
2. Ensure write access to current directory
3. Create manually: mkdir -p .agent-brain
4. Check if .claude exists and is writable
Not in a Project Directory
Warning: No git repository or project markers found.
Agent Brain will initialize here, but consider:
1. Navigate to your project root first
2. Initialize git: git init
3. Then run: agent-brain init
Parent Directory Issues
Error: Cannot create .claude directory
The parent directory may not exist or is not writable.
Check:
1. Current directory exists: pwd
2. You have write permissions: ls -la .
3. Disk is not full: df -h .
Re-initialization
To completely reset a project's Agent Brain configuration:
# Stop server if running
agent-brain stop
# Remove existing configuration
rm -rf .agent-brain
# Re-initialize
agent-brain init
**Warning**: This deletes all indexed documents. You will need to re-index after re-initiali
Read more
name: agent-brain-init
description: Initialize Agent Brain for the current project
parameters:
- name: path
description: "Project path (default: auto-detect project root)"
required: false
- name: host
description: "Server bind host (default: 127.0.0.1)"
required: false
- name: port
description: "Preferred server port (default: auto-select from range)"
required: false
- name: force
description: Overwrite existing configuration
required: false
default: false
- name: state-dir
description: Custom state directory for index data
required: false
- name: json
description: Output as JSON
required: false
default: false
context: fork
agent: setup-assistant
skills:
- configuring-agent-brain
last_validated: 2026-03-16Initialize Agent Brain Project
Purpose
Initializes the current project for Agent Brain by creating the necessary configuration directory and files. This sets up per-project isolation, allowing each project to have its own Agent Brain instance with separate configuration and data.
Usage
/agent-brain:agent-brain-init [--path <path>] [--host <host>] [--port <port>] [--force] [--state-dir <dir>] [--json]
Parameters
| Parameter | Required | Default | Description | |-----------|----------|---------|-------------| | --path / -p | No | Auto-detect | Project path (auto-detects git root or project markers) | | --host | No | 127.0.0.1 | Server bind host | | --port | No | Auto-select | Preferred server port (disables auto-port if set) | | --force / -f | No | false | Overwrite existing configuration | | --state-dir / -s | No | .agent-brain | Custom state directory for index data | | --json | No | false | Output as JSON |
Examples
/agent-brain:agent-brain-init /agent-brain:agent-brain-init --path /my/project /agent-brain:agent-brain-init --port 8080 /agent-brain:agent-brain-init --state-dir /custom/path /agent-brain:agent-brain-init --force
Execution
Run Initialization
agent-brain init agent-brain init --path /my/project agent-brain init --port 8080 agent-brain init --state-dir /custom/path agent-brain init --force
This creates the `.agent-brain/` directory structure in the project root.
Verify Initialization
ls -la .agent-brain/
Output
Agent Brain Initialization
==========================
Initializing Agent Brain for current project...
Running: agent-brain init
Created directory structure:
.agent-brain/
config.json - Project configuration
chroma_db/ - Vector store (created on first index)
bm25_index/ - Keyword index (created on first index)
Project initialized successfully!
Configuration file: .agent-brain/config.json
{
"project_name": "my-project",
"created_at": "2025-01-31T12:00:00Z",
"mode": "project"
}
Next steps:
1. Start server: /agent-brain:agent-brain-start
2. Index documents: /agent-brain:agent-brain-index ./docs
3. Search: /agent-brain:agent-brain-search "your query"What Gets Created
The initialization creates the following structure:
.agent-brain/
config.json # Project configuration (bind_host, port, chunk settings, exclude patterns)
data/
chroma_db/ # ChromaDB vector store (created on index)
bm25_index/ # BM25 keyword index (created on index)
llamaindex/ # LlamaIndex persistence (created on index)
logs/ # Server logsconfig.json
Contains project-specific settings:
{
"bind_host": "127.0.0.1",
"port_range_start": 8000,
"port_range_end": 8100,
"auto_port": true,
"chunk_size": 512,
"chunk_overlap": 50,
"exclude_patterns": [
"**/node_modules/**",
"**/__pycache__/**",
"**/.venv/**",
"**/venv/**",
"**/.git/**",
"**/dist/**",
"**/build/**",
"**/target/**"
],
"project_root": "/path/to/project"
}config.yaml (Optional)
Create a `config.yaml` in the project's `.agent-brain/` directory for project-specific provider settings:
# .agent-brain/config.yaml project: state_dir: null # Use default embedding: provider: "openai" api_key: "sk-proj-..." # Or use api_key_env: "OPENAI_API_KEY" summarization: provider: "anthropic" api_key: "sk-ant-..."
**Note**: Agent Brain searches for config.yaml in multiple locations. Project-level config takes precedence over user-level (`~/.agent-brain/config.yaml`).
runtime.json
Created when server starts, contains:
{
"port": 49321,
"pid": 12345,
"started_at": "2025-01-31T12:00:00Z",
"state_dir": ".agent-brain"
}Error Handling
Already Initialized
Project already initialized. Existing configuration found at: .agent-brain/config.json Options: - Continue using existing configuration - Reset with: rm -rf .agent-brain && agent-brain init - Check status: agent-brain status
Permission Denied
Error: Cannot create directory .agent-brain/ Permission denied. Solutions: 1. Check directory permissions: ls -la . 2. Ensure write access to current directory 3. Create manually: mkdir -p .agent-brain 4. Check if .claude exists and is writable
Not in a Project Directory
Warning: No git repository or project markers found. Agent Brain will initialize here, but consider: 1. Navigate to your project root first 2. Initialize git: git init 3. Then run: agent-brain init
Parent Directory Issues
Error: Cannot create .claude directory The parent directory may not exist or is not writable. Check: 1. Current directory exists: pwd 2. You have write permissions: ls -la . 3. Disk is not full: df -h .
Re-initialization
To completely reset a project's Agent Brain configuration:
# Stop server if running agent-brain stop # Remove existing configuration rm -rf .agent-brain # Re-initialize agent-brain init
**Warning**: This deletes all indexed documents. You will need to re-index after re-initiali
A RAG-based (Retrieval-Augmented Generation) document indexing and semantic search system for AI agents and applications. Agent Brain enables intelligent querying of documentation and source code using natural language.
Other commands on agent-brain.
ag-install-local
Local build-and-install of Agent Brain server + CLI + plugin for rapid testing.
ag-remove-local-instal…
Remove all local Agent Brain installations, caches, and plugin artifacts.
implement-python
Senior Python engineer implementation command. Launches a background agent with full Python development capabilities and access…
speckit.analyze
Perform a non-destructive cross-artifact consistency and quality analysis across spec.md, plan.md, and tasks.md after task…

